Genomic and Personalized Medicine
This page intentionally left blank
Genomic and Personalized Medicine Volume 1 Edited by Huntington F. Willard, Ph.D. Director Duke Institute for Genome Sciences & Policy Nanaline H. Duke Professor of Genome Sciences Howard Hughes Medical Institute Professor Duke University Durham, North Carolina 27708
and Geoffrey S. Ginsburg, M.D., Ph.D. Center Director, Center for Genomic Medicine Duke Institute for Genome Sciences & Policy Professor of Medicine Duke University Durham, North Carolina 27708
AMSTERDAM • BOSTON • HEIDELBERG • LONDON • NEW YORK • OXFORD • PARIS SAN DIEGO • SAN FRANCISCO • SYDNEY • TOKYO
Academic Press is an imprint of Elsevier
Academic Press is an imprint of Elsevier 525 B Street, Suite 1900, San Diego, CA 92101-4495, USA 30 Corporate Drive, Suite 400, Burlington, MA 01803, USA 32 Jamestown Road, London NW1 7BY, UK First edition 2009 Copyright © 2009 Elsevier Inc. All rights reserved No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (44) (0) 1865 843830; fax (44) (0) 1865 853333; email:
[email protected]. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material Notice No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-12-369420-1 (set) ISBN: 978-0-12-370888-5 (vol. 1) ISBN: 978-0-12-370889-2 (vol. 2) For information on all Academic Press publications visit our web site at elsevierdirect.com Typeset by Charon Tec Ltd., A Macmillan Company. (www.macmillansolutions.com) Printed and bound in China 09 10 11 12 13 10 9 8 7 6 5 4 3 2 1
Contents in Brief Foreword
xxv
Preface
xxvii
Acknowledgements
xxix
Abbreviations
xxxi
Advisory Board
xxxix
Contributors
xli
PART I GENOMIC APPROACHES TO BIOLOGY AND MEDICINE Section 1 Principles of Human Genomics 1. Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine 2. Concepts of Population Genomics 3. Genomic Approaches to Complex Disease 4. Human Health and Disease: Interaction Between the Genome and the Environment 5. Epigenomics and its Implications for Medicine 6. Systems Biology and the Emergence of Systems Medicine Section 2 Technology Platforms for Genomic Medicine 7. DNA Sequencing for the Detection of Human Genome Variation and Polymorphism 8. Genome-Wide Association Studies and Genotyping Technologies 9. Copy Number Variation and Human Health 10. Inter-Species Comparative Sequence Analysis: A Tool for Genomic Medicine 11. DNA Methylation Analysis: Providing New Insight into Human Disease 12. Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine 13. DNA Microarrays in Biological Discovery and Patient Care 14. Proteomics: The Deciphering of the Functional Genome
3
4 22 33 47 60 74
87 88 101 108 120 131 143 157 173
15. Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms 16. Comprehensive Analysis of Gene Function: RNA interference and Chemical Genomics
180 193
Section 3 Informatic and Computational Platforms for Genomic Medicine 205 17. Bioinformatic and Computational Analysis for Genomic Medicine 206 18. Fundamentals and History of Informatics for Genomic and Personalized Medicine 226 19. Electronic Medical Records in Genomic Medicine Practice and Research 233 20. Clinical Decision Support in Genomic and Personalized Medicine 242 21. Online Health Information Retrieval by Consumers and the Challenge of Personal Genomics 252
PART II TRANSLATIONAL APPROACHES IN GENOMIC AND PERSONALIZED MEDICINE Section 4 Enabling Strategies in the Translation of Genomics into Medicine 22. Translational Genomics: From Discovery to Clinical Practice 23. Principles of Study Design 24. Biobanking in the Post-Genome Era 25. Application of Biomarkers in Human Population Studies 26. Validation of Candidate Protein Biomarkers 27. Pharmacogenetics and Pharmacogenomics 28. The Role of Genomics and Genetics in Drug Discovery and Development 29. Role of Pharmacogenomics in Drug Development 30. Clinical Implementation of Translational Genomics 31. Translating Innovation in Diagnostics: Challenges and Opportunities 32. The Role of Genomics in Enabling Prospective Health Care
261 262 275 284 299 308 321 335 343 357 367 378
v
vi
■
Contents in Brief
Section 5 Policy Challenges in Genomic and Personalized Medicine 33. From Sequence to Genomic Medicine: Genome Policy Considerations 34. Educational Strategies in Genomic Medicine 35. Federal Regulation of Genomic Medicine 36. Economic Issues and Genomic Medicine 37. Public–Private Interactions in Genomic Medicine: Research and Development
387 388 401 414 424 434
Section 6 Genomic Medicine and Public Health 445 38. What Is Public Health Genomics? 446 39. Why Do We Need Public Health in the Era of Genomic Medicine? 454 40. Principles of Human Genome Epidemiology 461 41. Genomics and Population Screening: Example of Newborn Screening 470 42. Family History: A Bridge Between Genomic Medicine and Disease Prevention 481 Section 7 Clinical Technologies Supporting Personalized Medicine 43. Molecular Imaging as a Paradigm for Genomic and Personalized Medicine 44. PET Imaging in Genomic Medicine 45. MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges 46. Fluorescence Imaging: Overview and Applications in Biomedical Research 47. Imaging Genetics: Integration of Neuroimaging and Genetics in the Search for Predictive Markers 48. Viral Chip Technology in Genomic Medicine 49. Vaccines Against Infectious Diseases: A BiotechnologyDriven Evolution 50. Cancer Vaccines: Some Basic Considerations 51. Biosensors for the Genomic Age 52. Stem Cells 53. Gene Therapy
493 494 500 512 524 532 538 562 573 590 599 610
PART III DISEASE-BASED GENOMIC AND PERSONALIZED MEDICINE: GENOME DISCOVERIES AND CLINICAL APPLICATIONS Section 8 Cardiovascular Genomic Medicine 623 54. The Genomics of Hypertension 624 55. Lipoprotein Disorders 634 56. Reactive Oxygen Species Signals Leading to Vascular Dysfunction and Atherosclerosis 652 57. Genomics of Myocardial Infarction 665 58. Acute Coronary Syndromes 680
59. Heart Failure in the Era of Genomic Medicine 60. Application of Genomic Methodology for Assessment of Cardiac Transplant Rejection 61. Hypertrophic Cardiomyopathy in the Era of Genomic Medicine 62. Genetics and Genomics of Arrhythmias 63. Hemostasis and Thrombosis 64. Peripheral Arterial Disease 65. Genomics of Congenital Heart Disease 66. Genomics of Perioperative and Procedural Medicine
692
Section 9 Oncology Genomic Medicine 67. Cancer Genes, Genomes, and the Environment 68. Immune Cells and the Tumor Microenvironment 69. Lymphomas 70. Genomics in Leukemias 71. Genomics of Lung Cancer 72. Breast Cancer and Genomic Medicine 73. Colorectal Cancer 74. Prostate Cancer 75. Molecular Biology of Ovarian Cancer 76. Pancreatic Neoplasms 77. The Multiple Endocrine Neoplasia Syndromes 78. Genomics of Head and Neck Cancer 79. Genomic Medicine, Brain Tumors and Gliomas 80. Molecular Therapeutics of Melanoma 81. Emerging Concepts in Metastasis 82. Diagnostic-Therapeutic Combinations in the Treatment of Cancer
807 808 818 830 844 856 869 879 898 913 921 931 945 956 967 977
705 716 729 755 773 781 794
990
Section 10 Inflammatory Disease Genomic Medicine 1009 83. Environmental Exposures and the Emerging Field of Environmental Genomics 1010 84. Molecular Basis of Rheumatoid Arthritis 1017 85. “Omics” in the Study of Multiple Sclerosis 1032 86. Inflammatory Bowel Disease 1040 87. Glomerular Disorders 1056 88. Spondyloarthropathies 1067 89. Asthma Genomics 1084 90. Genomic Aspects of Chronic Obstructive Pulmonary Disease 1098 91. Genomic Determinants of Interstitial Lung Disease 1110 92. Peptic Ulcer Disease 1122 93. Cirrhosis in the Era of Genomic Medicine 1138 94. Systemic Sclerosis 1155 Section 11 Metabolic Disease Genomic Medicine 95. Genomic Medicine of Obesity 96. Diabetes 97. Metabolic Syndrome 98. Nutrition and Diet in the Era of Genomics
1169 1170 1187 1194 1204
Contents in Brief
Section 12 Neuropsychiatric Disease Genomic Medicine 99. The Genetic Approach to Dementia 100. Parkinson’s Disease: Genomic Perspectives 101. Epilepsy Predisposition and Pharmacogenetics 102. Ophthalmology 103. Genomic Basis of Neuromuscular Disorders 104. Psychiatric Disorders 105. Genomics and Depression 106. Bipolar Disorder in the Era of Genomic Psychiatry Section 13 Infectious Disease Genomic Medicine 107. Genomic Approaches to the Host Response to Pathogens
1221 1222 1233 1243 1256 1265 1282 1289 1299
1313 1314
108. 109. 110. 111. 112.
Genomic Medicine and Aids Viral Genomics and Antiviral Drugs Host Genomics and Bacterial Infections Sepsis and the Genomic Revolution Diagnosis, Management and Disease Mechanisms of Hepatitis in the Era of Genomic Medicine
■
vii
1324 1340 1347 1362 1375
Glossary
1391
Index
1407
This page intentionally left blank
Table of Contents Foreword
xxv
Preface
xxvii
Acknowledgements
xxix
Abbreviations
xxxi
Advisory Board
xxxix
Contributors
xli
PART I GENOMIC APPROACHES TO BIOLOGY AND MEDICINE Section 1 Principles of Human Genomics 1. Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine Huntington F. Willard Introduction The Human Genome Variation in the Human Genome Expression of the Human Genome Genes, Genomes and Disease From Genome to Personalized Medicine Conclusion References Recommended Resources 2. Concepts of Population Genomics Mike E.Weale and David B. Goldstein Introduction Important Concepts in Population Genomics Human Population Genomics Application of Population Genomics to Genomic Medicine Conclusions References Recommended Resources 3. Genomic Approaches to Complex Disease Desmond J. Smith and Aldons J. Lusis Introduction Identifying Common and Rare Genomic Variations in the Population Relating DNA Variation to Phenotypes Integration of “Omic” Technologies with Genetics
3
4 4 6 9 11 13 15 18 18 21 22 22 22 26 28 29 30 32 33 33 33 36 40
Conclusions and Prospects Acknowledgements References Recommended Resources 4. Human Health and Disease: Interaction Between the Genome and the Environment Kenneth Olden Introduction Importance of the Environment The Environmental Genome Project Problematic Nature of Gene–environment Interaction Studies Polymorphism and Disease Susceptibility: Case–control Studies Epigenetics and the Environment Conclusion Acknowledgements References 5. Epigenomics and its Implications for Medicine Moshe Szyf Introduction DNA Methylation Patterns Chromatin Modification DNA Methylation and Chromatin States Co-operatively Determine the State of Activity of Genes Epigenetics and Human Disease Conclusions Acknowledgements References 6. Systems Biology and the Emergence of Systems Medicine Nathan D. Price, Lucas B. Edelman, Inyoul Lee, Hyuntae Yoo, Daehee Hwang, George Carlson, David J. Galas, James R. Heath and Leroy Hood Introduction Systems Science in Biology And Medicine Multi-parameter Blood-bourne Biomarkers Emerging in vivo and in vitro Technologies Computational and Mathematical Challenges in Systems Medicine Conclusions and Perspectives References Recommended Resources
43 43 43 46
47 47 48 50 52 53 56 57 57 57 60 60 61 63
65 67 69 70 70 74
74 75 76 78 81 81 82 85 ix
x
■
Table of Contents
Section 2 Technology Platforms for Genomic Medicine 7. DNA Sequencing for the Detection of Human Genome Variation and Polymorphism Samuel Levy and Yu-Hui Rogers Introduction DNA Sequencing Other Methodologies for Polymorphism Detection Future Directions Acknowledgements References 8. Genome-Wide Association Studies and Genotyping Technologies Kevin V. Shianna Introduction Principles of Genome-wide Association Studies Platform Overview Conclusion References 9. Copy Number Variation and Human Health Charles Lee, Courtney Hyland, Arthur S. Lee, Shona Hislop and Chunhwa Ihm Introduction Basic Principles of CNVs Detecting CNVs in a Genome-wide Manner Association of CNVs to Disease and Disease Susceptibility Implications of CNVs Conclusions Acknowledgements References Recommended Resources 10. Inter-Species Comparative Sequence Analysis: A Tool for Genomic Medicine Anthony Antonellis and Eric D. Green Introduction Performing Comparative Sequence Analysis: Resources and Methods Comparative Sequence Analysis and Human Genetic Disease CSA and the Future of Human Genetics and Genomic Medicine References Recommended Resources
87 88 88 89 95 96 97 97
101 101 101 103 106 106 108
108 108 112 114 116 118 118 118 119
120 120 121 124 128 128 130
11. DNA Methylation Analysis: Providing New Insight into Human Disease 131 Susan Cottrell,Theo deVos, Juergen Distler, Carolina Haefliger, Ralf Lesche, Achim Plum and Matthias Schuster Introduction Technology to Assess DNA Methylation
131 132
Clinical Impact of DNA Methylation Analysis Conclusion References Recommended Resources 12. Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine Michelle M. Kittleson, Rafael Irizarry, Bettina Heidecker and Joshua M. Hare Introduction Gene Expression Technology Gene Discovery Molecular Signature Analysis Gene Discovery Versus Molecular Signature Analysis Current Issues in Gene Expression Analysis Alternative Technologies for Analysis of the Transcriptome Conclusion Acknowledgements References Recommended Resources 13. DNA Microarrays in Biological Discovery and Patient Care Andrew J.Yee and Sridhar Ramaswamy Introduction Microarray Technology Data Analysis Applications Limitations and Challenges Future Directions Conclusions References Recommended Resources
136 139 139 142
143
143 143 146 148 151 151 153 153 154 154 156 157 157 157 160 161 165 165 168 168 172
14. Proteomics: The Deciphering of the Functional Genome 173 Li-Rong Yu, Nicolas A. Stewart and Timothy D.Veenstra Introduction 173 Gel-based and Solution-based Proteomics 174 Mass Spectrometry 175 Bioinformatics 176 Impact of Proteomics on Understanding Diseases 178 Conclusions 178 Acknowledgements 179 References 179 Recommended Resources 179 15. Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms 180 Christopher B. Newgard, Robert D. Stevens, Brett R.Wenner, Shawn C. Burgess, Olga Ilkayeva, Michael J. Muehlbauer, A. Dean Sherry and James R. Bain Introduction 180 Current Metabolomics Platforms: Basic Tools and General Features 181
Table of Contents
Comparison of NMR and MS Technologies for Unbiased Metabolic Profiling MS Methods for Targeted Metabolic Profiling Examples of NMR-based Metabolic Profiling in Disease Research Examples of Targeted MS-based Metabolic Profiling for Understanding of Disease Mechanisms Integration of Metabolic Profiling with Other “Omics” Technologies Future Directions References 16. Comprehensive Analysis of Gene Function: RNA interference and Chemical Genomics Bjorn T. Gjertsen and James B. Lorens Introduction RNA Interference Gene Function Analysis: An Overview Chemical Genomics Gene Function Studies Conclusions Acknowledgements References Recommended Resources Section 3 Informatic and Computational Platforms for Genomic Medicine 17. Bioinformatic and Computational Analysis for Genomic Medicine Atul J. Butte Introduction Vignettes: How Specific Bioinformatics Methods Can Change the Practice Ofmedicine Analytic Methods Where Data for Studies May be Found Bioinformatics Vocabularies and Ontologies Freely Available Bioinformatics Tools New Questions for Genomic Medicine Acknowledgements References Recommend Resources 18. Fundamentals and History of Informatics for Genomic and Personalized Medicine A. Jamie Cuticchia Introduction Databases for Genomic Medicine Conclusion References Recommended Resources 19. Electronic Medical Records in Genomic Medicine Practice and Research Glenn S. Gerhard, Robert D. Langer, David J. Carey and Walter F. Stewart
182 183 184 184 189 189 190
193 193 194 194 197 199 200 201 201 203
205 206 206 207 213 214 215 216 220 221 221 225 226 226 227 230 231 232
233
Introduction EMRs and Genomic Medicine Clinical Practice EMRs and Genomic Medicine Research Conclusion Acknowledgements References Recommended Resources 20. Clinical Decision Support in Genomic and Personalized Medicine Kensaku Kawamoto and David F. Lobach Introduction CDS Background: History, Examples, Evidence of Effectiveness, and Desirable Attributes Potential Uses of CDS to Support Genomic and Personalized Medicine Limited Deployability: The Potential Achilles’ Heel of CDS Systems for Genomic Medicine Challenges to Widespread Deployment of Effective CDS Systems Conclusions Disclosures References Recommended Resources 21. Online Health Information Retrieval by Consumers and the Challenge of Personal Genomics Mark S. Boguski Introduction Characteristics of Consumer Searches for Health Information What and Where Are Consumers Searching? Personalized Genomics for Consumers Summary and Conclusions References Wikipedia References
■
xi
233 234 236 240 240 240 241 242 242 243 245 246 247 249 249 249 251
252 252 252 253 255 256 256 257
PART II TRANSLATIONAL APPROACHES IN GENOMIC AND PERSONALIZED MEDICINE Section 4 Enabling Strategies in the Translation of Genomics into Medicine 22. Translational Genomics: From Discovery to Clinical Practice Geoffrey S. Ginsburg Introduction A Roadmap for Translation Where Can Genomics Have Impact in the Continuum of Health and Disease? The Genomics “Gold Rush”
261 262 262 263 263 264
xii
■
Table of Contents
The Personal Genome: Precious Code or Fool’s Gold? “Grand Challenges” in Translation of Genomics to Human Health Translational Genomics: Enabling Competencies How Are We Going to Do This? Developing Environments That Foster Translational Genomics to Health Applications References 23. Principles of Study Design Peter Grass Introduction Principles of Experimental Design Design Issues in Genomic Medicine References Recommended Resources 24. Biobanking in the Post-Genome Era Theresa Puifun Chow, Chia Kee Seng, Per Hall and Edison T. Liu Introduction The Biobanking Evolution The Past Imperfect Resources The Evolving Face of Biobanking Existing Models: Biobanking in Europe and the USA Singapore’s National Biobank and National Aspirations In Biomedical Research The Future of National Biobanks References Recommended Resources 25. Application of Biomarkers in Human Population Studies Stefano Bonassi and Monica Neri Introduction Biomarkers in Medicine Biomarkers of Exposure Biomarkers of Early Disease Risk Biomarkers of Genetic Susceptibility to Disease Conclusions References 26. Validation of Candidate Protein Biomarkers Ingibjörg Hilmarsdóttir and Nader Rifai Introduction Optimization of the Candidate Protein Research Assay Analytical Evaluation Reference Intervals Pre-analytical Variation Clinical Evaluation Indicators of Diagnostic Accuracy And Predictability Diagnostic Research Studies Design of Diagnostic Studies
266 266 267
270 272 275 275 276 281 283 283 284
284 285 285 285 286 288 294 296 297
299 299 299 302 303 304 306 306 308 308 308 309 311 311 312 313 316 316
Transferability Of Test Performance Assay Transfer to Diagnostic Company Regulatory Requirements References Recommended Resources 27. Pharmacogenetics and Pharmacogenomics Iris Grossman and David B. Goldstein Introduction Pharmacogenetic Studies: From Concept to Practice Marker Selection – Strategy and Application From Bench to Bedside: Integration of Pharmacogenetic Testing into Clinical Practice Examples of PGx Tests: Promising New Developments and Marketed Products Future Developments Required for the Field to Fully Meet its Expectations References Recommended Resources 28. The Role of Genomics and Genetics in Drug Discovery and Development Robert I.Tepper and Ronenn Roubenoff Introduction The Drug Discovery Process Genomics in Target Discovery Genomic Approaches to Drug Identification Pharmacogenomics and Drug Development Pharmacodynamic Markers and their Role in Drug Discovery and Development Toxicogenomics Genetics and Genomics in Clinical Trial Design Genomics in Drug Approval and Regulation Conclusion References Recommended Resources 29. Role of Pharmacogenomics in Drug Development Colin F. Spraggs, Beena T. Koshy, Mark R. Edbrooke and Allen D. Roses Introduction Drug Development Critical Path Drug Development Economics Methods for Identification of Genetic Classifiers Pharmacogenomics in the Drug Development Pipeline Efficacy Pharmacogenetics – here and Now Drug Exposure Pharmacogenetics to Tune Efficacy and Safety Profiles Investigation and Management of Safety in Clinical Trials Other Genomic Methods: RNA Interference to Direct Drug Usage “No Samples, No Science” Conclusions Acknowledgements References
317 317 318 318 320 321 321 323 325 328 329 331 332 334
335 335 335 336 337 337 338 340 340 340 341 341 342 343
343 344 344 344 346 348 349 350 352 352 354 355 355
Table of Contents
30. Clinical Implementation of Translational Genomics Wendy K. Chung Introduction Genetic Stratification Will Allow Medical Care to be Individualized After a Diagnosis is Made Population-based Germline Genomic Screening Newborn Screening Pharmacogenetics Somatic Genomic Variation Novel Sources of Genomic Variation Laboratory Standards to Ensure Analytic Validity Clinical Validation and Clinical Utility Cost Reimbursement Who Will Provide Genomic Medical Care? Genomic Literacy Ethical, Legal, and Social Issues Conclusions Acknowledgements References Recommended Resources 31. Translating Innovation in Diagnostics: Challenges and Opportunities Matthew P. Brown, Myla Lai-Goldman and Paul R. Billings Introduction Novel Diagnostics Conclusions: Translational Challenges for Innovative Diagnostics References 32. The Role of Genomics in Enabling Prospective Health Care Ralph Snyderman Introduction Predictive Models Predictive Factors Risk Assessment for Breast Cancer Pharmacogenomics Conclusion Acknowledgements References Recommended Resources Section 5 Policy Challenges in Genomic and Personalized Medicine 33. From Sequence to Genomic Medicine: Genome Policy Considerations Susanne B. Haga Introduction Genome Research after the Human Genome Project Policy Issues in Large-scale Genetics and Genomics Research
357 357 357 358 359 360 360 361 361 362 362 363 363 364 364 365 365 365 366
367
367 368 374 375
378 378 380 380 382 383 383 383 384 385
■
Integrating Genomic Medicine Applications in Healthcare Conclusion References
xiii
392 396 397
34. Educational Strategies in Genomic Medicine Charles J. Epstein Introduction Genetic and Genomic Literacy of the Public and Makers of Public Policy Education of the Providers of Health Care Conclusion References
401
35. Federal Regulation of Genomic Medicine Janet Woodcock Introduction Regulation of Genomic Tests Pharmacogenomics in Drug Development And Clinical Medicine: the Role of Regulation Fda Efforts to Advance Genomic Product Development Conclusions References Recommended Resources
414
36. Economic Issues and Genomic Medicine David L.Veenstra, Louis P. Garrison and Scott D. Ramsey Introduction Economic Evaluation and Cost-effectiveness Analysis Evaluating Genomic Technologies Economic Incentives and the Future of Genomic Medicine Establishing Value-based Reimbursement For Genomic Technologies Conclusions References 37. Public–Private Interactions in Genomic Medicine: Research and Development Subhashini Chandrasekharan, Noah C. Perin, Ilse R.Wiechers and Robert Cook-Deegan Introduction Landscape of Private Sector Genomics Future Trends Acknowledgements References
401 402 404 411 411
414 416 418 421 422 422 423 424 424 424 426 428 430 431 432 434
434 435 441 442 442
387 388 388 389 391
Section 6 Genomic Medicine and Public Health 445 38. What Is Public Health Genomics? 446 Alison Stewart and Ron Zimmern Introduction 446 The Emergence of Public Health Genomics 446 The Definition of Public Health Genomics 447 Key Concepts in Public Health Genomics 447
xiv
■
Table of Contents
The “Enterprise” of Public Health Genomics Core Activities in Public Health Genomics Moving Public Health Genomics Forward: Leadership And Networks Conclusion References Recommended Resources
448 449 451 452 452 453
39. Why Do We Need Public Health in the Era of Genomic Medicine? 454 Muin J. Khoury and Marta Gwinn Introduction 454 The Continuum from Genetics to Genomics in Health Practice 454 The Role of Public Health in the Translation of Human Genome Discoveries into Health Applications 455 The Focus on Disease Prevention and Health Promotion 457 The Population Perspective: Crucial Role of Public Health Sciences 457 The Role of Knowledge Integration Across Disciplines 458 The Role of Health Services Research and Population Health Assessment, Assurance, and Evaluation 458 Conclusion 458 References 459 40. Principles of Human Genome Epidemiology Marta Gwinn and Muin J. Khoury Introduction Human Genome Epidemiology Epidemiologic Study Designs Epidemiologic Measures Of Disease Frequency, Association, and Risk Measurement and Bias Gene–environment Interaction Probability and Personalized Medicine Building the Evidence Base Conclusion References Recommended Resources 41. Genomics and Population Screening: Example of Newborn Screening John D.Thompson and Michael Glass Introduction Components of the NBS System Screening Technology: Simple Ideas, Complex Realities Variability Among NBS Programs Influence of Genetics and -omic Technologies on NBS Acknowledgements
461 461 461 462 463 464 465 466 466 467 467 469
470 470 471 471 476 476 477
References Recommended Resources 42. Family History: A Bridge Between Genomic Medicine and Disease Prevention Maren T. Scheuner and Paula W.Yoon Introduction Clinical Approach Conclusion Acknowledgements References Recommended Resources Section 7 Clinical Technologies Supporting Personalized Medicine 43. Molecular Imaging as a Paradigm for Genomic and Personalized Medicine Ralph Weissleder Introduction Molecular Imaging and Cancer Detection Molecular Imaging to Determine Treatment Efficacy Molecular Imaging and Drug Development Near-term Needs and Opportunities Acknowledgement References 44. PET Imaging in Genomic Medicine Vikas Kundra and Osama Mawlawi Introduction Physics Imaging Agents and Methods in Analysis of Biological Samples Conclusion References Recommended Resources 45. MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges Dmitri Artemov Introduction Basics of MRI Contrast MR Contrast Agents for Molecular Imaging Applications Molecular Imaging Applications of MRI Conclusions Acknowledgements References 46. Fluorescence Imaging: Overview and Applications in Biomedical Research Vasilis Ntziachristos Introduction Imaging Technology
477 479
481 481 481 488 488 488 492
493 494 494 494 496 497 497 498 498 500 500 500 503 510 510 511
512 512 513 515 517 520 521 521
524 524 525
Table of Contents
Fluorescence Applications in Genomic Medicine References 47. Imaging Genetics: Integration of Neuroimaging and Genetics in the Search for Predictive Markers Ahmad R. Hariri Introduction Conceptual Basis of Imaging Genetics Basic Principles of Imaging Genetics Imaging Genetics and the Neurobiology of the 5-httlpr Future Directions References 48. Viral Chip Technology in Genomic Medicine Zeno Földes-Papp Introduction Role of Viruses in Human Infectious Disease Microfabrication Nanofabrication Are there Additional Alternatives to Diagnostic Microarrays? Conclusions Acknowledgements References Recommended Resources 49. Vaccines Against Infectious Diseases: A Biotechnology-Driven Evolution Vega Masignani, Hervé Tettelin and Rino Rappuoli Introduction The Genomic Era: From Microbial Genome to Vaccine Development Impact of Whole Genome Analyses In Vivo Gene Expression: Ivet And Stm Microarray Expression Technology Proteomics From Microbial to Human Genome Sequencing: Genomic Medicine Metagenomics: Deciphering Host–microbe Interactions Conclusions References 50. Cancer Vaccines: Some Basic Considerations Hans-Georg Rammensee, Harpreet Singh-Jasuja, Niels Emmerich and Steve Pascolo Introduction Immune Suppression by Tumors and by Regulatory T-cells The Ideal Therapeuticcancer Vaccine Molecularly Undefinedcancer Vaccines Peptides
530 530
532 532 533 534 535 536 536 538 538 538 541 552 553 554 554 554 560
562 562 563 564 566 567 568
■
Proteins and Carbohydrates Nucleic Acids: Plasmid DNA and Messenger RNA Viral and Bacterial Vectors Adjuvants, Formulations, and Route of Application Immunomonitoring Conclusions References 51. Biosensors for the Genomic Age Meghan B. O’Donoghue, Lin Wang,Yan Chen, Gang Yao and Weihong Tan Introduction Biosensors for Detection of Oligonucleotides for the Detection of Disease Nucleic Acid as Tools for Biosensing Outlook References 52. Stem Cells Rikkert L. Snoeckx, Kris Van Den Bogaert and Catherine M.Verfaillie Introduction Types of Stem Cells: Embryonic and Adult Stem Cells How to Define The Molecular Signature of Stem Cells Future Directions to Identify the Global Integrated Regulatory Network Future Directions in Stemcell Therapies Conclusion References 53. Gene Therapy James M.Wilson and Nelson A.Wivel Introduction Gene Delivery Vehicles Gene Therapy Clinical Trials Conclusion References
xv
578 579 582 582 583 584 584 590
590 591 592 596 597 599
599 599 600 601 604 606 606 610 610 611 614 617 617
568 569 570 570 573
573 574 576 576 577
PART III DISEASE-BASED GENOMIC AND PERSONALIZED MEDICINE: GENOME DISCOVERIES AND CLINICAL APPLICATIONS Section 8 Cardiovascular Genomic Medicine 54. The Genomics of Hypertension Chana Yagil and Yoram Yagil Introduction Predisposition Screening Diagnosis
623
624 625 627 627
xvi
■
Table of Contents
Prognosis Pharmacogenomics Monitoring Novel and Emerging Therapeutics Conclusion Acknowledgements References Recommended Resources 55. Lipoprotein Disorders Sekar Kathiresan and Daniel J. Rader Introduction Overview of Lipoprotein Metabolism Plasma Lipid and Lipoprotein Levels and Atherosclerotic Cardiovascular Disease Inherited Basis for Blood Lipid Traits Screening for Lipid Disorders Genetics of Ldl-c Genetics of Hdl-c Genetics of Triglycerides Genetic Lipid Disorders Without Current Proven Molecular Etiology Influence of Lipid-modulating Mutations on Risk of Atherosclerotic Cardiovascular Disease Future Directions in Genetics And Genomics of Lipoproteins Pharmacogenetics of Lipid-modulating Therapies Implications of Genomics of Lipoprotein Metabolism For The Development of Novel Therapies Clinical Recommendations for Genetic Testing for Lipid Disorders Acknowledgements References Recommended Resources 56. Reactive Oxygen Species Signals Leading to Vascular Dysfunction and Atherosclerosis Nageswara R. Madamanchi, Aleksandr E.Vendrov and Marschall S. Runge Introduction Sources of ROS in vascular cells Vascular Dysfunction and Atherosclerosis ROS-induced inflammatory gene expression in vascular cells Association of ROS modulators with atherosclerosis ROS signaling in atherosclerotic risk factors ROS-regulated signaling pathways Regulation of transcription factors by ROS ROS signaling in advanced atherosclerosis Polymorphisms in ROS production genes and atherosclerosis Inhibitors of ROS signaling and vascular disease Conclusion Acknowledgements References
628 630 631 631 631 632 632 633 634 634 634 636 636 637 637 639 642 643 643 644 646 646 647 648 648 651 652
652 652 653 653 654 655 655 656 658 658 658 659 660 660
57. Genomics of Myocardial Infarction Carlos A. Hubbard and Eric J.Topol Introduction Predisposition Screening Strategies Diagnosis of Acute MI Prognostic Implications of MI Pharmacogenomics of MI Novel and Emerging Therapies Conclusion References Recommended Resources
665
58. Acute Coronary Syndromes L. Kristin Newby Introduction Predisposition Screening Diagnosis Prognosis Pharmacogenomics Monitoring Novel and Emerging Therapeutics Conclusion Acknowledgements References Recommended Resources
680
59. Heart Failure in the Era of Genomic Medicine Ivor J. Benjamin and Jeetendra Patel Introduction Predisposition (Genetic and Non-genetic) Screening Pathophysiology Diagnosis Prognosis Pharmacogenomics Monitoring Novel Therapeutics and Future Directions Conclusions and Recommendations Acknowledgements References
692
60. Application of Genomic Methodology for Assessment of Cardiac Transplant Rejection Michael Pham, Mario C. Deng, Jay Wohlgemuth and Thomas Quertermous Introduction Cardiac Allotransplantation as a Definitive Therapy for End-stage Heart Failure The Problem of Allograft Rejection Immunosuppression Strategies to Prevent Rejection Current Strategies for Monitoring Transplant Rejection
665 665 672 673 673 673 674 675 676 679
680 680 681 682 685 686 687 687 687 687 688 690
692 692 693 694 695 696 697 698 701 702 703 703
705
705 705 706 706 707
Table of Contents
The Cargo Clinical Study Development of a Gene Expression Signature For Cardiac Transplant Rejection Pathways Monitored by the Gep (Allomap™) Test Variability of the Biopsy Gold Standard and Relationship to the Gep (Allomap™) Score Discordance Between Biopsy Grade and Molecular Score Effect of Time Post-transplantation on Performance of the Gep Test Relationship of Gep Score to Corticosteroid Dose Relationship of Gep Scores to Cytomegalovirus Infection Prediction of Future Acr by Molecular Score Clinical Use of the Allomap™ Test Future Directions and Ongoing Research With Gep Testing Further Application of Genomic Science to Transplant Rejection References 61. Hypertrophic Cardiomyopathy in the Era of Genomic Medicine J. Martijn Bos, Steve R. Ommen and Michael J. Ackerman Introduction Definitions, Clinical Presentation, and Diagnosis Molecular Genetics of HCM Screening And Treatment for HCM Conclusions References 62. Genetics and Genomics of Arrhythmias Jeffrey A.Towbin and Matte Vatta Introduction Specific Cardiac Arrhythmias Primary Abnormalities in Cardiac Rhythm: Ventricular Tachyarrhythmias Complex Forms of Lqts Short Qt Interval Syndrome Familial Vt/cpvt Primary Conduction Abnormalities References 63. Hemostasis and Thrombosis Richard C. Becker and Felicita Andreotti Introduction Genetics of Coagulation Human Hemostatic Variability Genotype–phenotype Influences Gene-environment Influences on Hemostasis Circulating Cellular and Protein Influences on Hemostasis And Thrombosis
709 709 710 710 710 711 711 711 712 712 713 713 714
716 716 716 718 722 725 725 729 729 729 729 736 741 741 742 745 755 755 755 756 757 758 759
■
Race-related Influences On Hemostasis and Thrombosis Linkage Studies in Thrombosis Association Studies in Thrombosis Heritability And Thrombosis: Existing Complexities A Personalized Approach to Hemostasis and Thrombosis Patient Screening: A Traditional Paradigm Patient Screening: A Comprehensive And Population-based Approach Prognostic Considerations Emerging Platform for Hemostasis and Thrombosis Research References
xvii
760 762 762 763 763 764 767 768 768 769
64. Peripheral Arterial Disease 773 Ayotunde O. Dokun and Brian H. Annex Introduction 773 Epidemiology and Risk Factors for PAD 773 Clinical Manifestations of PAD 773 Therapeutic Strategies for PAD 774 Ic And Cli Are Distinct Clinical Outcomes of PAD 775 Genetic Background as a Risk Factor for PAD 775 Gene Polymorphisms Contributing to Atherosclerosis and PAD 775 Polymorphisms in Pro-atherothrombotic Genes and PAD 776 Genetic Locus Conferring Susceptibility to PAD 776 Identification of Novel Gene Polymorphisms Involved in PAD 776 Identification of a Quantitative Trait in a Preclinical Model of PAD 777 Refining a QTL Using Haplotype Analysis 777 Identification of Candidate Genes 777 Future Potential Use of Genomic Methodologies in PAD 777 Acknowledgements 778 References 778 65. Genomics of Congenital Heart Disease Jessie H. Conta and Roger E. Breitbart Introduction CHD Gene Discovery by Conventional Genetics Genomic Strategies for CHD Gene Discovery Cytogenetic and Molecular Genetic testing Medical Evaluation and Counseling Recommendations Conclusion Acknowledgements References Recommended Resources
781 781 781 786 787 788 789 790 790 793
66. Genomics of Perioperative and Procedural Medicine 794 Simon C. Body, Mihai V. Podgoreanu and Debra A. Schwinn Introduction 794 Why Perioperative Insults are not Equivalent to Chronic Ambulatory Disease 795
xviii
■
Table of Contents
Perioperative Atrial Fibrillation Perioperative Venous and Arterial Thrombosis Perioperative Stroke and Neurocognitive Dysfunction Hemorrhage and Cardiac Surgery Dynamic Genomic Markers of Perioperative Outcomes Conclusion References
796 796 798 799 799 800 800
Section 9 Oncology Genomic Medicine 67. Cancer Genes, Genomes, and the Environment Robert L. Strausberg Introduction Acquired Functions of Cancer Cells Chromosomal Aberrations and Cancer Cancer Genes and their Functions Inherited Predisposition Cellular Progression Toward Cancer Through Somatic Changes From Genome to the Clinic Comprehensive Sequencing of the Kinome Expanding the Search Multiple Molecular Mechanisms for Oncogene Activation Microarrays and Cancer Genomics Environmental Cancer Genomics Cancer Genomic Databases Expedite Progress Mouse Models of Cancer Future Directions References
807 808
68. Immune Cells and the Tumor Microenvironment David S. Hsu, Michael Morse,Timothy Clay, Gayathri Devi and H. Kim Lyerly Introduction Immune Cells of the Tumor Microenvironment Examples of Tissue or Gene Microarrays used to Study Tumors Studies of Genomic Immune Stimulation within the Microenvironment Genomic Analysis of Tumor Microenvironment in Immunotherapy Studies Proteomics of Immune Cells and the Tumor Microenvironment Conclusion References
818
69. Lymphomas Lisa Rimsza Introduction Diffuse Large B-cell Lymphoma Primary Mediastinal Large B-cell Lymphoma Hodgkin Lymphoma Follicular Lymphoma Mantle Cell Lymphoma
808 808 808 809 809 810 811 811 812 812 813 813 813 814 814 815
818 819 822 823 824 825 825 826 830 830 833 835 835 836 838
Burkitt Lymphoma Miscellaneous Lymphomas Clinical Applications of Molecular Assays in Lymphoma References
839 839 840 840
70. Genomics in Leukemias Lars Bullinger, Hartmut Dohner and Jonathan R. Pollack Introduction Genomics in Leukemias-insights into Leukemia Biology Genomics in Leukemias – Evaluation of Drug Effects Genomics in Leukemias – Clinical Outcome Prediction Conclusions Acknowledgments References Recommended Resources
844
71. Genomics of Lung Cancer Hasmeena Kathuria, Avrum Spira and Jerome Brody Introduction Early Diagnosis/screening of Lung Cancer Classification And Prognosis Pathogenesis And Treatment of Lung Cancer Conclusion References Recommended Resource
856
72. Breast Cancer and Genomic Medicine Erich S. Huang and Andrew T. Huang Introduction The Promise Genetic Bases Molecular Bases Prognosis and Prediction Molecular Markers Genomic Insights Netherlands Cancer Institute Study Duke-taipei Study Nsabp Study Pathway Prediction The Reality of Clinical Genomics References 73. Colorectal Cancer G.L.Wiesner,T.P. Slavin and J.S. Barnholtz-Sloan Introduction Genomic Model of CRC Predisposition for CRC Risk Assessment, Evaluation, and Genetic Testing Screening and Surveillance Prognosis and Treatment
844 846 847 849 851 852 852 855
856 857 859 862 865 866 868 869 869 869 871 871 872 872 873 873 875 875 876 876 877 879 879 880 882 887 889 889
Table of Contents
Pharmacogenetics/genomics of Chemoprevention and Chemotherapy Novel and Emerging Therapeutics Conclusion References 74. Prostate Cancer Phillip G. Febbo and Philip W. Kantoff Introduction Genetic Predisposition and Alterations in Prostate Cancer Prostate Cancer Detection Genomic Changes Associated with Prostate Cancer Behavior Genomic Changes associated with HormoneRefractory Prostate Cancer Future Prospects of Genomics in Prostate Cancer Care References 75. Molecular Biology of Ovarian Cancer Tanja Pejovic, Matthew L. Anderson and Kunle Odunsi Introduction Inherited Ovarian Cancer Syndromes Options for Screening and Prevention Genomic Instability and Ovarian Cancer Fanconi/anemia Pathway Somatic Mutations in Ovarian Cancer Oncogenes and Growth Factors Tumor Suppressor Genes Epigenetics in Ovarian Carcinogenesis Ovarian Cancer Metastases Angiogenesis Summary References 76. Pancreatic Neoplasms Asif Khalid and Kevin McGrath Introduction Predisposition (Genetic and Non-Genetic) Screening Diagnosis Prognosis Pharmacogenomics Monitoring Novel and Emerging Therapies Conclusion References
890 892 892 893 898 898 898 901 904 906 907 908 913 913 913 914 914 914 915 916 917 917 917 918 919 919 921 921 922 922 923 926 927 927 927 927 928
77. The Multiple Endocrine Neoplasia Syndromes Y. Nancy You,Vipul Lakhani and Samuel A.Wells
931
Introduction The Multiple Endocrine Neoplasia Syndromes
931 932
Conclusion Acknowledgements References Recommended Resources
■
xix
941 941 942 944
78. Genomics of Head and Neck Cancer Giovana R.Thomas and Yelizaveta Shnayder Introduction Head and Neck Squamous Cell Carcinoma Conclusion References Recommended Resources
945
79. Genomic Medicine, Brain Tumors and Gliomas Sean E. Lawler and E. Antonio Chiocca Introduction Predisposition Screening Diagnosis and Prognosis Pharmacogenomics Monitoring Novel and Emerging Therapeutics Conclusions References
956
80. Molecular Therapeutics of Melanoma Jiaqi Shi,Yonmei Feng, Robert S. Krouse, Stanely Leong and Mark A. Nelson Introduction Diagnosis Genetics of Melanoma Pharmacogenomics Novel and Emerging Therapeutics Conclusions References 81. Emerging Concepts in Metastasis Nigel P.S. Crawford and Kent W. Hunter Introduction Tools to Investigate the Mechanisms of Metastasis Assessement of Prognosis and New Treatments for Metastasis: the Role of New Technologies Conclusion References Recommended Resources 82. Diagnostic-Therapeutic Combinations in the Treatment of Cancer Jeffrey S. Ross Introduction Targeted Therapies for Cancer The Ideal Target The First Diagnostic-therapeutic Combination in Cancer Therapy: Hormonal Therapy for Breast Cancer
945 945 952 953 955
956 957 958 958 961 962 962 963 964 967
967 967 968 969 969 973 973 977 977 979 986 987 988 989 990 990 990 990
992
xx
■
Table of Contents
Diagnostic-therapeutic Combinations for Leukemia and Lymphoma 994 Her-2 Positive Breast Cancer and Trastuzumab 994 (Herceptin®) Other Targeted Anticancer Therapies Using Antibodies 995 Selected Targeted Anticancer Therapies Using Small Molecules 999 Pharmacogenomics 1002 Conclusion 1003 References 1003 Section 10 Inflammatory Disease Genomic Medicine 83. Environmental Exposures and the Emerging Field of Environmental Genomics David A. Schwartz Introduction Importance of Environmental Exposures in Human Health Importance of Environmental Exposures in Studying Disease Processes Comparative Environmental Genomics Exposure Assessment in the Gene-environment Paradigm Challenges and Future of Environmental Genomics Conclusion References
1009 1010 1010 1010 1012 1012 1013 1015 1015 1015
84. Molecular Basis of Rheumatoid Arthritis Robert M. Plenge and Michael E.Weinblatt Introduction Clinical Features Predisposition Screening Diagnosis, Prognosis, and Monitoring Pharmacogenomics Conclusions References
1017
85. “Omics” in the Study of Multiple Sclerosis Francisco J. Quintana and Howard L.Weiner Introduction Genomics in MS Transcriptomics in MS Immunomics in MS Proteomics in MS Conclusion References
1032
86. Inflammatory Bowel Disease Ad A. van Bodegraven and Cisca Wijmenga
1040
Introduction Predisposition (Genetic and Non-genetic)
1017 1018 1018 1024 1024 1024 1026 1026
1032 1032 1034 1035 1036 1037 1037
1040 1041
Screening Diagnosis Prognosis Pharmacogenomics Monitoring Novel and Emerging Therapeutics Conclusions References 87. Glomerular Disorders Tadashi Yamamoto, Hidehiko Fujinaka and Visith Thongboonkerd Introduction Techniques for Detection, Quantification, and Profiling of mRNA Expression in the Kidney mRNA Expression Profiles of Glomerular Disorders Genome Variations in Glomerular Disorders Genetics of Congenital Glomerular Disorders Genomic Medicine for Glomerular Disorders Conclusions References 88. Spondyloarthropathies Dirk Elewaut, Filip De Keyser, Filip Van den Bosch, Dieter Deforce and Herman Mielants Introduction Characteristics of SPA Role of Bowel Inflammation Histopathology of Synovitis in SPA Gut and Synovium Transcriptomes Proteome Analysis Novel and Emerging Therapeutics and Biomarkers Conclusions References 89. Asthma Genomics Scott T. Weiss, Benjamin A. Raby and Juan C. Celedón Introduction Asthma: Basic Pathobiology Predisposition (Genetic and Non-genetic) to Asthma Genome-wide Linkage Analyses of Asthma and its Intermediate Phenotypes Candidate-gene Association Studies of Asthma Genome-wide Association Studies of Asthma Asthma Genomics Screening Diagnosis Prognosis Pharmacogenetics Monitoring Novel and Emerging Therapeutics Conclusions Acknowledgements References Recommended Resources
1043 1044 1045 1047 1047 1048 1049 1050 1056
1056 1056 1059 1060 1062 1062 1064 1064 1067
1067 1067 1070 1072 1074 1075 1075 1077 1078 1084 1084 1084 1085 1085 1087 1088 1088 1088 1091 1091 1091 1092 1093 1093 1093 1093 1097
Table of Contents
90. Genomic Aspects of Chronic Obstructive Pulmonary Disease Peter J. Barnes Introduction Predisposition Pathophysiology Cellular and Molecular Mechanisms Diagnosis and Screening Prognosis Management New Treatments Conclusions References 91. Genomic Determinants of Interstitial Lung Disease P.W. Noble and M.P. Steele Introduction Genetic Determinants of Dpld in Mouse Strains Genetic Determinants of Sarcoidosis Surfactant Proteins and Dpld Genetic Determinants of Pulmonary Fibrosis Identified in Rare Inherited Disorders Genetic Determinants of Fip Conclusion References 92. Peptic Ulcer Disease J. Holton Introduction Clinical and Physiological Aspects of Pud Pathophysiology of Ulcer Formation The Helicobacter Genome Human Polymorphism and Pud Genomics in the Management of Disease Future Developments in the use of Genomic Techniques in Relation to Pud Conclusions Acknowledgements References Recommended Resources 93. Cirrhosis in the Era of Genomic Medicine N.A. Shackel, K. Patel and J. McHutchison Introduction Liver Structure Fibrosis and Cirrhosis Diagnosis of Cirrhosis Treatment of Cirrhosis Genetics of Cirrhosis The Liver Transcriptome The Liver Proteome Development of Liver Fibrosis Transcriptome Analysis of Liver Disease Proteomic Studies of Liver Disease
1098 1098 1098 1099 1101 1104 1105 1105 1107 1107 1108 1110 1110 1111 1111 1114 1115 1115 1117 1118 1122 1122 1123 1124 1125 1130 1131 1132 1132 1133 1133 1137 1138 1138 1138 1140 1140 1141 1142 1143 1143 1144 1146 1148
Proteomics in Other Liver Disease Future Impact of Genomics Studies Conclusion References 94. Systemic Sclerosis Ulf Müller-Ladner Introduction Predisposition Screening Diagnosis Prognosis Pharmacogenomics Monitoring and Genomic Factors Therapeutic Strategies Novel and Emerging Therapeutics Conclusions Acknowledgements References
■
xxi
1150 1151 1151 1151 1155 1155 1155 1158 1159 1159 1159 1159 1162 1164 1165 1165 1165
Section 11 Metabolic Disease Genomic Medicine 95. Genomic Medicine of Obesity J. Alfredo Martínez Introduction Obesity: Causes and Genetic Predisposition Search of Genes Involved in Obesity Diagnosis and Characterization of Genes Associated with Obesity Screening and Diagnosis Prognosis and Gene Based-treatments Novel and Emerging Therapeutics Nutrigenomics, Pharmacogenomics and Gene Therapy Conclusions Acknowledgements References
1169 1170
96. Diabetes Maggie Ng and Nancy J. Cox Introduction GWAS in Type 2 Diabetes Future Research in Type 2 Diabetes Genetics GWAS in Type 1 Diabetes Future Studies in Type 1 Diabetes Clinical Utility of Genetic Research in Diabetes Conclusion References
1187
97. Metabolic Syndrome Rebecca L. Pollex and Robert A. Hegele Introduction Diagnosis: Definition of the MetS Phenotype Pathophysiology of MetS in Brief Genetics of MetS
1170 1170 1172 1175 1179 1180 1182 1183 1183 1184
1187 1188 1190 1191 1192 1192 1192 1192 1194 1194 1194 1196 1196
xxii
■
Table of Contents
Conclusion Acknowledgements References Recommended Resources 98. Nutrition and Diet in the Era of Genomics Jose M. Ordovas and Dolores Corella Introduction Methodological Issues Gene–Nutrient Interactions Path Forward Conclusions Acknowledgements References Section 12 Neuropsychiatric Disease Genomic Medicine 99. The Genetic Approach to Dementia Robert L. Nussbaum Introduction Incidence of Dementia Primary Dementias Clinical Approach to the Dementias Future Prospects for Genomic Medicine in the Dementias Conclusion References Recommended Resources 100. Parkinson’s Disease: Genomic Perspectives Shushant Jain and Andrew B. Singleton Introduction Clinical Characteristics of PD Genetics of PD Genetics of Sporadic PD Conclusion References Recommended Resources
1200 1200 1200 1202 1204 1204 1205 1206 1214 1214 1215 1215
1221 1222 1222 1223 1224 1229 1229 1230 1230 1232 1233 1233 1233 1235 1238 1241 1241 1242
101. Epilepsy Predisposition and Pharmacogenetics Nicole M.Walley and David B. Goldstein Introduction Mendelian Epilepsies Common Epilepsies Future Program of Work References
1243
102. Ophthalmology Janey L.Wiggs Introduction Extraocular Muscles Cornea Lens
1256
1243 1243 1248 1251 1252
1256 1257 1257 1259
Iris Trabecular Meshwork Optic Nerve Retina Genetic Testing for Ocular Disorders Summary References
1259 1259 1260 1260 1261 1261 1261
103. Genomic Basis of Neuromuscular Disorders Erynn S. Gordon and Eric P. Hoffman Introduction Motor Neuron Disease Disorders of the Neuromuscular Junction Disorders of the Muscle Predisposition Screening Diagnosis Prognosis Monitoring Current, Novel, and Emerging Therapies Advances in Genomics and Proteomics Conclusion References Recommended Resources
1265
104. Psychiatric Disorders Stephan Züchner and Ranga Krishnan Introduction Great Prospect, But are We There Yet? Classification Reconsidered How Complex Can it Be? The Value of Rare Genetic Variation Converging Methods Personalized Medicine Conclusion References 105. Genomics and Depression Brigitta Bondy Introduction Diagnosis, Prevalence and Course of Depression Pathophysiological Mechanisms Pharmacogenomics of Antidepressants Current Concepts Future Aspects References 106. Bipolar Disorder in the Era of Genomic Psychiatry Ayman H. Fanous, Frank Middleton, Carlos N. Pato and Michele T. Pato Introduction Diagnosis Predisposition Pharmacogenetics
1265 1265 1266 1268 1268 1273 1274 1274 1275 1276 1277 1279 1279 1281 1282 1282 1282 1283 1283 1284 1284 1285 1286 1287 1289 1289 1289 1290 1293 1294 1296 1296 1299
1299 1299 1300 1304
Table of Contents
The Role of New Technologies in Elucidating the Genetics of BPD 1305 Conclusion 1308 References 1308 Section 13 Infectious Disease Genomic Medicine 107. Genomic Approaches to the Host Response to Pathogens M. Frances Shannon Introduction Genetic Susceptibility to Pathogens Exploring the Host Response Through Expression Profiling Genetical Genomics and Systems Biology: the New Frontiers Application to Clinical Practice Acknowledgement References Recommended Resources 108. Genomic Medicine and AIDS Thomas Hirtzig,Yves Lévy and Jean-François Zagury Introduction Context of HIV and AIDS Predisposition: Susceptibility to HIV-1 Infection Diagnosis Prognosis Monitoring Pharmacogenomics Novel and Emerging Therapeutics Conclusion References 109. Viral Genomics and Antiviral Drug Roberto Patarca Introduction Viral Genomics and the Antiviral Drug Revolution Era Conclusion References
1313 1314 1314 1315 1317 1320 1320 1320 1320 1323 1324 1324 1324 1328 1328 1329 1331 1331 1333 1334 1334 1340 1340 1341 1344 1345
■
xxiii
110. Host Genomics and Bacterial Infections Melissa D. Johnson and Mihai Netea Introduction Genomics and the Study of Bacterial Infections Host Genomics and Gram-positive, Gram-negative and Mycobacterial Infections Future Directions Conclusions References
1347
111. Sepsis and the Genomic Revolution Christopher W.Woods, Robert J. Feezor and Stephen F. Kingsmore Introduction Genetic Polymorphisms Associated with Sepsis Molecular Signatures and Sepsis Therapeutics Conclusion References
1362
112. Diagnosis, Management and Disease Mechanisms of Hepatitis in the Era of Genomic Medicine N.A. Shackel, K. Patel and J. McHutchison Introduction Virology of Hepatitis Viruses Acquisition and Predisposition to Viral Hepatitis Screening and Diagnosis of Viral Hepatitis Pathogenesis of Viral Hepatitis Therapeutics and Pharmacogenomics Future Impact of Genomics Studies Conclusion References
1347 1347 1351 1357 1357 1357
1362 1363 1368 1371 1371 1372
1375 1375 1375 1377 1378 1380 1384 1386 1386 1386
Glossary
1391
Index
1407
This page intentionally left blank
Foreword In the health care field today, there are high expectations for a paradigm shift in care delivery over the coming years. According to this view, we have the opportunity to achieve fundamental change and improvement in the delivery of care, with better health outcomes across the board, led by advances in genomics and molecular medicine. The prospect is exciting: a continuum of health maintenance and medical care that is truly tailored to the individual, based on his or her individual biology. We could know our own genetic profiles from birth. Prevention could be more individually charted, based on individual genetic factors. Disease could be detected and treated earlier through molecular diagnostics. Health care dollars might be spent to much better effect. And our lives could be healthier, longer. The stepping stones to such a future are now widely known. Building on the work of the Human Genome Project and its offspring, medical care can be expected to acquire powerful new tools, especially in diagnosis, that could render care much more effective by making it more precise, more individually targeted, and more predictive. We should be able to prescribe drugs more safely because genetic or other characteristics would help clinicians identify which patients would respond well to a given therapy. As diseases come to be understood at a new level, we should be better able to achieve the right diagnosis and the right treatment for each person, without the trial-and-error process that has long characterized medical treatment. As biomarkers are identified, we should be able to intervene in disease at much earlier stages. And when we are able to know our personal genomic profiles, we should be able to better pinpoint our individual health susceptibilities. Our physicians could give more individualized prevention advice – and perhaps it will even come to pass that we will be more motivated to follow it. I am a believer in this view of the possible future – and I am confident it is not just wishful thinking. The pace of discovery in the genomic field today is unprecedented. Furthermore, enough successful applications of this paradigm already exist to give us reason to look toward a new era of effectiveness in medical care, with new information and new tools for both the clinician and the consumer. At the same time, we must be realistic in our assessment of this future, especially the extent of the efforts that will be needed to achieve it. However desirable the idea of this “paradigm shift” may be, the realization of such a shift rests on a foundation that is still very much under construction.
The shift will require development of new capacities that will enable us to differentiate among the needs of individual patients. In turn, these new capacities will depend on data to be derived, analyzed, and employed on a new scale. Such information demands will surely need to be supported by sophisticated electronic data networks that are yet to be created, informatics tools that are yet to be invented, and clinical decision-support systems that are yet to be devised or adopted. Finally, at the bedrock, the trust and understanding of the medical community and of society at large must be won even as the edifice is being created. This is the work of a generation. It is work that spans professions, economic sectors, and even nations. It is driven by science – but it makes new demands for the rapid translation of scientific discovery into clinical practice and improved health outcomes. One part of the work before us is continued discovery. Phenomenal achievements have been made in mapping the human genome, and a rush of findings is occurring today in understanding associations between genomic factors and health. Yet vast areas still remain to be explored. That work is underway on a global scale. Another part of our work might be called the engineering. This includes the development of interoperable health information technology, with all the implications of that goal: development of technical and clinical standards, adoption of health IT across the health sector, and security of personal health information. In the long term, it should also include the use of health IT to enable us to make faster progress in medical research – and then to feed back what we’ve learned into clinical practice, using IT decision-support tools that are physician-, nurse-, and consumer-friendly. Another element of our task is less defined but equally challenging: the collaboration and cooperation needed to bring this vision to reality. Personalized medicine means care that is information-based at a new level. Gathering that information and using it successfully will transcend many disciplines and make new demands on a health system that is often characterized as “fractured.” It may seem ironic that delivery of individualized care depends on standardization, partnerships, and networks. But these kinds of collaboration are among the most important element of the work that lies ahead. Finally, in today’s health care environment, the success of new products, services, and models of care will depend more than ever on the value that is delivered. At its core, personalized medicine is about care that can achieve new levels in predicting, preventing, and detecting disease. Medical effectiveness of this
xxv
xxvi
■
Foreword
kind should translate into cost effectiveness as well. Achieving and demonstrating high value expectations will be challenging, but this discipline will serve the success of personalized medicine over the long term. The paradigm shift of personalized medicine depends on an extensive foundation of scientific knowledge, professional leadership, health information technology, and a spirit of cooperation, collaboration, and dialogue. The many perspectives offered by the
contributors to this text help us appreciate the breadth of these issues, challenges, debates, and opportunities that lie ahead.
Michael O. Leavitt Secretary Department of Health and Human Services Washington, DC
Preface It seems just yesterday that we were first getting used to the notion of introducing the seemingly freshly uncovered concepts of genetics into the practice of medicine. And yet, with the completion of the Human Genome Project and the rapid development and application of new advances in our ability to understand and query the human genome and its gene set, it is time already to anticipate and outline the early stages of what must be called a transformation of medicine. We are beginning to see the first signs of a fundamental shift in how we behold human physiology and pathology, how we view the concept of what is “normal,” how we consider individuals and their prospects for lifelong health, and how we design healthcare systems that are equally adaptable to the demands of population-wide epidemics and the opportunities for personalized care that utilizes genome-based information to consider individual susceptibility to disease and therapeutic options. Genome-based data, information, knowledge, and eventually wisdom will make possible the kind of healthcare that has been dreamed of since the advent of disease-based medicine early in the 20th century. A system of healthcare that harnesses the might of the genome and its derivatives, along with imaging, clinical and environmental information, will empower physicians and other healthcare providers to do what they have always aspired to do – make medical care as individualized as possible. But this newfound information and knowledge will also allow each of us as consumers of healthcare to take more control of our futures and to develop a more strategic and a prospective approach to health. We stand at the dawn of a profound change in science and medicine’s predictive nature and in our understanding of the biological underpinnings of health and disease. Even in this early light, we can see the outlines of a coming ability to: ●
●
●
● ●
predict individual susceptibility to disease, based on genetic, genomic and other factors; provide more useful tools and individualized programs for disease prevention, based on knowledge of one’s susceptibility; detect the onset of disease earlier and before it is clinically evident, based on newly discovered biological markers that arise from changes at the molecular level; preempt disease progression, as a result of early detection; target medicines and their dose more precisely and safely to each patient, on the basis of a deep understanding of disease
mechanism and the role that genetic and genomic factors play in the individual response to drugs. This revolution in genomic and personalized medicine was anticipated nearly three decades ago by Nobel laureate Paul Berg, who stated so presciently: Just as our present knowledge and practice of medicine relies on a sophisticated knowledge of human anatomy, physiology, and biochemistry, so will dealing with disease in the future demand a detailed understanding of the molecular anatomy, physiology, and biochemistry of the human genome. . . . We shall need a more detailed knowledge of how human genes are organized and how they function and are regulated. We shall also have to have physicians who are as conversant with the molecular anatomy and physiology of chromosomes and genes as the cardiac surgeon is with the structure and workings of the heart.
That time has come. This book is intended to lay out the foundations of this new science, to outline the early opportunities for the practice of medicine to incorporate genome-based analysis into healthcare, and to anticipate the many conditions to which genomic and personalized medicine will apply in the years ahead. The chapters in these volumes are designed to be read either sequentially – introducing the scientific underpinnings of this revolution, exploring aspects of translational medicine and genomics that will be critical for bringing about this revolution, and presenting practical aspects of the first applications of genomic and personalized medicine in the context of specific medical conditions – or one-at-a-time for those interested in particular disorders or approaches. These volumes also describe a field in its infancy, with many challenges for society at large, in addition to those associated with healthcare systems strife with inefficiencies and heterogeneity in their ability to deliver the basics of healthcare. There are “grand challenges” for the visionary science and the clinical care highlighted in these pages. Such challenges include the potential for these innovations to exaggerate existing health disparities, information technology systems that have been described as a “tower of Babel,” an unprepared healthcare work force, and economic incentives that are inadequately aligned for the various stakeholders to fully embrace genomic and personalized medicine. Nonetheless, we are optimistic that the appropriate delivery models and economic incentives will be developed
xxvii
xxviii
■
Preface
in a trustworthy framework that will be embraced by societies around the globe. As an indicator of the importance of personalized medicine, international communities are working together on strategies to overcome these obstacles. In the United States, the Department of Health and Human Services has played a leadership role in this area and a plan was published in 2007 (Personalized Healthcare: Opportunities, Pathways, Resources; see http://www. hhs.gov/myhealthcare/). Other governments (in the United Kingdom, Iceland, Estonia, Luxembourg, and Singapore – to name just a few) have funded initiatives that will secure their place in developing genome-based knowledge and its translation into day-to-day patient care. A collective and global approach to what might be arguably one of the most complex scientific and clinical undertakings in the history of healthcare is undoubtedly what is required. Our international collective of contributors to this work reflects the early adopters and members of a global community of physicians, scientists, and policy makers who will make this happen.
Our intended audience is broad, ranging from medical students (and even the intrepid undergraduate eager to explore this new era of personalized and prospective medicine) to residents and fellows to practitioners in any of the healthcare professions – physicians in any of the medical specialties, surgeons, nurses, genetic (and genomic) counselors, and laboratory directors – and, finally, to members of the genomic and personalized medicine research communities who will, we trust, help write future editions of this text. In times of transformation, we are all students. We hope that this book will help usher in this new era of genomic and personalized medicine and will provide a useful and thorough introduction to the science and practice of this new approach to human health.
Huntington F. Willard, Ph.D. Geoffrey S. Ginsburg, M.D., Ph.D.
Terminology Throughout this book, the terms “genetics” and “genomics” are used repeatedly, both as nouns and in their adjectival forms. Although these terms seem similar, they in fact describe quite distinct (though frequently overlapping) approaches in biology and in medicine. Here, we provide operational definitions to distinguish the various terms and the subfields of medicine to which they contribute. The field of genetics is the scientific study of heredity and of the genes that provide the physical, biological, and conceptual bases for heredity and inheritance. To say that something – a trait, a disease, a code or information – is “genetic” refers to its basis in genes and in DNA. Heredity refers to the familial phenomenon whereby traits (including clinical traits) are transmitted from generation to generation, due to the transmission of genes from parent to child. Genomics is the scientific study of a genome, the complete DNA sequence, containing the entire genetic information of a gamete, an individual, a population or a species. The word “genome” was first used as an analogy with the earlier term “chromosome,” referring to the physical entities (visible under the microscope) that carry genes from one cell to its daughter cells or from one generation to the next. Over the past two decades, “genomics” has given birth to a series of other “-omics” that refer to the comprehensive study of the full complement of, for example, proteins (hence, proteomics), transcripts (transcriptomics), or metabolites (metabolomics). The essential feature of the “-omes” is that they refer to the complete collection of genes, proteins, transcripts, or metabolites, not just to the study of individual entities. Medical genetics is the application of genetics to medicine and is one of the 24 medical specialties recognized by The American Board of Medical Specialties, the preeminent medical organization overseeing physician certification in the United States. Genetic medicine is a term sometimes used to refer to the application of genetic principles to the practice of medicine and thus overlaps medical genetics. Both medical genetics and genetic
medicine approach clinical care largely through consideration of individual genes and their effects on patients and their families. Genomic medicine, by contrast, refers to the use of large-scale genomic information and to consideration of the full extent of an individual’s genome, proteome, transcriptome, or metabolome in the practice of medicine and medical decision-making. The principles and approaches of genomic medicine are relevant well beyond the traditional purview of individual medical specialties and include, as examples, gene expression profiling to characterize tumors or to define prognosis in cancer, genotyping variants in the set of genes involved in drug metabolism or action to determine an individual’s correct therapeutic dosage, scanning the entire genome for millions of variants that influence one’s susceptibility to disease, or analyzing multiple protein biomarkers to monitor therapy and to provide predictive information in presymptomatic individuals. Finally, personalized medicine refers to a rapidly advancing field of healthcare that is informed by each person’s unique clinical, genetic, genomic, and environmental information. The goals of personalized medicine are to take advantage of a molecular understanding of disease to optimize preventive healthcare strategies and drug therapies while people are still well or at the earliest stages of disease. Because these factors are different for every person, the nature of disease, its onset, its course, and how it might respond to drug or other interventions are as individual as the people who have them. For personalized medicine to be used by healthcare providers and their patients, these findings must be translated into precision diagnostic tests and targeted therapies. Since the overarching goal is to optimize medical care and outcomes for each individual, treatments, medication types and dosages, and/or prevention strategies may differ from person to person – resulting in unprecedented customization of patient care. The principles underlying genomic and personalized medicine and their applications to the practice of clinical medicine are presented throughout the chapters that comprise this volume.
Acknowledgements We wish to express our appreciation and gratitude to our many colleagues, especially in the Duke Institute for Genome Sciences & Policy, who have shared their knowledge and ideas about genomic and personalized medicine and who, by doing so, inspired this project. We particularly thank our first editor at Academic Press/Elsevier, Luna Han, who encouraged us to develop the concept of a text on genomic and personalized medicine. We are also grateful to Sally Cheney, Kirsten Funk, and Christine Minihane, our Senior Editors at Academic Press/ Elsevier; to Rogue Shindler, our Developmental Editor; and to Ganesan Murugesan, our Production Project Manager, for their patience, advice, and professionalism in all stages of the project.
We acknowledge our Advisory Board for their suggestions and support and especially thank the nearly 300 authors of the 112 chapters that comprise these volumes. Needless to say, without their efforts, this project could never have come to fruition. We also thank Secretary Mike Leavitt for providing a Foreword for this book, as well as for his enthusiastic support of the concept of genomic and personalized medicine. It gives us pleasure to give special thanks to Kathy Hay and to Lynne Skinner, whose tireless efforts kept us on track and saw this project through to completion. Lastly, we thank our families for their patience and understanding for the many hours we spent creating this, the inaugural edition of Genomic and Personalized Medicine.
xxix
This page intentionally left blank
Abbreviations 2D-DIGE 2D-PAGE
Positron Neutrino Two-dimensional difference gel electrophoresis Two-dimensional polyacrylamide gel electrophoresis 3-MT 3-Methoxytyramine 5-HT 5-Hydroxytriptamine or serotonin 5-HTT 5-HT transporter 5-HTTLPR 5-HTT gene-linked polymorphic region 6-MMP 6-Methylmercaptopurine 6-MP 6-Mercaptopurine 6-MTIMP 6-Methyl-thioinosinemonophosphate 6-TGN 6-Thioguanine nucleosides 6-TIMP 6-Thioinosinemonophosphate 18 18 F-FLT F-labeled 3-deoxy-3-fluorothymidine 18 18 F-FMAU F-1abeled-2-deoxy-2-fluoro--d-arabinofuranosyl thymine AAU Acute anterior uveitis ABCA1 Adenosine triphosphate-binding cassette protein A1 ABCB1 Adenosine triphosphate-binding cassette, sub-family B ABI Ankle brachial index ACC American College of Cardiology ACE Angiotensin converting enzyme aCGH Array comparative genomic hybridization ACR Acute cellular rejection AD Alzheimer’s disease AD Aortic dilation ADH Autosomal dominant hypercholesterolemia ADHD Attention deficit hyperactivity disorder ADIGE Autosomal dominant idiopathic generalized epilepsy
ADJME ADME ADNFLE ADP ADR AED Ago2 AHA AIDS AIMs AJCC ALL ALOX5AP AMACR AML Amol AMR AOO AOTF Apo(a) ApoE APOE AR AR ARACNe ARE ARH ARJP ARM
Autosomal dominant juvenile myoclonic epilepsy Absorption, distribution, metabolism, and elimination Autosomal dominant nocturnal frontal lobe epilepsy Adenosine 5-diphosphate Adverse drug reaction Antiepileptic drug Argonaute 2 mRNA endonuclease American Heart Association Acquired immune deficiency syndrome Ancestry informative markers American Joint Committee on Cancer Acute lymphoblastic leukemia Arachidonate 5-lipooxygenase-activating protein Alpha methylacyl CoA reductase Acute myeloid leukemia Attomole Antibody mediated rejection Age of onset Acousto-optic tunable filters Apolipoprotein(a) Apolipoprotein E Apolipoprotein E (gene name) Androgen receptor Autosomal recessive Algorithm for the Reconstruction of Accurate Cellular Networks Androgen response element Autosomal recessive hypercholesterolemia Autosomal recessive juvenile Parkinsonism Armadillo
xxxi
xxxii
■
ARVC AS ASCA ASCO ASCVD ASD ASO ASPE ATM ATP ATP ATR AZT B1-AR BAC BBB BCAA Bcl BCL-2 Bcl-Abl BDNF BER BEST BH BLAST BLAT BMC BMI BMP-7 BOLD BPDE BPH BRCA BRCA1 BRCA2 BRD2 BSML CA CAD CADASIL CAE CALGB CAP CARD CARGO CAV CBD CBP CBZ CCD
Abbreviations
Arrhythmogenic right ventricular cardiomyopathy Ankylosing spondylitis Anti-Saccharomyces cerevisiae antibodies American Society for Clinical Oncology Atherosclerotic cardiovascular disease Atrial septal defect Anti-sense oligonucleotides Allele-specific primer extension Ataxia telangiectasia mutated Adenosine triphosphate Adult Treatment Panel Ataxia-telangiectasia and RAD3-related Azidothymidine 1-adrenergic receptor Bacterial artificial chromosome Blood–brain barrier Branched-chain amino acids B-cell lymphoma protein B-cell lymphoma 2 Breakpoint cluster region – Abelson kinase fusion protein Brain-derived neurotrophic factor Base excision repair -Blocker Evaluation of Survival Trial Bcl-2 homology domain Basic local alignment and search tool BLAST-like alignment tool Bone marrow cell Body mass index Bone morphogenic protein-7 Blood oxygen level-dependent Benzo (a) pyrene diol epoxide Benign prostatic hypertrophy Breast cancer gene Breast cancer 1, early onset, protein Breast cancer type 2 susceptibility protein Bromodomain containing 2 (gene) Bioinformatic sequence markup language Capsid (viral protein) Coronary artery disease Cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy Childhood absence epilepsy Cancer and leukemia group B College of American Pathologists Caspase recruitment domain (gene) Cardiac Allograft Rejection Gene Expression Observational study Cardiac allograft vasculopathy Cortical basal degeneration CREB-binding protein Carbamazepine Charge coupled device
CCL CCND1 CCP CD CD CDAI CDC CDDP cDNA CDS CED-4 CETP CF CFTR CGAP CGH CGHa CGP CHD CHD CHF CHMP CHRNA4 CIBEX CIC CICR CKD CLI CLIA CML CMT CMV CNS CNV COPA COR COX COX-2 CPOE CRC CREB CRF CRP CS CSA CSF CSF1 CT CTC CTGF CTL CTLA
Chemokine (C-C motif) ligand Cyclin D1, a proto-oncogene Citrullinated cyclic peptide Cluster of differentiation Crohn’s disease Crohn’s disease activity index Centers for Disease Control Cisplatin (cis-diamminedichloroplatinum Complementary deoxyribonucleic acid Clinical decision support Cell death abnormality gene 4 Cholesteryl ester transfer protein Cystic fibrosis Cystic fibrosis transmembrane conductance regulator Cancer Genome Anatomy Project Comparative genomic hybridization Comparative genomic hybridization array Cancer Genome Project Congenital heart disease Coronary heart disease Congestive heart failure Committee for Medicinal Products for Human Use Cholinergic receptor, nicotinic, alpha 4 (gene) Center for Information Biology Gene Expression Citrate/isocitrate carrier Calcium-induced calcium release Chronic kidney disease Critical limb ischemia Clinical Laboratory Improvement Amendments Chronic myelogenous leukemia Charcot-Marie-Tooth disease Cytomegalovirus or human herpes virus 5 Central nervous system Copy number variant Cancer outlier profile analysis C-terminal of ROC domain Cyclooxygenase Cyclooxygenase-2 Computerized provider order entry Colorectal cancer Cyclic AMP response binding protein Circulating recombined form C-reactive protein Cell signaling factor Comparative sequence analysis Cerebrospinal fluid Colony stimulating factor 1 Computed tomography Circulating tumor cells Connective tissue growth factor Cytotoxic T lymphocytes Connective tissue late antigen
Abbreviations
CTSA CX32 CX37 CYP CYP2C19 CYP2D6
Clinical and Translational Science Awards Connexin-32 Connexin-37 Cytochrome P450 Cytochrome P450 2C19 Cytochrome P450 2D6
D2R DAI DBS dbSNP DCM DGGE DGI DHF DHLPC
Dopamine-2 receptor Disease Activity Score (for Ulcerative Colitis) Deep brain stimulation NCBI SNP database Dilated cardiomyopathy Denaturing gradient gel electrophoresis Diabetes Genetics/Initiative Diastolic heart failure Denaturing high performance liquid chromatography Diffuse Lewy body disease Diffuse Lewy body disease Diagonalized linear discriminant analysis Disc large homologue (gene) Drug-metabolizing enzymes Deoxyribonucleic acid Deoxyribonuclease DNA methyltransferases Dihydroxyphenylacetic acid Double stranded Double-stranded ribonucleic acid Direct-to-consumer Duodenal ulcer Dizygotic
DLB DLBD DLDA DLG DMEs DNA DNase DNMTs DOPAC ds dsRNA DTC DU DZ EAE EASS EBI EBV ECL ECM EDG-1 EGAPP EGF EGFR EGP EHR EIA EL ELISA ELMO1 EM EMB EMBL-EBI EMEA ENCODE EOAD
Experimental autoimmune encephalomyelitis East Asia specific sequence European Bioinformatics Institute Epstein–Barr virus Eenterochromaffin-like cells Extracellular matrix Endothelial differentiation gene-1 Evaluation of Genomic Applications in Practice and Prevention Epidermal growth factor Epidermal growth factor receptor Environmental Genome Project Electronic health record Enzyme immunoassay Endothelial lipase Enzyme-linked immunosorbent assay Engulfment and cell motility 1 Extracellular matrix protein Endomyocardial biopsy European Molecular Biology Laboratory– European Bioinformatics Institute European Medicines Evaluation Agency Encyclopedia of DNA Elements Early-onset Alzheimer’s disease
EPIYA eQTL ERK ERNA ESI ESI-MS/MS ESR ESRD EST FA FACS FANCA FANCD2 FAP FCCS FCHL fCJD FCS FCS FD FDA FDB FDG FFPE FGF FGF-2 FH FHTG FISH FKBP FL FLT-3 Fmol fMRI FMT FNA FOV FRET FSGS FTD FT-ICR-MS FUSION GABBR1 GABRG2 GAIN GAPDH GBM GBS GCKR GC-MS
■
xxxiii
Single letter codes for amino acids Expression quantitative trait loci Extracellular signal-regulated kinase Equilibrium radionucleotide angiography Electrospray ionization Electrospray ionization tandem mass spectrometry Erythrocyte sedimentation rate End-stage renal disease Expressed sequence tag Fanconi anemia Fluorescent-activated cell sorting Fanconi anemia proteins A Fanconi anemia protein D2 Familial adenomatous polyposis Two-color fluorescence cross-correlation spectroscopy Familial combined hyperlipidemia Familial Creutzfeldt–Jakob disease Familial hyperchylomicronemia syndrome Fluorescence correlation spectroscopy Familial dysbetalipoproteinemia Food and Drug Administration (USA) Familial defective apoB-100 18 F-labeled 2-deoxy-d-glucose Formalin-fixed, paraffin-embedded Fibroblast growth factor Fibroblast growth factor-2 Familial hypercholesterolemia Familial hypertriglyceridemia Fluorescence in situ hybridization FK506-binding protein Follicular lymphoma FMS-like tyrosine kinase-3 Femtomole Functional MRI Fluorescence molecular tomography Fine needle aspirate Field of view Fluorescence resonance energy transfer Focal segmental glomerulosclerosis Frontotemporal dementia Fourier transform ion cyclotron mass spectrometry Finland–United States Investigation of NIDDM genetics GABA B receptor 1 (gene) GABA A receptor, gamma 2 (gene) Genetic Association Information Network Glyceraldehyde-3-phosphate dehydrogenase Glomerular basement membrane Guillain–Barre Syndrome Glucokinase regulatory protein Gas chromatography–mass spectrometry
xxxiv
■
GCSF GDNF GEFS GEO GEP GFAP GFP GLEPP1 GLP GM-CSF GO GPI GR GSEA GSIS GSTT GU GWAS GZM H HAART HAV HBV HCC HCM HCV HDAC HDL HDM2 HDV HeLa HEPS HER2 HEV HF HFPEF HGF HGPRT HHS HHV-6 HITSP HIV HKMT HL HL7 HLA/HLA HLI HMT HNPCC HNSCC
Abbreviations
Granulocyte colony-stimulating factor Glial cell line-derived neurotrophic factor Generalized epilepsy with febrile seizures plus Gene Expression Omnibus Gene expression profiling Glial fibrillary acidic protein Green fluorescent protein Glomerular epithelial protein 1 Good laboratory practice Granulocye macrophage–colony stimulating factor Gene ontology Globus pallidus interna Glucocorticoid receptor gene Gene set enrichment analysis Glucose-stimulated insulin secretion Glutathione S-transferase Gastric ulcer Genome-wide association study Granzyme Heterozygosity Highly active anti-retroviral therapy Hepatitis A Hepatitis B Hepatocellular carcinoma Hypertrophic cardiomyopathy Hepatitis C Histone deacetylase High-density lipoproteins Human double minute 2 Hepatitis D cervical cancer cell line, also the first cancer cell line established Highly exposed persistently seronegative individuals Human epidermal growth factor receptor 2 Hepatitis E Heart failure Heart failure with preserved ejection fraction Hepatocyte growth factor Hypoxanthine phosphoribosyl transferase Health and Human Services (U.S.) Human herpes virus 6 Healthcare Information Technology Standards Panel Human immunodeficiency virus Histone lysine methyltransferases Hepatic lipase Health Level 7 Human leukocyte antigen/(gene) Hind limb ischemia Histone methyltransferase Hereditary nonpolyposis colorectal cancer Head and neck squamous cell carcinoma
HP-1 HPV HPV-16 HR HRPC HS HSC HSP HSV-1 HSV1-TK HSV-2 hTERT HTLV-I HTLV-III HTS HVJ
Heterochromatin-associated protein-1 Human papilloma virus Human papilloma virus-16 Homologous recombination Hormone refractory prostate cancer Hyper-spectral Hepatic stellate cell Heat shock protein Human herpes simplex virus type 1 Herpes simplex virus type 1-thymidine kinase Human herpes simplex virus type 2 Human telomerase reverse transcriptase Human T-cell leukemia/lymphoma virus type I Human T-cells lymphotropic virus type 3 High-throughput screening Hemagglutinating virus of Japan
IBD IBDQ IC IC ICAM ICAT ICD ICD ICDc
Inflammatory bowel disease Inflammatory bowel disease questionnaire Intermittent claudication Interstitial cystitis Intracellular adhesion molecules Isotope-coded affinity tags Implantable cardioverter defibrillator International Classification of Diseases Cytosolic, NADP-dependent isocitrate dehydrogenase Immature dendritic cells Idiopathic dilated cardiomyopathy Insulin-dependent diabetes mellitus Intermediate-density lipoproteins Indolemaine 2,3-dioxygenase Interferon Interferon-beta Interferon-gamma Immunoglobulin Immunoglobulin-A Insulin-like growth factor-I Immunohistochemistry Integrating the Healthcare Enterprise Inhibitor of B-like (gene) Interleukin Interleukin (4/13) Interleukin-1 beta (gene) Interleukin-1 receptor antagonist Interleukin-6 Interleukin-10 Invasive Monitoring Attenuation through Gene Expression study Immature myeloid cells Integrase (viral protein) Institute of Medicine (U.S.) Insertion sequence Institute for Systems Biology Intreferon stimulated gene
iDC IDCM IDDM IDL IDO IFN IFN- IFN- Ig IgA IGF-I IHC IHE IKBL IL IL (4/13) IL-1 IL-1ra IL-6 IL-10 IMAGE iMC IN IOM IS ISB ISG
Abbreviations
ISHLT IT IVDMIA IVVM JCA JHDM1
International Society for Heart and Lung Transplantation Information technology In vitro diagnostic multivariate index assay In vivo videomicroscopy
JME
Juvenile chronic arthritis JmjC domain-containing histone demethylase-1 Juvenile myoclonic epilepsy
Kb KEGG
Kilobase Kyoto Encyclopedia of Genes and Genomes
LARGO
Lung Allograft Rejection Gene expression Observational study Lymphadenopathy-associated virus Lewy body Lecithin cholesterol acyltransferase Laser capture microdissection Liquid chromatography–mass spectrometry Liquid chromatography/mass spectrometry/ mass spectrometry Liquid crystal tunable filters Linkage disequilibrium Low-density lipoproteins Levodopa Late-onset Alzheimer’s disease Logarithm of the odds Loss of heterozygosity Logical Observation Identifiers, Names, and Codes Line of response Lipoprotein(a) Lysophosphatidic acid Lipoprotein lipase Lipopolysaccharide Leucine-rich repeat Leucine rich repeat kinase 2 Lysine-specific demethylase-1 Lymphotoxin alpha Leukotriene A4 hydrolase Leukotriene B4 Long-term non-progression Left ventricle Left ventricular assist device Left ventricular end-diastolic pressure Left ventricular ejection fraction Left ventricular hypertrophy Left ventricular outflow tract (obstruction)
LAV LB LCAT LCM LC-MS LC-MS/MS LCTF LD LDL l-DOPA LOAD LOD LOH LOINC LOR Lp(a) LPA LPL LPS LRR LRRK2 LSD1 LTA LTA4H LTB4 LTNP LV LVAD LVEDP LVEF LVH LVOT(O) MA mAb MADH6 MAGE-ML MAGUK
Matrix (viral protein) Monoclonal antibodies Mothers against decapentaplegic homolog 6 MicroArray and Gene Expression Markup Language Membrane-associated guanylate kinase
MALDI MALDI-TOF MAM MAO MAOA MAP MAPK MAPK MAQC Mb MBD MCD MCD MCP MCP-1 MCS M-CSF MDM2 MDP MDR MedDRA MEF2A Megsin MEN MeSH MeV MGED MGMT MHC MI MIAME MIM miR-192 miRNA MITF MLEM MLPA MMP MMP-3 MMR MPSS MPTP mQTL MR MRI mRNA mRSS
■
xxxv
Matrix-assisted laser desorption and ionization Matrix-assisted laser desorption ionization time-of-flight Metastasis-averse microenvironment Monoamine oxidase Monoamine oxidate A Mitogen-activated protein Microtubule-associated protein kinase Mitogen-activated protein kinase Microarray quality control Megabase Methylated DNA binding domain protein Malonyl CoA decarboxylase Minimal change disease Monocyte/macrophage chemoattractant protein Monocyte chemoattractant protein 1 Multi-species conserved sequence Macrophage colony-stimulating factor Mouse double minute 2 Muramyl dipeptide Multidrug resistance (gene) Medical Dictionary for Regulatory Activities Terminology Myocyte-enhancing factor 2A Mesangial cell-specific gene with homology to serpin Multiple endocrine neoplasm Medical Subject Headings Multiexperiment viewer Microarray Gene Expression Data society O6-Methylguanine-DNA methyltransferase Major histocompatability complex Myocardial infarction Minimal Information About a Microarray Experiment Metastasis-inclined microenvironment MicroRNA-192 Micro-ribonucleic acid Microphthalmia transcription factor Maximum likelihood expectation maximization Multiplex ligation-dependent probe amplification Matrix metalloproteinase Matrix metalloproteinase-3 Mismatch repair Massively parallel signature sequencing 1-Methyl-4-phenyl-1, 2, 3, 6-tetrahydropyridine Metabolite quantitative trait loci Magnetic resonance Magnetic resonance imaging Messenger ribonucleic acid Modified Rodnan Skin Score
xxxvi
■
MS MS MS MS/MS MSA MTDNA MTHFR mTOR MTP MTX MudPIT MVP MZ NAAT NAD(P)H
Abbreviations
Mass spectrometry Multiple sclerosis Multispectral Tandem mass spectrometry Multiple system atrophy Mitochondrial DNA Methylenetetrahydrofolate reductase (gene) Mammalian target of rapamycin Microsomal triglyceride transfer protein Methotrexate Multidimensional protein identification technology Mitral valve prolapse Monozygotic
NRTI NSAID nt NUD NYHA
Nucleic acid amplification testing Nicotinamide adenine dinucleotide phosphate reduced form Non-alcoholic fatty liver disease Non-alcoholic steatohepatitis N-Acetyltransferase 2 Normal appearing white matter Nucleocapsid (viral protein) Neural cell adhesion molecule National Center for Biotechnology Information New chemical entity National Cholesterol Education Program Refers to a set of 60 cancer cell lines curated at the NCI (National Cancer Institute) Non-coding RNA NEgative regulatory Factor (viral protein) Nucleotide excision repair Neurofibromatosis1/2 National Human Genome Research Institute Non-homologous recombination National Institute of Environmental Health Sciences (NIH) National Institutes of Health Natural killer Nuclear magnetic resonance Non-nucleosidic reverse transcriptase inhibitor Nucleotide oligomerization domain (gene) Negative predictive value Nicotinamide adenine dinucleotide phosphate oxidoreductase Nucleosidic reverse transcriptase inhibitor Non-steroidal anti-inflammatory drug Nucleotides Non-ulcer dyspepsia New York Heart Association
OBO OCTN OHCA OMIM OMP
Open Biomedical Ontologies Organic cation transporter (gene) Out of hospital cardiac arrest Online Mendelian Inheritance in Man Outer membrane protein
NAFLD NASH NAT2 NAWM NC N-CAM NCBI NCE NCEP NCI60 ncRNA NEF NER NF1/2 NHGRI NHR NIEHS NIH NK NMR NNRTI NOD NPV NQO1
OmpC opn OPN OR ORESTES ORF OSEM
Outer membrane porin C of E. coli Osteopontin (gene) Osteopontin (protein) odds ratio Open reading frame expressed sequence tags Open reading frame Ordered subset expectation maximization
p53BP1 PAD PAGE PAI PAM pANCA
p53 Binding protein 1 Peripheral arterial disease Polyacrylamide gel electrophoresis Pathogenicity island Prediction analysis of microarrays Perinuclear antineutrophil cytoplasmatic antibodies Peripheral arterial occlusive disease Papanicolaou Human parvovirus B-19 Primary biliary cirrhosis Peripheral blood lymphocytes Peripheral blood mononuclear cell Pyruvate carboxylase Principal components analysis Proliferating cell nuclear antigen Polymerase chain reaction Proprotein convertase subtilisin/kexin type 9 Procarbazine/CCNU/vincristine Parkinson disease Pharmacodynamic Patent ductus arteriosus Personal digital assistant Parkinson disease with dementia Platelet-derived growth factor Platelet-derived growth factor receptor Pyruvate dehydrogenase Pyrrolidine dithiocarbamate prodynorphin (gene) Positron emission tomography Pulse field gel electrophoresis Prostaglandin E2 Pharmacogenetics and pharmacogenomics Phenytoin Protease inhibitor Phosphotidylinositol triphosphate kinase Prostatic intraepithelial neoplasia Pharmacokinetics Phospholipid transfer protein Polony Multiplex Analysis of Gene Expression Peripheral blood mononuclear cells Pro-myelocytic leukemia retinoic acid receptor oncoprotein Photomultiplier tube Paraoxonase Percutaneous peripheral intervention Positive predictive value
PAOD1 Pap Parvo B19 PBC PBL PBMC PC PCA PCNA PCR PCSK9 PCV PD PD PDA PDA PDD PDGF PDGFR PDH PDTC PDYN PET PFGE PGE2 PGx PHT PI PI3K PIN PK PLTP PMAGE PMBC PML-RAR PMT PON1 PPI PPV
Abbreviations
PR PrEC PRIDE PRNP PS PSA PsA PSC PSP PTEN PTM PUD PZ
Protease (viral) Prostate epithelial cells Proteomics Identifications Database Prion protein (gene) Pulmonary stenosis Prostate-specific antigen Psoriatic arthritis Primary sclerosing cholangitis Progressive supranuclear palsy Phosphatase and tensin homolog Pre-trans-splicing molecule Peptic ulcer disease Plasticity zone
QPCR qRT qRT-PCR
Quantitative PCR Quantitative real-time PCR Quantitative reverse-transcriptase polymerase chain reaction Quantitative trait locus Quadrupole time-of-flight-mass spectrometry Quantitative real-time polymerase chain reaction Quality value
QTL Q-TOF-MS QT-PCR QV R&D RA RAAS RAPD-PCR RB RB RCM REPAIR-AMI
RET REV RFLP RISC RNA RNAi ROC ROCO ROS RP RPA RPLC RRMS RSV RT RT-PCR RT-PCR
Research and Development Rheumatoid arthritis Renin-angiotensin-aldosterone system Random amplified polymorphic deoxyribonucleic acid–polymerase chain reaction Retinoblastoma Retinoblastoma (gene) Restrictive cardiomyopathy Reinfusion of enriched progenitor cells and infarct remodeling in acute myocardial infarction Ret proto-oncogene REgulator of Virion protein expression (viral protein) Restriction fragment length polymorphism RNA interference silencing complex Ribonucleic acid Ribonucleic acid interference RAS in complex proteins ROC (RAS of complex proteins)/COR (C-terminal of ROC) Reactive oxygen species Rapid progression Ribonuclease protection assay Reversed-phase liquid chromatography Relapsing-remitting MS Respiratory syncytial virus Reverse transcriptase (viral protein) Real-time polymerase chain reaction Reverse transcription – polymerase chain reaction
■
xxxvii
RV RVHF
Right ventricle Right ventricular heart failure
SACGHS
Secretary’s Advisory Committee on Genetics, Health and Society (U.S.) Serial Analysis of Gene Expression Scoring algorithm for spectral analysis S-Adenosyl-l-methionine (AdoMet) Significance analysis of microarrays Systolic anterior motion of mitral valve Shrimp alkaline phosphatase/exonuclease I Severe acute respiratory syndrome Single-base extension Sudden cardiac death Severe combined immunodeficiency Voltage-gated sodium channel, type 1, alpha subunit (gene) Selective capture of transcribed sequences Strong cation exchange Standard deviation Standards development organization Shared epitope of HLA (human leukocyte antigen) genes Spongiform encephalopathies System for Evidence-Based Advice through Simultaneous Transaction with an Intelligent Agent across a Network Sinusoidal endothelial cell Surface-enhanced laser desorption and ionization Surface-enhanced laser desorption ionization time-of-flight Surface-enhanced laser desorption/ionization time-of-flight mass spectrometry Serological proteome analysis Systolic heart failure Short hairpin ribonucleic acid Smad-interacting protein 1 Small interfering ribonucleic acid Simian immunodeficiency virus Stevens–Johnson syndrome/toxic epidermal necrolysis Solute carrier family 12 member 3 Systemic lupus erythematosus Sentinel lymph node Stanford Microarray Database Substantia nigra Systematized Nomenclature of Medicine, Clinical Terms Single nucleotide polymorphism Substantia nigra pars compacta Substantia nigra pars reticulata SRY-box containing gene 10 Spondyloarthropathy
SAGE SALSA SAM SAM SAM SAP/Exo I SARS SBE SCD SCID SCN1A SCOTS SCX SD SDO SE SE SEBASTIAN
SEC SELDI SELDI-TOF SELDI-TOF-MS SERPA SHF shRNA SIP1 siRNA SIV SJS/TEN SLC12A3 SLE SLN SMD SN SNOMED CT SNP SNpc SNR SOX10 SpA
xxxviii
■
SPARC SPECT SPMS SPR SPSM-FCS
SR-BI ss SSc SSCP SSRP SSTR2 STAT STEMI STK11 gene STN SU SVAS SVR SWOG TAM TAT TBB TCF7L2 TCR TDM TF TFAP2B TGF TGF- TH Th1 Th2 TIGR TIMP TIMP-1 TK TLE TLR TLS TM TNF
Abbreviations
Secreted protein, acidic, and rich in cysteine, osteonectin Single photon emission computed tomography Secondary progressive MS Surface plasmon resonance Solution (single)-Phase Single-Molecule Fluorescence auto- and two color crossCorrelation Spectroscopy Scavenger receptor, class BI Single-stranded Systemic sclerosis Single strand conformation polymorphism Simple sequence repeat polymorphism Somatostatin receptor type 2 Signal transducers and activators of transcription ST segment Elevation Myocardial Infarction Serine/threonine kinase 11 Subthalmic nucleus Surface (viral protein) Supravalvar aortic stenosis Sustained viral response Southwest Oncology Group Tumor associated macrophages TransActivator of Transcription (viral protein) Transcriptomic-based biomarkers Transcription factor 7-like 2 T-cell receptor Therapeutic drug monitoring Transcription factor Transcription factor activating protein 2 Transforming growth factor Transforming growth factor beta Tyrosine hydroxylase T-helper cell, type 1 T-helper cell, type 2 The Institute for Genomic Research Tissue inhibitor of metalloproteinase Tissue inhibitor of metalloproteinase-1 Tyrosine kinase Temporal lobe epilepsy Toll-like receptor Trans-lesion DNA synthesis Trans-Membrane (viral protein) Tumor necrosis factor
TNF- TNF- TNM TOF TOF TPMT/TPMT Treg TSA TSG TSP
Tumor necrosis factor-alpha Tumor necrosis factor-beta Tumor-node-metastasis stage Tetralogy of Fallot Time-of-flight Thiopurine methyl transferase/(gene) Regulatory T-cells Trichostatin A Tumor suppressor gene Thrombospondin
UADT UC UCSC UMLS UNAIDS US US USF1 USpA UTR
Upper aerodigestive tract Ulcerative colitis University of California at Santa Cruz Unified Medical Language System United Nation mission on AIDS Ultrasound United States of America Upstream transcription factor 1 Undifferentiated spondyloarthropathy Untranslated region (three prime or five prime)
V-ATPase V-CAM VEGF VF VIF VLDL VPR VPU VS VT VZV
Vacuolar adenosine triphosphatase Vascular cell adhesion molecule Vascular endothelial growth factor Ventricular fibrillation Virion infectivity factor Very low-density lipoproteins Viral protein R Viral protein U Ventricular septal defect Ventricular tachycardia Varicella zoster virus
WGA WGAS WGS WHO AIDS
Whole-genome amplification Whole Genome Association Studies Whole genome shotgun World Health Organization mission on AIDS World Health Organization West Nile virus Western-specific sequence Welcome Trust Case Control Consortium
WHO WNV WSS WTCCC XMRV
Xenotropic murine leukemia virus-related virus
ZDV ZFP
Zidovudine Zinc finger protein
Advisory Board Paul R. Billings President and Chief Executive Officer, CELLPOINT DIAGNOSTICS, 265 N. Whisman Road, Mountain View, CA 94043
Raju Kucherlapati Center for Genetics and Genomics, Harvard Medical School, 77 Avenue Louis Pasteur, Ste 250, Boston, MA 02115
Robert Cook-Deegan Center for Genome Ethics, Law & Policy, Duke Institute for Genome Sciences and Policy, Box 90141, Durham, NC 27708
Elizabeth G. Nabel National Heart, Lung and Blood Institute, 31 Center Drive MSC 2486, Building 31, Room 5A52, Bethesda, MD 20892
Kay E. Davies Department of Human Anatomy and Genetics, Oxford University, South Parks Road, Oxford, QX1 3QX, UK
Robert L. Strausberg Human Genomic Medicine, J. Craig Venter Institute, 9704 Medical Center Drive, Rockville, MD 20850
Brian Druker Department of Medicine, Oregon Health & Sciences University, 3181 S.W. Sam Jackson Park Road, Portland, Oregon 97239
Robert I. Tepper Third Rock Ventures, LLC, 29 Newbury St., Boston, MA 02116
Victor Dzau Office of the Chancellor, Duke University Health System, Box 3701, Durham, NC 27710
Janet A. Warrington External RNA Controls Consortium, Affymetrix, 3380 Central Expressway, Santa Clara, CA 95051
Eric Green National Institutes of Health, National Human Genome Research Institute, 50 South Drive, Room 5222, Bethesda, MD 20892-8002
Ralph Weissleder Molecular Imaging Research Center, Massachusetts General Hospital, 149 13th Street, Room 5406, Charlestown MA 02129
Muin J. Khoury Office of Genomics & Disease Prevention, Centers for Disease Control and Prevention, 1600 Clifton Road, Atlanta, GA 30333
Janet Woodcock Food and Drug Administration, 5600 Fishers Lane, Parklawn Buildling, Rm 14-71, Rockville, MD 20857
Mary-Claire King University of Washington, 1705 Northeast Pacific Street, Box 357720, Seattle, WA 98195-7720
xxxix
This page intentionally left blank
Contributors Michael J. Ackerman Mayo Clinic, Windland Smith Rice Sudden Death Genomics Laboratory, Rochester, MN 55905, USA. Matthew L. Anderson Departments of Obstetrics and Gynecology and Pathology, Baylor College of Medicine, Houston, Texas, USA. Felicita Andreotti Institute of Cardiology, Catholic University Medical School, Rome-Italy. Brian H. Annex Division of Cardiovascular Medicine and Department of Medicine, Duke University Medical Center, Durham, NC 27710, USA. Anthony Antonellis Genome Technology Branch, National Human Genome Research Institute, National Institutes of Health, Bethesda, MD, USA. Dmitri Artemov Department of Radiology – MR Research, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA. James R. Bain Departments of Radiology and Internal Medicine, Advanced Imaging Research Center, University of Texas Southwestern Medical Center, Dallas, TX 75235, USA. Peter J. Barnes National Heart and Lung Institute, Section of Airway Disease, London, SW3 6LY, UK. J.S. Barnholtz-Sloan Case Comprehensive Cancer Center, Case Western Reserve University School of Medicine; Department of Epidemiology and Biostatistics, Case Western Reserve University School of Medicine, Cleveland, Ohio, USA. Richard C. Becker Department of Medicine – Cardiovascular Thrombosis Center, Duke University Medical Center, Durham, NC 27705, USA.
Mark Boguski Harvard Medical School, Center for Biomedical Informatics, 10 Shattuck St., Boston, MA 02115 Stefano Bonassi Unit of Molecular Epidemiology, National Cancer Research Institute, Genova, Italy. Brigitta Bondy Section Psychiatric Genetics and Neurochemistry, Psychiatric Hospital of University Munich, Munich, Germany. J. Martijn Bos Mayo Clinic, Windland Smith Rice Sudden Death Genomics Laboratory, Rochester, MN 55905, USA. Roger E. Breitbart Department of Cardiology, Children’s Hospital Boston, Boston, MA 02115, USA. Jerome Brody Department of Medicine, Boston University School of Medicine, Boston, MA 02118, USA. Matthew P. Brown USA.
Omics Consulting LLC, Clayton, CA,
Lars Bullinger Department of Internal Medicine III, University of Ulm, Ulm, Germany. Shawn C. Burgess Sarah W. Stedman Nutrition and Metabolism Center, Department of Pharmacology and Cancer Biology, Duke University Medical Center, Durham, NC 27704, USA. Atul J. Butte Department of Medicine and Department of Pediatrics, Stanford Center for Biomedical Informatics Research, Stanford University School of Medicine, Stanford, CA 94305, USA. David J. Carey Weis Center for Research/Geisinger Clinic, Danville, PA 17822-2607, USA. George Carlson McLaughlin Research Institute, Great Falls, MT, USA
Ivor J. Benjamin Division of Cardiology, Center for Cardiovascular Translational Biomedicine, University of Utah Health Sciences Center, Salt Lake City, UT, USA.
Juan C. Celedón Channing Laboratory, Brigham and Women’s Hospital, Boston, MA; Division of Pulmonary and Critical Care, Brigham & Women’s Hospital, Boston, MA; Harvard Medical School, Boston, MA; Center for Genomic Medicine, Brigham and Women’s Hospital, Boston, MA.
Paul R. Billings President and Chief Executive Officer, CELLPOINT DIAGNOSTICS, 265 N. Whisman Road, Mountain View, CA 94043
Subhashini Chandrasekharan Center for Genome Ethics, Law and Policy, Duke Institute for Genome Sciences and Policy, Durham, NC 27708, USA.
Simon C. Body Department of Anesthesiology, Perioperative & Pain Medicine, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA.
Wing C. (John) Chang Department of Pathology, Center for Lymphoma and Leukemia Research, University of Nebraska Medical Center, Omaha, NE, USA. xli
xlii
■
Contributors
Yan Chen Department of Chemistry, Shands Cancer Center and UF Genetics Institute, University of Florida, Gainesville, FL 32611, USA. Antonio Chiocca Department of Neurological Surgery, Dardinger Laboratory for Neuro-oncology and Neurosciences, The Ohio State University Medical Center and Comprehensive Cancer Center, Columbus, OH 43210, USA. Theresa Puifun Chow Agency for Science, Technology and Research, Singapore Tissue Network, Singapore. Wendy K. Chung Department of Pediatrics, Division of Molecular Genetics, Columbia University, New York, NY 10032, USA. Robert Cook-Deegan Center for Genome Ethics, Law and Policy, Duke Institute for Genome Sciences and Policy, Durham, NC 27708, USA. Dolores Corella Nutrition and Genomics Laboratory, Jean Mayer–U.S. Department of Agriculture, Human Nutrition Research Center on Aging at Tufts University, Boston, MA, USA; Genetic and Molecular Epidemiology Unit, School of Medicine, University of Valencia, Valencia, Spain. Susan Cottrell Amgen, Inc., Seattle, WA 98119, USA. Nancy J. Cox Departments of Medicine and Human Genetics, University of Chicago, Chicago, IL 60637, USA. Nigel P.S. Crawford CCR/NCI/NIH, Laboratory of Cancer Biology and Genetics, National Cancer Institute, Bethesda, MD 20892-4264, USA. A. Jamie Cuticchia Bioinformatics Group, Duke Comprehensive Cancer Center, Durham, NC 27708, USA. Dieter Deforce Laboratory of Pharmaceutical Biotechnology, Faculty of Pharmaceutical Sciences, Ghent University, Ghent, Belgium. Mario C. Deng
Columbia University, New York, NY, USA.
Gayathri Devi Department of Surgery, Duke University Medical Center, Durham, NC 27710, USA.
Dirk Elewaut Department of University Hospital, Ghent, Belgium.
Rheumatology, Ghent
Charles J. Epstein Department of Pediatrics and Institute for Human Genetics, University of California, San Francisco, CA 94143, USA. Ayman H. Fanous Washington VA Medical Center, Washington, DC, 20422, USA. Phillip G. Febbo Institute for Genome Sciences and Policy, Department of Medicine – Oncology, Duke University Medical Center, Durham, NC 27710, USA. Robert J. Feezor Division of Vascular Surgery and Endovascular Therapy, University of Florida College of Medicine, Gainesville, FL. Yonmei Feng Department of Pathology and the Arizona Cancer Center, University of Arizona, Tucson, AZ, USA. Zeno Földes-Papp ISS, National Center of Fluorescence, Champaign, Illinois 61822, USA. Hidehiko Fujinaka Institute for Clinical Research, Niigata National Hospital, Niigata 951-8585, Japan. David J. Galas Institute for Systems Biology, Seattle, WA, USA; Battelle Memorial Institute, Columbus, OH, USA. Louis, P. Garrison Department of Pharmacy, University of Washington, Seattle, WA 98195, USA. Glenn S. Gerhard Weis Center for Research/Geisinger Clinic, Danville, PA 17822-2607, USA. Geoffrey S. Ginsburg Center for Genomic Medicine, Duke Institute for Genome Sciences and Policy, Durham, NC 27710, USA. Bjorn T. Gjertsen Institute of Medicine, Hematology, Haukeland University Hospital, University of Bergen, Bergen, Norway. Michael Glass Newborn Screening Program, Washington State Department of Health, Shoreline,WA 98155, USA.
Juergen Distler Epigenomics, AG, Berlin, Germany.
David B. Goldstein Center for Population Genomics and Pharmacogenetics, Duke Institute for Genome Sciences and Policy, Durham, NC 27710, USA.
Harmut Dohner Department of Internal Medicine III, University of Ulm, Germany.
Erynn S. Gordon Division of Pediatric Genetics, University of Maryland School of Medicine, Baltimore, MD 21201, USA.
Ayotunde O. Dokun Metabolism and Nutrition.
Endocrinology
Tucker Gosnell Massachussetts General Hospital Cancer Center, Boston, MA, USA.
Mark R. Edbrooke Pharmacogenetics, GlaxoSmithKline, Durham, NC 27709, USA.
Peter Grass Biomarker Development, Novartis Pharma AG, Basel, Switzerland.
Lucas B. Edelman Department of Bioengineering and Institute for Genomic Biology, University of Illinois, Urbana-Champaign.
Eric D. Green Genome Technology Branch, National Human Genome Research Institute, National Institutes of Health, Bethesa, MD 20892, USA.
Theo deVos
Epigenomics, Inc, Seattle, WA, USA.
Division
of
Contributors
■
xliii
Iris Grossman Pharmacogenetics, Research and Development, GlaxoSmithKline, Research Triangle Park, NC 27709, USA.
Chunhwa Ihm Department of Pathology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA.
Marta Gwinn National Office of Public Health Genomics, Centers for Disease Control and Prevention, Atlanta, GA 30341, USA.
Olga Ilkayeva Sarah W. Stedman Nutrition and Metabolism Center, Department of Pharmacology and Cancer Biology, Duke University Medical Center, Durham, NC 27704, USA.
Carolina Haefliger
Rafael Irizarry Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health.
Epigenomics, AG, Berlin, Germany.
Susanne B. Haga Duke Institute for Genome Sciences and Policy, Durham, NC 27710, USA. Per Hall Department of Medical Epidemiology Biostatistics, Karolinska Institute, Stockholm, Sweden.
and
Joshua M. Hare Division of Cardiology and Interdisciplinary Stem Cell Institute, University of Miami, Miller School of Medicine, Miami FL 33136, USA. Ahmad Hariri Developmental Imaging Genetics Program, University of Pittsburgh, Pittsburgh, PA 15213, USA. James R. Heath Department of Chemistry, California Institute of Technology, Los Angels, CA, USA.
Shushant Jain Laboratory of Neurogenetics, National Institute on Aging, National Institutes of Health, Bethesda, MD, USA; Reta Lila Weston Institute of Neurological Studies, University College London W1T 4JF, UK; Department of Molecular Neuroscience, Institute of Neurology, University College London, London, UK. Melissa D. Johnson Department of Medicine – Infectious Diseases, Duke University Medical Center, Durham, NC 27710, USA. Philip W. Kantoff Dana-Farber Cancer Institute, Department of Medical Oncology, Boston, MA, USA.
Robert A. Hegele Robarts Research Institute, Blackburn Genetics Laboratory, London, ONT, Canada.
Sekar Kathiresan Massachusetts General Hospital, Cardiovascular Disease Prevention Center, Boston, MA, 02114, USA.
Bettina Heidecker Division of Cardiology and Interdisciplinary Stem Cell Institute, University of Miami, Miller School of Medicine, Miami, FL, USA.
Hasmeena Kathuria Department of Medicine, Boston University School of Medicine, Boston, MA 02118, USA.
Shona Hislop Department of Pathology, Brigham and Women’s Hospital, Boston, MA 02115, USA. Eric P. Hoffman Research Center for Genomic Medicine, Children’s National Medical Center, Washington, DC 20010, USA. John Holton Centre for Infectious Diseases and International Health, Windeyer Institute of Medical Sciences, Royal Free and University College London Medical School, London, W1T 4JF, UK. Leroy Hood The Institute for Systems Biology, Seattle, WA 98103-8904, USA. Andrew T. Huang Koo Foundation Sun Yat-Sen Cancer Center, Taiwan. Erich S. Huang Duke University Medical Center, Durham, NC, USA. Carlos A. Hubbard Southern Medical Group, Tallahassee, FL 32308, USA. Kent Hunter CCR/NCI/NIH, Laboratory of Cancer Biology and Genetics, National Cancer Institute, Bethesda, MD 20892-4264, USA. Courtney Hyland Department of Pathology, Brigham and Women’s Hospital, Boston, MA 02115, USA.
Kensaku Kawamoto Division of Clinical Informatics, Duke University Medical Center, Durham, NC 27710, USA. Filip De Keyser Department of Rheumatology, Ghent University Hospital, Ghent, Belgium. Asif Khalid Division of Gastroenterology, Hepatology and Nutrition, University of Pittsburgh, Pittsburgh, PA 15213, USA. Muin Khoury National Office of Public Health Genomics, Centers for Disease Control and Prevention, Atlanta, GA 30341, USA. Stephen F. Kingsmore Resources.
National Center for Genome
Michelle M. Kittleson Division of Cardiology, UCLA School of Medicine, Los Angeles, CA, USA. Beena T. Koshy Pharmacogenetics, GlaxoSmithKline, Durham, NC 27709, USA. Ranga Krishnan Department of Psychiatry and Behavioral Science, Duke University Medical Center, Durham, NC 27710, USA. Robert S. Krouse Southern Arizona Veterans Affairs Health Care System, Tucson, AZ, USA. Vikas Kundra Diagnostic Radiology, University of Texas MD Anderson Cancer Center, Houston, TX 77030, USA.
xliv
■
Contributors
Myla Lai-Goldman Laboratory Corporation of America, Holdings, Burlington, NC, USA.
Herman Mielants Department of Rheumatology, Ghent University Hospital, Ghent, Belgium.
Vipul Lakhani Division of Endocrinology, Department of Medicine, Duke University School of Medicine, Durham, NC, 27710, USA.
Michael J. Muehlbauer Sarah W. Stedman Nutrition and Metabolism Center, Department of Pharmacology and Cancer Biology, Duke University Medical Center, Durham, NC 27704, USA.
Robert Langer Weis Center for Research/Geisinger Clinic, Danville, PA 17822-2607, USA. Sean E. Lawler Department of Neurological Surgery, The Ohio State University Medical Center, Wiseman Hall Columbus, OH 43210, USA. Inyoul Lee
Institute for Systems Biology, Seattle, WA, USA.
Ulf Müller-Ladner Department of Internal Medicine and Rheumatology, Justus-Liebig University Giessen, Bad Nauheim, Germany. Mark A. Nelson Department of Pathology, University of Airzona, Tucson, AZ 85724, USA.
Charles Lee Department of Pathology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA 02115, USA.
Monica Neri Unit of Molecular Epidemiology, National Cancer Research Institute, Genoa, Italy.
Arthur S. Lee Department of Pathology, Brigham and Women’s Hospital, Boston, MA 02115, USA.
Mihai Netea Department of Medicine, University Medical Center St. Radboud, Nijmegen, The Netherlands.
Stanely Leong Department of Surgery, UCSF Medical Center at Mount Zion, University of California, San Francisco, CA, USA.
L. Kristin Newby Department of Medicine – Cardiology, Duke University Medical Center, Durham, NC 27710, USA.
Ralf Lesche Epigenomics, AG, Berlin, Germany. Samuel Levy J. Craig Venter Institute, Rockville, MD 20850, USA. Edison T. Liu Agency for Science, Technology and Research, Singapore Tissue Network, Genome Institute of Singapore, Singapore. David F. Lobach Division of Endocrinology and Metabolism, Duke University Medical Center, Durham, NC 27710, USA. James B. Lorens Department of Biomedicine, University of Bergen, Bergen, Norway. Aldons J. Lusis Department of Medicine, Division of Cardiology, University of California, Los Angeles, CA 900951679, USA. H. Kim Lyerly Duke Comprehensive Cancer Center, Duke University Medical Center, Durham, NC 27705, USA. Nageswara R. Madamanchi Department of Medicine, Carolina Cardiovascular Biology Center, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-7126, USA. J. Alfredo Martinez Department of Physiology and Nutrition, University of Navarra, Pamplona, Spain.
Christopher B. Newgard Duke Independence Park Facility, Duke University Medical Center, Durham, NC 27704, USA. Maggie Ng Departments of Medicine and Human Genetics, Chicago, IL 60637, USA. Paul W. Noble Division of Pulmonary, Allergy and Critical Care Medicine, Duke University Medical Center, Box 3171, Durham, NC 27710, USA. Vasilis Ntziachristos Technical University of Munich, Helmholtz Center Munich, Munich, Germany. Robert L. Nussbaum Division of Medical Genetics, Department of Medicine & Institute for Human Genetics, University of California, San Francisco, 513 Parnassus Ave, San Francisco, CA 94143 Meghan B. O’Donoghue Department of Chemistry, Shands Cancer Center and UF Genetics Institute, University of Florida, Gainesville, FL 32611, USA. Kunle Odunsi Departments of Gynecologic Oncology and Immunology, Roswell Park Cancer Institute, Buffalo, NY, USA.
Kevin McGrath Division of Gastroenterology, Hepatology and Nutrition, University of Pittsburgh, Pittsburgh, PA 15213, USA.
Kenneth Olden Laboratory of Molecular Carcinogenesis, NIEHS, National Institutes of Health, Department of Health and Human Services, Research Triangle Park, NC 27709-2233, USA.
John McHutchison Department of Gastroenterology, Duke University Medical Center, Duke University Clinic Research Institute, Durham, NC, USA.
Steve R. Ommen Mayo Clinic, Windland Smith Rice Sudden Death Genomics Laboratory, Rochester, MN 55905, USA.
Frank Middleton Upstate Medical University, Washington VA Medical Center, Washington, DC, USA.
Jose M. Ordovas Nutrition and Genomics Laboratory, Tufts University, Boston, MA 02111-1524, USA.
Vega Masignani
Novartis Vaccines, Siena, Italy.
Contributors
Roberto Patarca E. M. Papper Lab of Clinical Immunology and Molecular Biology, University of Miami School of Medicine, Sunny Isles Beach, FL 33160, USA. Jeetendra Patel Division of Cardiology, Department of Medicine, University of Utah School of Medicine, Salt Lake City, UT, USA. K. Patel Department of Gastroenterology, Duke University Medical Center, Duke University Clinic Research Institute, Durham, NC, USA. Carlos N. Pato Center for Neuropsychiatric Genetics and Department of Psychiatry, Department of Neuroscience and Physiology, Upstate Medical University, Syracuse, NY.
■
xlv
Sridhar Ramaswamy Medical Oncology, Dana-Farber Cancer Institute, Boston, MA 02115, USA. Hans-Georg Rammensee Interfakultares Institut fur Zellbiologie, Abt. Immunologie, D-72076 Tubingen, Germany. Scott D. Ramsey Department of Pharmacy, University of Washington, Seattle, WA 98195, USA. Rino Rappuoli Novartis Vaccines, Siena, Italy. Nader Rifai Laboratory Medicine, Children’s Hospital Boston, Boston, MA 02115, USA. Lisa Rimsza Department of Pathology, University of Arizona, Tucson, AZ 85724-5043, USA.
Michele T. Pato Center for Neuropsychiatric Genetics and Department of Psychiatry, Department of Neuroscience and Physiology, Upstate Medical University, Syracuse, NY.
Yu-Hui Rogers J. Craig Venter Institute, Rockville, MD 20851, USA.
Tanja Pejovic Department of Obstetrics and Gynecology, Oregon Health Sciences University, Portland, OR 97239, USA.
Allen D. Roses Deane Drug Discovery Institute, Duke University Medical Center, One Science Drive, Suite 342, Durham, NC 27708.
Noah C. Perin Tularik, Inc., South San Francisco, CA 94080, USA. Michael Pham Division of Cardiovascular Stanford University, Stanford, CA 94305, USA.
Medicine,
Robert M. Plenge Division of Rheumatology, Harvard Medical School, Brigham and Women’s Hospital, Boston, MA 02115, USA. Achim Plum Epigenomics, AG, Berlin, Germany. Mihai V. Podgoreanu Department of Anesthesiology, Duke University Medical Center, Durham, NC 27710, USA. Jonathan R. Pollack University, CA, USA.
Department of Pathology, Stanford
Rebecca L. Pollex Robarts Research Institute, London, ONT, Canada. Nathan D. Price Department of Chemical and Biomolecular Engineering, Department of Bioengineering and Institute for Genomic Biology, University of Illinois, UrbanaChampaign, IL USA. Thomas Quertermous Division of Cardiovascular Medicine, Stanford University, Stanford, CA 94305, USA. Francisco J. Quintana Center for Neurologic Diseases, Harvard Medical School, Boston, USA. Benjamin A. Raby Channing Laboratory, Brigham and Women’s Hospital, Boston, MA; Division of Pulmonary and Critical Care, Brigham & Women’s Hospital, Boston, MA; Harvard Medical School, Boston, MA; Center for Genomic Medicine, Brigham and Women’s Hospital, Boston, MA. Daniel J. Rader University of Pennsylvania School of Medicine, Philadelphia, PA 19104-6160, USA.
Jeffrey S. Ross Department of Pathology and Laboratory Medicine, Albany Medical College MC-80, Albany, NY 12208, USA. Ronenn Roubenoff Biogen Idec, Inc.
Immunology
Medical
Research,
Marschall S. Runge Department of Medicine, University of North Carolina-Chapel Hill, Chapel Hill, NC 27514, USA. Maren T. Scheuner 90407-2138, USA.
Rand Corporation, Santa Monica, CA
Matthias Schuster Epigenomics, AG, Berlin, Germany. David A. Schwartz Center for Genetics and Therapeutics, National Jewish Medical and Research Center, 1400 Jackson Street, Denver, CO 80206 Debra A. Schwinn Department University of Washington, Seattle, USA.
of
Anesthesiology,
Chia Kee Seng Centre for Molecular Epidemilogy, Yong Loo Lin School of Medicine, National University of Singapore, Singapore. Nicholas A. Shackel Department of Gastroenterology, Royal Prince Alfred Hospital, Sydney, Australia. M. Frances Shannon Division of Molecular Bioscience, John Curtin School of Medical Research, Australian National University, Canberra ACT 2601 Australia. A. Dean Sherry Departments of Radiology and Internal Medicine, Advanced Imaging Research Center, University of Texas Southwestern Medical Center, Dallas,TX 75235, USA. Jiaqi Shi Department of Pathology and the Arizona Cancer Center, University of Arizona, Tucson, Arizona, USA.
xlvi
■
Contributors
Kevin Shianna Duke Institute for Genome Sciences and Policy, Duke University, Durham, NC 27710, USA. Yelizaveta Shnayder Department of Otolaryngology – Head and Neck Surgery, University of Kansas School of Medicine, Kansas City, KS, USA. Andrew B. Singleton Laboratory of Neurogenetics, Molecular Genetics Unit, NIH, National Institute on Aging, Bethesda, MD 20892, USA. T.P. Slavin Center for Human Genetics, University Hospitals Case Medical Center; Case Comprehensive Cancer Center, Case Western Reserve University School of Medicine; Department of Pediatrics, University Hospitals Case Medical Center, Cleveland, Ohio, USA. Desmond J. Smith Department of Molecular and Medical Pharmacology, Geffen School of Medicine, UCLA, Los Angeles, CA 90095-1735, USA. Rikkert L. Snoeckx Stem Cell Institute Leuven, Catholic University Leuven, Leuven, Belgium.
Hervé Tettelin
Novartis Vaccines, Siena, Italy.
Giovana R. Thomas Department of Otolaryngology – Head & Neck Surgery, University of Miami School of Medicine, Miami, FL 33136, USA. John D. Thompson Newborn Screening Program,Washington State Department of Health, Shoreline,WA 98155, USA. Visith Thongboonkerd Medical Molecular Biology Unit, Office for Research and Development, Siriraj Hospital, Mahidol University, Bangkoknoi, Bangkok 10700, Thailand. Eric J. Topol Scripps Clinic Division of Cardiovascular Disease, La Jolla, CA 92037, USA. Jeffrey A. Towbin Pediatric Cardiology, Baylor College of Medicine, Houston, TX 77030, USA. Ad A. van Bodegraven Department of Gastroenterology, VU University Medical Centre, Amsterdam, The Netherlands. Kris Van Den Bogaert Stem Cell Institute Leuven, Catholic University Leuven, Leuven, Belgium.
3059,
Filip Van den Bosch Department of Rheumatology, Ghent University Hospital, Ghent, Belgium.
Avrum Spira Department of Medicine, Boston University School of Medicine, Boston, MA 02118, USA.
Matte Vatta Departments of Pediatrics (Cardiology), Baylor College of Medicine, Texas Children’s Hospital, Houston, TX, USA.
Colin F. Spraggs Pharmacogenetics, Durham, NC 27709, USA.
GlaxoSmithKline,
Timothy D. Veenstra SAIC Frederick, Inc., National Cancer Institute, Frederick, MD 21702-1201, USA.
Robert D. Stevens Sarah W. Stedman Nutrition and Metabolism Center, Department of Pharmacology and Cancer Biology, Duke University Medical Center, Durham, NC 27704, USA.
David L. Veenstra Department of Pharmacy, University of Washington, Seattle, WA 98195, USA.
Ralph Snyderman Duke Durham, NC 27710, USA.
University,
DUMC
Nicolas A. Stewart SAIC Frederick, Inc., National Cancer Institute, Frederick, MD 21702-1201, USA. Alison Stewart Strangeways Research Laboratory, Public Health Genetics Unit, Worts Causeway, Cambridge, CB1 8RN, UK. F. Stewart Weis Center for Research/Geisinger Clinic, Danville, PA 17822-2607, USA. Robert L. Strausberg J. Craig Venter Institute, Rockville, MD 20850, USA.
Aleksandr E. Vendrov Department of Medicine, Carolina Cardiovascular Biology Center, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-7126, USA. Catherine M. Verfaillie Stem Cell Institute Leuven, Catholic University Leuven, Leuven, Belgium. Nicole M. Walley Center for Population Genomics and Pharmacogenetics, Duke University Medical Center, Durham, NC 27710, USA. Ling Wang Department of Chemistry, Shands Cancer Center and UF Genetics Institute, University of Florida, Gainesville, FL 32611, USA.
Moshe Szyf Department of Pharmacology and Therapeutics, McGill University, Montreal, Quebec H3G 1Y6, Canada.
Mike Weale Statistical Genetics Unit, Department of Medical and Molecular Genetics, King’s College London, Guy’s Hospital, London SE1 9RT, UK.
Weihong Tan Department of Chemistry, Shands Cancer Center and UF Genetics Institute, University of Florida, Gainesville, FL 32611, USA.
Michael E. Weinblatt Division of Rheumatology, Harvard Medical School, Brigham and Women’s Hospital, Boston, MA 02115, USA.
Robert I. Tepper Third Rock Ventures, LLC, Boston, MA 02116, USA.
Howard L. Weiner Center for Neurologic Diseases, Harvard Medical School, Boston, MA 02115, USA.
Contributors
Scott T. Weiss Department of Medicine, Harvard Medical School, Boston, MA 02115, USA. Ralph Weissleder Center for Systems Massachusetts General Hospital, Boston, MA.
Biology,
Samuel A. Wells Department of General Surgery, Duke University Medical Center, Durham, NC 27705, USA. Brett R. Wenner Sarah W. Stedman Nutrition and Metabolism Center, Department of Pharmacology and Cancer Biology, Duke University Medical Center, Durham, NC 27704, USA. Ilse R. Wiechers Massachusetts General/McLean Hospital, Adult Psychiatry Residency Program, Boston, MA 02114, USA. Georgia L. Wiesner School of Medicine – Department of Genetics, Case Western Reserve University, Cleveland, Ohio 44106-4955, USA. Janey L. Wiggs Department of Opthalmology, Harvard Medical School and the Massachusetts Eye and Ear Infirmary, Boston, MA 02114, USA. Cisca Wijmenga Department of Genetics, University Medical Center Groningen, Groningen, The Netherlands.
■
xlvii
Christopher W. Woods Department of Medicine – Infectious Diseases, Duke University Medical Center, Durham, NC 27705, USA. Paula W. Woon National Center for Public Health Genomics, Centers for Disease Control and Prevention, Atlanta, GA 30341, USA. Chana Yagil Department of Nephrology and Hypertension and Laboratory for Molecular Medicine, Ben-Gurion University, Barzilai Medical Center Campus, Ashkelon, Israel. Yoram Yagil Department of Nephrology and Hypertension, Ben-Gurion University, Barzilai Medical Center Campus, Ashkelon, Israel. Tadashi Yamamoto Department of Structural Pathology, Institute of Nephrology, Niigata University Graduate School of Medical and Dental Sciences, Niigata 951-8510, Japan. Gang Yao Department of Chemistry, Shands Cancer Center and UF Genetics Institute, University of Florida, Gainesville, FL 32611, USA. Andrew J. Yee Massachusetts General Hospital Cancer Center, Harvard Medical School, Boston, MA 02115, USA. Hyuntae Yoo Institute for Systems Biology, Seattle, WA, USA.
Huntington F. Willard Duke Institute for Genome Sciences and Policy, Durham, NC 27710, USA.
Y. Nancy You Department of Surgery, Mayo Clinic, Rochester, MN 55905, USA.
James M. Wilson Gene Therapy Program, Translational Research Lab, University of Pennsylvania, Philadelphia, PA 19104-3403, USA.
Li-Rong Yu SAIC Frederick, Inc., National Cancer Institute, Frederick, MD 21702-1201, USA.
Nelson A. Wivel Gene Therapy Program, Translational Research Lab, University of Pennsylvania, Philadelphia, PA 19104-3403, USA.
Jean-François Zagury Conservatoire National des Arts et Metiers, Chaire de Bioinformatique, 292, rue Saint-Martin, 75003 Paris.
Jay Wohlgemuth Via Pharmaceuticals, San Francisco, CA, USA.
Ron Zimmern Strangeways Research Laboratory, Public Health Genetics Unit, Worts Causeway, Cambridge, CB1 8RN, UK.
Janet Woodcock MD 20857, USA.
Stephan Züchner University of Miami School of Medicine, Miami, FL 33136.
Food and Drug Administration, Rockville,
This page intentionally left blank
PART
One
Genomic Approaches to Biology and Medicine
1
This page intentionally left blank
Section
Principles of Human Genomics
1. 2. 3. 4. 5. 6.
1
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine Concepts of Population Genomics Genomic Approaches to Complex Disease Human Health and Disease: Interaction Between the Genome and the Environment Epigenomics and Its Implications for Medicine Systems Biology and the Emergence of Systems Medicine
CHAPTER
1 Organization, Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine Huntington F. Willard
INTRODUCTION The genetic variation that can influence health and disease has been a central, if not widely practiced, principle of medicine for over a hundred years, since the prescient observations of the British physician Sir Archibald Garrod established the concept of “chemical individuality” over a century ago (Garrod, 1902). What has limited broad application of this principle until now has been the generally presumed rarity or special nature of clinical circumstances or conditions to which genetic variation was relevant – rare disorders such as Garrod’s alkaptonuria, inherited conditions limited to specific populations such as sickle cell anemia, or specialized situations such as the role of ABO incompatibility in blood transfusion. Now, however, with the availability of a “reference sequence” of the human genome, with emerging appreciation of the extent of genome variation among different individuals, and with a growing understanding of the role of common, not just rare, variation in disease, we are poised to begin to exploit the impact of that variation on human health on a broad scale, in the context of genomic and personalized medicine. Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 4
Variation in the human genome has long been the cornerstone of the field of human genetics (see Box 1.1), and its study led to the establishment of the medical specialty of medical genetics (Nussbaum et al., 2007). A crucial set of connections joining Mendel’s principles of heredity, Garrod’s concept of chemical individuality, the practice of medicine and the sequence of the genome came with Pauling’s discovery of the molecular basis of sickle cell anemia and the direct correspondence between an individual’s genetic make-up and the type of hemoglobins present in that individual’s red cells (Pauling et al., 1949). The general nature and frequency of gene variants in the human genome became apparent with the classic work in the 1960s on the incidence of polymorphic protein variants in populations of healthy individuals (Harris and Hopkinson, 1972; Lewontin, 1967; reviewed in Harris, 1980). Calculations based on those protein polymorphism data – now extended in a robust and comprehensive manner with the analysis of variation on a genome scale – lead to the inescapable conclusion that virtually every individual should be found to have his or her own unique constitution of gene products, the implications Copyright © 2009, Elsevier Inc. All rights reserved.
Introduction
BOX 1.1
■
5
Genetics and Genomics
Throughout this and the many other chapters in this volume, the terms “genetics” and “genomics” are used repeatedly, both as nouns and in their adjectival forms. While these terms seem similar, they in fact describe quite distinct (though frequently overlapping) approaches in biology and in medicine. Having said that, there are inconsistencies in the way the terms are used, even by those who work in the field. To some, genetics is a subfield of genomics; to others, genomics is a subfield of genetics. Arguably, depending on the perspective one has in mind, both may be right! Here, we provide operational definitions to distinguish the various terms and the subfields of medicine to which they contribute. The field of genetics is the scientific study of heredity and of the genes that provide the physical, biological and conceptual bases for heredity and inheritance. To say that something – a trait, a disease, a code or information – is “genetic” refers to its basis in genes and in DNA. Heredity refers to the familial phenomenon whereby traits (including clinical traits) are transmitted from generation to generation, due to the transmission of genes from parent to child. A disease that is said to be inherited or hereditary is certainly genetic; however, not all genetic diseases are hereditary (witness cancer, which is always a genetic disease, but is only occasionally an inherited disease). Genomics is the scientific study of a genome or genomes. A genome is the complete DNA sequence, containing the entire genetic information of a gamete, an individual, a population or a species. As such, it is a subfield of genetics when describing an approach taken to study genes. The word “genome” originated as an analogy with the earlier term “chromosome,” referring to the physical entities (visible under the microscope) that carry genes from one cell to its daughter cells or from one generation to the next. “Genomics” gave birth to a series of other “-omics” that refer to the comprehensive study of the full complement of genome products – for example, proteins (hence, proteomics), transcripts (transcriptomics) or metabolites (metabolomics). The essential feature of the “omes” is that they refer to the complete collection of genes or their derivative proteins, transcripts or metabolites, not just to the study of individual entities. While formally the field of genomics refers to the study of genomes (and hence, DNA) only, it sometimes takes on the broader meaning of referring to any large-scale approach; the less specific term “genome sciences” is also sometimes used to refer to all of the “-omics” to connote global and comprehensive approaches to the study of biology and medicine. By analogy with genetics and genomics, epigenetics and epigenomics refer to the study of factors that affect gene (or, more globally, genome) function, but without an accompanying change in genes or the genome. As presented later in this chapter and others, some typical epigenetic factors involve changes in DNA methylation or modifications to chromatin that change genome structure and hence influence gene expression even in the absence of changes in the DNA sequence. The epigenome is the comprehensive set of epigenetic changes in a given individual, tissue, tumor or population.
of which provide a conceptual foundation for what today we call “personalized medicine” as a new-age rediscovery of Garrod’s “chemical individuality.” Thus, with the availability of the human genome sequence (International Human Genome Sequencing Consortium, 2001, 2004; Venter et al., 2001) and determination of
It is the paired combination of the genome and the epigenome that best characterize and determine one’s phenotype. Medical genetics is the application of genetics to medicine with a particular emphasis on inherited disease. Medical genetics is a broad and varied field, encompassing many different subfields, including clinical genetics, biochemical genetics, cytogenetics, molecular genetics, the genetics of common diseases and genetic counseling. Medical genetics is one of 24 medical specialties recognized by The American Board of Medical Specialties, the preeminent medical organization overseeing physician certification in the United States. As of 2007, there were approximately 2300 board-certified medical geneticists in the United States. Genetic medicine is a term sometimes used to refer to the application of genetic principles to the practice of medicine and thus overlaps medical genetics. However, genetic medicine is somewhat broader, as it is not limited to the specialty of Medical Genetics but is relevant to health professionals in many, if not all, specialties and subspecialties. Both medical genetics and genetic medicine approach clinical care largely through consideration of individual genes and their effects on patients and their families. By contrast, genomic medicine refers to the use of large-scale genomic information and to consideration of the full extent of an individual’s genome, proteome, transcriptome, metabolome and/or epigenome in the practice of medicine and medical decision-making. The principles and approaches of genomic medicine are relevant well beyond the traditional purview of medical genetics and include, as examples, gene expression profiling to characterize tumors or to define prognosis in cancer, genotyping variants in the set of genes involved in drug metabolism or action to determine an individual’s correct therapeutic dosage, scanning the entire genome for millions of variants that influence one’s susceptibility to disease, or analyzing multiple protein biomarkers to monitor therapy and to provide predictive information in presymptomatic individuals. Finally, personalized medicine refers to a rapidly advancing field of health care that is informed by each person’s unique clinical, genetic, genomic and environmental information. The goals of personalized medicine are to take advantage of a molecular understanding of disease to optimize preventive health care strategies and drug therapies while people are still well or at the earliest stages of disease. Because these factors are different for every person, the nature of disease, its onset, its course, and how it might respond to drug or other interventions are as individual as the people who have them. In order for personalized medicine to be used by health care providers and their patients, these findings must be translated into precision diagnostic tests and targeted therapies. Since the overarching goal is to optimize medical care and outcomes for each individual, treatments, medication types and dosages, and/or prevention strategies may differ from person to person – resulting in unprecedented customization of patient care. The principles underlying genomic and personalized medicine and their applications to the practice of clinical medicine are presented throughout the chapters that comprise this volume.
the extent of human genome variation, both within and among populations (International HapMap Consortium, 2003, 2007) and within individual genomes (Levy et al., 2007; Wheeler et al., 2008), awareness of widespread human variation can begin to be applied generally to the exploration of common human disease.
6
CHAPTER 1
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
In this chapter, the organization, variation and expression of the human genome is presented as a foundation for the many chapters to follow on human genomics, on genome technology and informatics, on approaches in translational genomics and, finally, on the principles of genomic and personalized medicine as applied to specific diseases.
THE HUMAN GENOME The typical human genome consists of approximately 3 billion (3 109) bp of DNA, divided among the 24 types of nuclear chromosomes (22 autosomes, plus the sex chromosomes, X and Y) and the much smaller mitochondrial chromosome (Tables 1.1, 1.2). The genome can be represented and evaluated in different ways, with different levels of resolution and degrees of sensitivity, depending on the clinical need (Figure 1.1). Individual chromosomes can best be studied at metaphase in dividing cells, and karyotyping of patient chromosomes has been a valuable and routine clinical laboratory procedure for decades (Trask, 2002); various staining or hybridizationbased analytical techniques have the ability to detect chromosome abnormalities ranging from an extra or missing whole chromosome (aneuploidy), to translocations or rearrangements involving just a portion of a chromosome(s), to deletions or duplications involving as little as perhaps a megabase (106 bp; Mb) of DNA. More recent technologies involving overlapping sets (called “tiling paths”) of isolated segments of the genome arrayed on microscope slides have provided vastly improved resolution and precision capable of evaluating in a rapid and comprehensive way the proper dosage (and in some cases the organization) of the corresponding DNA segments within an individual’s genome (Figure 1.1b) (see Chapter 9). The ultimate resolution, of course, comes from direct sequence analysis, and a number of new technologies have reduced the cost and improved the throughput of sequencing individual genomes, facilitating comparisons with the reference human genome sequence and enabling medical resequencing of patient samples (see later section in this chapter) to search for novel variants or mutations that might be of clinical importance (Bentley, 2006) (Figure 1.1c, d) (see Chapter 7). Genes in the Human Genome While the human genome contains a currently estimated 20,000–25,000 genes (Clamp et al., 2007; International Human Genome Sequencing Consortium, 2004), the coding segments of those genes comprise less than 2% of the genome; as represented in Figure 1.1c, most of the genome, therefore, consists of DNA that lies between genes, far from genes or in vast areas spanning several Mb that appear to contain no genes at all (“gene deserts”). A caveat for this statement is that the process of gene identification and genome annotation remains very much a work-in-progress; despite the apparent robustness of recent estimates (Clamp et al., 2007), it is conceivable that
TABLE 1.1
Characteristics of the human genomea
Length of the human genome (basepairs)
3,253,037,807
Number of known protein-coding genes
21,541
b
Average gene density (genes/Mb)
6.6
Number of non-coding RNA genes
4421
Number of SNPsb
13,022,900
a b
From Ensembl v. 48 (accessed February 2008) Mb megabasepairs; SNP single nucleotide polymorphism
TABLE 1.2 Chromosome
Variation among human chromosomesa Mb
Genes/ Mb
miRNA genes
1
247.25
2153
8.7
68
2
242.95
1315
5.4
60
3
199.50
1105
5.5
57
4
191.27
786
4.1
42
5
180.86
894
4.9
46
6
170.90
1109
6.5
36
7
158.82
1008
6.3
43
8
146.27
743
5.1
38
9
140.27
904
6.4
40
10
135.37
819
6.1
35
11
134.45
1368
10.2
37
12
132.35
1069
8.1
43
13
114.14
356
3.1
23
14
106.37
662
6.2
62
15
100.34
634
6.3
21
16
88.83
902
10.2
20
17
78.77
1217
15.5
40
18
76.12
289
3.8
15
19
63.81
1427
22.4
82
20
62.44
603
9.7
28
21
46.94
283
6.0
10
22
49.69
508
10.2
18
X
154.91
874
5.6
97
Y
57.77
80
1.4
3
22
–
–
Mitochondrial a
Proteincoding genes
From Ensembl v. 48
0.016
The Human Genome
(a)
(d)
1
2
3
6
7
8
13
14
15
19
20
4
9
21
22
5
10
11
12
16
17
18
X
Y
(b)
0Mb
1Mb
2Mb
3Mb
■
7
>chr1:2040588,2043588 actgcaacctccacctcctgggttcaagtgattctgctgcctcagcctcctgagtagctg ggattacaggtgcccaccaccatgcccaactattttttgtatttttagtagaggcagggt ttcaccatattgaccaggctggtatcgaattcctggcctcaagtgatctgtctgccttgg cctcccaaagtgctggg[t/a]ttacaggcatgagccactgtgcctggcctaattattct tctttccttattgttagtttgtgctattattttatcagtctttgtgctgttattatcatg cctgtaaattctacgtgtatttcagacccacaaaccaagtgttgtcttagacagtggtcc ttcagatttacccccaggttacccttctagtcttcctgcaggacggcgcttacatggaga ccagcttccttctgcctgaagtagtccctttagtattcctttcagcacagacttgtaatt aattctttttatttcttttcttttcttttttttttttttgagatggatttttgctcttgt tgcccaggctggagtgcagtggtgtgattttggctcactgcagcctccacctcccaggtt caagcgattctcctggctcagcctcctgaggagctaggattgcaggtgtgcgccaccacg cccagttgttttttgtttgtgtgggaaatgtctttggcattctttctggagggtgttctc cactctgtgtggagttctaggcaggtagggggtttcccccaacaggtttttgtgttggct tggagtgtt[t/g]gtcatttctgtggtgagggcgccttccagcctcactgccacccctg gaaggcaacatctcttttctctgactcctgttaaaagtgttttcatcacaacagcagcct tgtgaaggacagaggaatcgagaatttctcctaattgagattggtagagcttcttgaatc agggacatgatagcttttgtctcttttggaaaatatcagcccttgacttttcgttttttt ttttttttttttttttttttttgagtctcgctcttgttgcccaggctggagtgcaatggc gcgatctcgactcactgcaatctccacctccccggttcaagtgattctcctgcctcagcg tcccgagtagctgggattacaggcacttgccaccatgaccggctaattttttttgcattt ataggagagacagggtttcaccatgttgaccaggctggtctggaactcctgatcatacat ccaccttggcctcccaaagtgctgggattacaggtgtgagccaccgtgcccggccagccc ttggcttttcaaatagcatcctgttctctctcccctgggacccccacacttcacacctgt gtgtctaatgtgctcttttttctgggtttcttctgcgttggttttttcccgctttgtgct tcaatgtggatttttttctactgttatctcttatttcacccaatctactcttaaatctac cctttaaattattaatttcagtcacttcattttttacttttagaatttccatttgattct ttttttttttttttttgcccaggatggcaatggcacgctctcggctcactgcaacctccg cctcccaggttcaagcaatattcctgccccagcctcccaagcagctgggattacagggtc acactaccacgccccactaatttttatgtttttattagagacggggttttgccatgttgg ccaggctggtctcgaactcctgaccttgggtgatccg[c/t]ttgcctcagcctcccaaa gtgttgggattacaggcgtgagccactgcgcctggcatcgtagttctctcttctggggtg ggaatgtctattctgtgtccttctcacgtgcaaaatactgtcattacatcccaatggccc cagaacccttaactcctcccagtgtggcgggggcagtcttgtctgaacaaggcatggggg agcctggaggcccattcctcctgaggccaagt[t/a]cctccctggctgtgggccagcat aagcgaacaaggcgtgtacttccggaatgctatggactgagtgtgtgtctccccagaatc catatgttgaagccctaaccctccagtgtgatggtgtttggagacgaagcctttgacagg tagttagagtcatggcggtagttagttagagtcatggcggtagttagttagggtcacggt ggtagttaggatcatggtggtacttaaggtcatggcagtagttagggttatatcagtagt tagggctatggctgtagttagggtgatggtggtagttaaggtcacagcagtaattagggt catggtggtggttagggtcacagtggtagttagggtcacggtggtggttagggtcgtggt ggtggttagggtcacggtggtggttagggtcacggtggtagttagggtcacggcggtact tagggtcacggcggtggttagggtcacggcggtggttagggtcacggtggtggttagggt cacggcggtggttagggtcacggtggtggttagggtcgtggtagttaggttcatggtggt ggttagggtcgtggtggttagggtcacggtggtggttagggtcacggtggtagttagggt cacggctgtagttagcgtcatggtggtggttagggtcacggcggtggttagggtcacggt ggtggttagggtcacggcggtggttagggtcacggtggtggttagggtcgtggtagttag gttcatggtggtggttagggtcgtggtggttagggtcacggtggtagttagggtcgtggt ggttagggtcatggtggtggttagggtcacggtggtggttagggtcgtggtggttagggt cgtggtggttagggtcgtggtggttagggttgtggtggttagggtggtggtggttagggt cgtggcggtggttagggtcgtggcggtggttagggttgtggtggttagggtcacggtggt ggttagggtcacggtgg…
4Mb
5Mb
6Mb
Chromosome 6
(c)
p25.2
Figure 1.1 Four views of the human genome. (a) Karyotype of a normal male donor, HuRef, whose genome was the first individual diploid genome to be sequenced (Levy et al., 2007). The 24 types of human chromosome are shown after conventional G-banding – 22 pairs of autosomes and the two sex chromosomes, X and Y. (b) An array of genomic segments, showing 244,000 genomic elements hybridized to DNA from HuRef. (c) Schematic representation of the content of 6 Mb from the short arm of chromosome 6, including the location of various genes and other features from the HuRef genome. (d) DNA sequence from the genome of James D. Watson, showing 3000 bp from chromosome 1. Watson’s sequence is heterozygous at four positions, three (in yellow) that are known polymorphisms in various populations and one (in red) that is a novel variant. Figures in (a) and (b) were provided courtesy of Steve Scherer, Hospital for Sick Children, Toronto, Canada. (c) is part of a large poster representing the complete diploid HuRef genome (Levy et al., 2007).
there are some genes, including clinically relevant genes, that are currently undetected or that display characteristics that we do not currently recognize as being associated with genes. A maximum of 5% of the genome consists of DNA that has been quite
well conserved through evolution, one indication of an important function. These and other considerations have led to the estimate that at most 20% of the genome is of functional importance (Pheasant and Mattick, 2007). Nonetheless, the statement
8
CHAPTER 1
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
that the vast majority of the genome consists of spans of DNA that are non-genic, of no obvious function, and of uncertain clinical relevance remains true. In addition to being relatively sparse in the genome, genes are distributed quite non-randomly along the different human chromosomes. Some chromosomes are relatively gene-rich, while others are quite gene-poor, ranging from a high of 22 genes/Mb to a low of 3 genes/Mb (excluding the Y chromosome and the mitochrondrial chromosome) (Table 1.2). And even within a chromosome, genes tend to cluster in certain regions or in particular bands, a point of clear clinical significance when evaluating genome integrity, dosage or arrangement in different patient samples. Coding and Non-Coding Genes There are a number of different types of gene in the human genome. Most genes are protein-coding and are transcribed into messenger RNAs (mRNAs) that are ultimately translated into their respective proteins; their products comprise the list of enzymes, structural proteins, receptors and regulatory proteins that are found in various human tissues and cell types. However, there are additional genes whose functional product appears to be the RNA itself. These so-called non-coding RNAs (ncRNAs) have a range of functions in the cell, and some do not as yet have any identified function. But the genes whose transcripts make up the collection of ncRNAs represent about a sixth of all identified human genes (Table 1.1). Some of the types of ncRNA play largely generic roles in cellular infrastructure, including transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs) involved in translation of mRNAs on ribosomes, spliceosomal RNAs involved in control of RNA splicing, and small nucleolar RNAs (snoRNAs) involved in modifying rRNAs (Griffiths-Jones, 2007; Mattick and Makunini, 2006). Other ncRNAs play roles in gene regulation, for example in epigenetic gene silencing (Ogawa and Lee, 2002). A class of small RNAs of growing importance are the microRNAs, ncRNAs of only 22 bases in length that suppress translation of target genes by binding to their respective mRNAs and thus regulate protein production from the target transcript(s) (Filipowicz et al., 2008). Some 255 microRNA genes were identified in the human genome initially (Lim et al., 2003), although the total number of such genes is now thought to be closer to 1000 (Bentwich, 2005; Griffiths-Jones, 2007) (Table 1.2). Some are evolutionarily conserved, while others appear to be of quite recent origin during primate evolution, thus underscoring the difficulty of determining the precise number and identity of human genes (Clamp et al., 2007). MicroRNAs have been shown to downregulate hundreds of mRNAs each, with different combinations of target RNAs in different tissues (Lim et al., 2005); combined, the microRNAs are thus predicted to control the activity of as many as 30% of all protein-coding genes in the genome (Filipowicz et al., 2008). While this is a fast-moving area of genome biology, several microRNAs have already been implicated in various human diseases, including cancer,
developmental disorders and heart disease (Chang and Mendell, 2007; van Rooij et al., 2008). Genome Composition and Landscape As observed earlier, the distribution of genes in the genome is non-random, both within and between chromosomes. This in part is a reflection of the distribution of different types of DNA sequence, as the genome is partitioned into domains spanning hundreds of kilobasepairs to megabases, reflecting large-scale variation in the GC content of DNA. These so-called “isochores” have been known for decades and, at a very gross level, mimic the pattern of light- and dark-staining bands that one observes in metaphase chromosomes (e.g., Figure 1.1a, c) (Eyre-Walker and Hurst, 2001). While the driving force behind the evolution of isochores is not clear, they influence the GC content of genes contained within them (and, by virtue of the genetic code, therefore, the amino acid composition of the encoded proteins), the patterns of mutation and polymorphism detected, and the nature of various families of repeated DNA that reside there. Further – and most strikingly – different isochore domains contain clusters of genes that are highly or weakly expressed in a coordinated manner in different tissues (Caron et al., 2001; Gierman et al., 2007; Hurst et al., 2004). Thus, isochores reflect both the functional as well as structural organization of the genome. (See later section on “Expression of the Human Genome” for further discussion.) Repetitive DNA Overall, only about half of the total linear length of the genome consists of so-called single-copy or unique DNA, whose sequence is represented only once or at most a few times (International Human Genome Sequencing Consortium, 2001, 2004; Venter et al., 2001). The rest of the genome consists of several classes of repetitive DNA and includes DNA whose sequence is repeated, either perfectly or with some variation, hundreds to millions of times in the genome. Several different categories of repetitive DNA are recognized. Clustered repeated sequences constitute an estimated 10–15% of the genome and consist of arrays of various short repeats organized tandemly in a head-to-tail fashion. Such arrays can stretch several Mb or more in length and constitute up to several percent of the DNA content of individual human chromosomes; a notable outlier in this respect is the male-specific Y chromosome, of which more than half consists of such repeated DNA families (Skaletsky et al., 2003). Other tandem repeat families are based on somewhat longer basic repeats. For example, the alpha-satellite family of DNA is composed of tandem arrays of different copies of an 171 bp unit, found at the centromere of each human chromosome, which is critical for proper segregation of chromosomes during cell division (Rudd and Willard, 2004; Schueler and Sullivan, 2006). Another highly significant family of repeats is found at the very ends of chromosomes, the telomeres. While the repeats at the functional telomeres consist of relatively short stretches of perfect (TTAGGG)n repeats, different subtelomeric regions (just proximal to the telomere repeats) share patterns of homology with other subtelomeres around the genome that
Variation in the Human Genome
create clinically relevant hotspots of interchromosomal recombination (Linardopoulou et al., 2005; Riethman et al., 2005). Other major types of repetitive DNA in the genome consist of related sequences that are dispersed throughout the genome rather than localized. Among the best-studied dispersed repetitive elements are short, interspersed nuclear elements (SINEs). The most prominent family of these contains repeats that are about 300 bp in length and are recognizably related to each other although not identical in DNA sequence. In total, members of this family make up at least 10% of human DNA, although they make up a much higher percentage of the DNA in some isochores. A second major dispersed, repetitive DNA family is called the LINE (where the L stands for long) family, whose members range in size up to 6 kp in length and account for about 20% of the genome. Families of repeats dispersed throughout the genome are clearly of medical importance. Both SINE and LINE sequences have been implicated as the cause of mutations in genetic disease. At least a few copies of these families generate copies of themselves that can integrate elsewhere in the genome, occasionally causing insertional inactivation of a medically important gene. The frequency of such events causing genetic disease in humans is largely unknown, but they have been suggested to account for as many as 1 in 500 mutations (Deininger et al., 2003; Kazazian and Moran, 1998). In addition, aberrant recombination events between different LINE or SINE repeats can also be a cause of mutation in some genetic diseases. Segmental Duplications An important subclass of repetitive DNA, distinct from the large families just mentioned, includes blocks of different sequences (hence, not defining a particular family of sequences) that are present in multiple copies, often with extraordinarily high sequence conservation, in many different locations around the genome. Duplications involving substantial segments of a chromosome, called segmental duplications, account for at least 5% of the genome (Bailey and Eichler, 2006). When the duplicated regions contain genes, genomic rearrangements can result in the deletion of the region (and the genes) between the copies and thus give rise to disease (Conrad and Antonarakis, 2007). In addition, rearrangements between duplicated segments are a source of significant variation between individuals in the number of copies of these DNA sequences (Sharp et al., 2005), as will be discussed in the next section.
With completion of the reference human genome sequence, attention turned to the discovery and cataloging of variation in that sequence among different individuals (including both healthy individuals and those with various diseases) and among different populations. It has been estimated that there are some 10–15 million common sequence variants that are of sufficient frequency (minor allele frequency 5%) in one or more
9
populations to be considered polymorphic in our species. In addition, there are countless very rare variants, many of which probably exist in only a single or a few individuals. In fact, given the number of individuals in our species, essentially each and every base pair in the human genome is expected to vary in someone somewhere around the globe. It is for this reason that the original genome sequence is considered a “reference” sequence, derived as a consensus of the limited number of individual genomes whose sequencing was part of the Human Genome Project, but actually identical to no individual’s genome. Types of Variation Early estimates were that any two randomly selected individuals have sequences that are 99.9% identical or, put another way, that an individual genome would be heterozygous at approximately 3–5 million positions, with different bases (i.e., a T or a G) at the maternally and paternally inherited copies of that particular sequence position. The majority of these differences involve simply a single unit in the DNA code and are referred to as single nucleotide polymorphisms (SNPs) (Table 1.1) (see Chapter 7). The remaining variation consists of insertions or deletions (in/dels) of (usually) short sequence stretches, variation in the number of copies of repeated elements or inversions in the order of sequences at a particular locus in the genome (Figure 1.2). The total amount of in/del variation is more than originally anticipated and approaches 0.5%, not 0.1%, between any two randomly selected individuals (Levy et al., 2007). Any and all of these types of variation can influence disease and thus must be accounted for in any attempt to understand the contribution of genetics to human health (Table 1.3).
A
B
C
D
Reference A
B
C
C
D
Segmental duplication – Biallelic CNV (C)2 A
B
C
C
C
D
Multiallelic copy number variant (C)0-n A
B
C
D
D
D
D
C
D
C
D
C
D
Complex CNV (D)4(CD)3 A
VARIATION IN THE HUMAN GENOME
■
C
B
D
Inversion (CB) Chromosome
Figure 1.2 Schematic representation of different types of structural polymorphism in the human genome, leading to deletions, duplications, inversions and CNV changes relative to the reference arrangement. From Estivill and Armengol (2007), with permission.
10
CHAPTER 1
TABLE 1.3
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
Common variation in the human genome
Type of variation
Size range (approx.)a
Effect(s) in biology and medicine
Single nucleotide polymorphisms (SNPs)
1 bp
Non-synonymous → functional change in encoded protein? Others → potential regulatory variants? Most → no effect? (“neutral”)
Copy number variants (CNVs)
10 kb to 1 Mb
Insertion/deletion polymorphisms (in/dels)
1 bp to 1 Mb
Inversions
Few bp to 100 kb
Segmental duplications
10 kb to 1 Mb
a
Gene dosage variation → functional consequences? Most → no effect or uncertain effect
In coding sequence: frameshift mutation? → functional change Most → uncertain effects ? break in gene sequence ? long-range effect on gene expression ? indirect effects on reproductive fitness Most → no effect? (“neutral”)
Hotspots for recombination → polymorphism (CNVs)
Abbreviations: bp basepair; kb kilobasepair; Mb megabasepair
While the overall estimate of SNP heterozygosity is approximately 1 in 1500 bp, there is much more variation in non-coding sequences than in the coding segments of genes, reflecting strong selective pressure during evolution against certain types of change in gene sequences. The combination of particular alleles along chromosomes is also non-random, with particular combinations (haplotypes) being more prevalent over short distances, due to the relative inefficiency of meiotic recombination to separate alleles at sites that are physically close together (International HapMap Consortium, 2007; Nussbaum et al., 2007). The resulting patterns of linkage disequilibrium are relevant for designing strategies to examine genetic variation genome-wide, both as a practical matter (i.e., reducing the number of SNPs that need to be tested to reveal the underlying patterns of variation) and for evaluating the potential functional importance of any particular SNP allele (see Chapters 2, 8 and 27). Copy Number Variation Over the past few years, a number of important studies have identified a previously unanticipated prevalence of structural variants in the genome, which collectively account for more variation in genome sequence than do SNPs (e.g., Levy et al., 2007; Redon et al., 2006; Sebat et al., 2004; Tuzun et al., 2005). The most common type of structural variation involves changes in the local copy number of sequences (including genes) in the genome, and these are generally referred to as copy number variants (CNVs) (Figure 1.2) (see Chapter 9). A number of different technology platforms are now being used to detect CNVs, including arrays and direct genome sequencing (Korbel et al., 2007; Levy et al., 2007; Wong et al., 2007). As many such CNVs encompass genes (including microRNA genes; Wong et al., 2007) and as a significant number of
novel CNVs are uncovered with every new population studied, a dedicated effort is underway to cataloged CNVs in the human genome worldwide and to associate these with clinical phenotype (Feuk et al., 2006; Scherer et al., 2007; Sharp et al., 2006). While most variation of this type is inherited, some CNVs occur de novo or even in somatic cells; in these cases, an individual will have different repeat lengths than do either of his or her parents. Array-based methods (see Chapter 9) have rapidly gained acceptance for evaluating the association of both inherited and de novo CNVs with mental retardation and other developmental disorders (de Vries et al., 2005; Friedman et al., 2006; Lee et al., 2007;Weiss et al., 2008). It is of considerable ongoing interest to evaluate the role of CNVs and other structural variants including deletions (Conrad et al., 2006) and inversions (Korbel et al., 2007; Stefansson et al., 2005; Tuzun et al., 2005) in the etiology of more common, complex diseases or traits of adulthood, including neurological and psychiatric conditions as well as pharmacogenetic traits (Beckmann et al., 2007; Buckland, 2003). Variation in a Single Genome The most extensive current inventory of the amount and type of variation to be expected in any given genome comes from the direct analysis of the diploid genome sequence of a single male individual, HuRef (Levy et al., 2007). Over 4 million variants were described, spanning some 12.3 Mb of DNA. About 20 Mb of “new” sequence was determined that was not previously available as part of the human reference sequence, reflecting in part the still unfinished nature of the human genome sequence and in part the particular patterns of inserted or deleted sequences that distinguish different genomes. Several hundred thousand in/dels were also found in this single genome. In addition, several hundred CNVs were detected, which overlapped at
Expression of the Human Genome
least 95 well-annotated genes. While most of these variants are identical to those found in other individuals in the population, others are likely to be what are termed “private” mutations, specific to HuRef and his family. In the HuRef genome, at least 850 genes known to be involved in inherited disease contained at least one heterozygous variant, and over 300 of them contained at least one nonsynonymous SNP (i.e., a SNP that, by virtue of the genetic code, is predicted to change the encoded amino acid). Of course, additional genes may also impact disease, and, overall, more than 4000 genes in the HuRef genome contained one or more non-synonymous SNP. Thus, at least 17% and perhaps as many as 44% of the genes in the HuRef genome were heterozygous and could encode proteins that differ in their amino acid sequence and/or are produced in different amounts (Levy et al., 2007). These estimates underscore the impact of gene and genome variation on human biology and on medicine. They also provide remarkable validation of the original estimates of Harris and Lewontin decades ago of the proportion of genes that are heterozygous in any given individual (Harris, 1980; Lewontin, 1967). Table 1.3 and Figure 1.2 capture the general types of and characteristics of the most common variation in the human genome and in human genes. However, it is clear that we are still in a mode of discovery, as relatively few genomes or populations have been assessed to date; no doubt millions of additional SNPs remain to be uncovered, as well as many additional in/ dels, inversions and CNVs, a portion of which will be expected to involve genes and other sequences of direct relevance to medicine. The issue of “what is normal?” – an essential concept in clinical medicine – remains very much an open question when it comes to the human genome (Shianna and Willard, 2006). Variation in Populations Most of the heterozygosity in the human genome is believed to be due to variants with a minor allele frequency of at least 1%. Taking advantage of major technological developments that have greatly increased the throughput of genotyping on a genome-wide scale, several large-scale projects have validated these estimates by gathering genotypic information on millions of SNPs worldwide (Hinds et al., 2005; International HapMap Consortium, 2003, 2007). Most of the studies to date, however, have been restricted to a small number of populations of Northern European, African and Asian origin used for SNP detection. From these and a large number of earlier studies that examined more populations but for many fewer variants, it has been concluded that some 85–90% of the variation found in our species is shared among different population groups; a relative minority of variants, therefore, are specific to or highly enriched/depleted in genomes from a particular population. It is possible to use population-specific variants to obtain information on the geographic origin of a genome or of particular segments within a genome. Given the many millions of SNPs now available, there are at least hundreds of thousands of
■
11
SNPs that are informative for such studies (so-called “ancestry informative markers,” AIMs) (Kittles and Weiss, 2003; Paschou et al., 2007; Tian et al., 2008). This had led to two related, but distinct applications of such markers. First is the use of admixture mapping, tracing the location of particular SNPs associated with disease in populations of patients whose genomes are a mixture from at least two original populations, for example, African-Americans or Latinos (Price et al., 2007; Smith et al., 2004). Such an approach has already been used to map genes associated with several phenotypes whose frequency differs markedly between different population groups, including prostate cancer (Freedman et al., 2006), hypertension (Deo et al, 2007), skin pigmentation (McEvoy et al., 2006) and white blood cell count (Nalls et al., 2008). The second use of AIMs is for ancestry testing unrelated to disease studies (Kittles and Weiss, 2003; Shriver and Kittles, 2004). While the motivations behind such testing and the potential uses (and, some fear, abuses) of biogeographic information are varied, the commercial availability and interpretation of genetic ancestry testing is controversial (Bolnick et al., 2007). Nonetheless, the availability of such information as an intentional or unwitting by-product of wide-scale genome analysis is inevitable, and both consumers/patients and health professionals need to be aware of this as genetic variation is explored in the context of individual genomes (see Chapter 33).
EXPRESSION OF THE HUMAN GENOME A key question in exploring the origins, structure and function of the human genome is to understand how proper expression of our 20,000–25,000 genes is determined, how it can be influenced by either genetic variation or by environmental exposures or inputs, and by what mechanisms such alterations in gene expression can lead to pathology evident in the practice of clinical medicine. The control of gene activity – in development, in different tissues, during the cell cycle, and during the lifetime of an individual both in sickness and in health – is determined by a complex interplay of genetic and epigenetic features. By “genetic” features, we here refer to those found in the genome sequence (see Box 1.1), which plays a role, of course, in determining the identity of each gene, its particular form (alleles), its level of expression (regulatory elements such as promoters, enhances, splice sites, etc.), and its particular genomic landscape (domains, isochores). By “epigenetic” features, here we mean packaging of the DNA into chromatin, in which it is complexed with a variety of histones as well as innumerable non-histone proteins that influence the accessibility and activity of genes and other genomic sequences. The structure of chromatin – unlike the genome sequence itself – is highly dynamic and underlies the control of gene expression that shapes in a profound way both cellular and organismal function (Felsenfeld and Groudine, 2003).
12
CHAPTER 1
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
Genomic and Epigenomic Aspects of Gene Expression Identifying the genomic sequences and features that direct spatial and temporal aspects of gene expression remains a formidable challenge in genome annotation. While several decades of work in molecular biology have defined critical regulatory elements such as promoters and enhancers for many individual genes, it is a tall order to perform such studies on a genomewide scale. At the genetic level, such studies have depended on identifying highly conserved elements, whose demonstrated sequence conservation over, in some cases, hundreds of millions of years provides prima facie evidence of their importance (Bejerano et al., 2004; Margulies et al., 2003) (see Chapter 10). Several such ultraconserved elements have been validated in vitro as gene enhancers, providing some confidence that others identified in this way will also play a role in regulating gene expression (Pennacchio et al., 2006). However, by definition, such approaches will overlook regulatory elements that are not well conserved at the sequence level and/or that are newly evolved on the human lineage. A complementary epigenetic approach has been to explore the characteristics of chromatin that are associated with active or repressed genes as a step towards identifying the transcriptional regulatory code for the human genome (Barrera and Ren, 2006; Bernstein et al., 2007). Such studies, largely employing the method of chromatin immunoprecipitation followed by array or sequence analysis (Hawkins and Ren, 2006), have uncovered predictive chromatin “signatures” for promoters and enhancers in the human genome (Heintzman et al., 2007; Kim et al., 2005). These analyses are part of a broad effort to explore epigenetic patterns in chromatin genome-wide to better understand control of gene expression in different tissues or disease states (ENCODE Project Consortium, 2007; Brena et al., 2006). Increasing evidence points to a role for epigenetic changes in human disease in response to environmental influences (Feinberg, 2007). The dynamic nature of epigenetic regulation – changes in DNA methylation or in histone modification over time, between different tissues, or in various disease states – allows for what has been called ”phenotypic plasticity,” relevant both to the origins and potential treatment of disease (Feinberg, 2007) (see Chapters 5 and 11). That monozygotic twins (who share identical genomes, but are frequently discordant with respect to clinical phenotypes) show epigenetic differences in DNA methylation demonstrates the potential for epigenetic effects to extend or modify information contained in the genome and thus to underlie at least some phenotypic differences (Fraga et al., 2005).
Genetic Control of Gene Expression Levels It has been appreciated for decades that there is high variability in gene expression levels among individuals. A number of groups have begun to examine this variation as a complex quantitative trait, under genetic control (Cheung and Spielman, 2002). Several thousand loci exhibiting such variation have now been studied, and the factor(s) controlling the variation mapped around the genome (Cheung et al., 2005; Morley et al., 2004; Stranger et al., 2007a, b). A significant proportion of such effects map to the genes themselves, a result consistent with local sequence variation influencing the expression of such genes; many such local regulatory variants map to the promoter region of the gene or to the 3 untranslated region (Cheung et al., 2005). Most of the mapped variation could be accounted for by SNP variation, but nearly 20% was due to CNVs (Stranger et al., 2007a, b), underscoring the functional importance of this type of genome variation. It is likely (though unproven in specific instances) that the discovered regulatory variants will correlate with the patterns of epigenetic modifications described earlier (Heintzman et al., 2007; Kim et al., 2005) and with the finding of widespread allelic imbalance around the genome (Serre et al., 2008). Other determinants of gene expression variation map not to the variable locus itself, but to another locus elsewhere in the genome. This implies a network of regulatory interactions among different gene products in the cell, a finding that lends itself to a systems biology approach to begin to better define such networks (see Chapter 6). Some of the differences in gene expression also vary among populations. Where examined, the phenotypes are attributable to common genetic variants that are more common in particular populations than in others (Spielman et al., 2007; Stranger et al., 2007a, b). These findings, therefore, may be relevant to complex genetic diseases whose prevalence differs among populations.
Allelic Imbalance Other than well-known examples of monoallelic gene expression where only one of the two copies of a gene in a diploid cell is expressed (such as genomic imprinting and X chromosome inactivation; Nussbaum et al., 2007), it is commonly assumed that the vast majority of autosomal genes in the genome are expressed
The Genome in Three-Dimensional Space In contrast to the impression one gets when viewing the genome as a linear string of sequence (e.g., Figure 1.1c,d), the genome adopts a highly ordered arrangement within the three-dimensional space of the nucleus (Figure 1.3) (Bolzer et al., 2005; Lanctot et al., 2007). This three-dimensional structure is highly predictive of the transcriptome map (Goetze et al., 2007) and
from both homologs at comparable levels. Recent evidence, however, using expressed SNPs to distinguish transcripts from the two alleles, has demonstrated widespread differential allelic expression for up to 20% of genes in the genome (Serre et al., 2008). This phenomenon, whose epigenetic or genetic basis is unknown, may be related to similar variability observed among X-linked genes in heterozygous females (Carrel and Willard, 2005). Further complexity is suggested by the finding of monoallelic gene expression, random with respect to parental origin, for approximately 5% of genes tested (Gimelbrant et al., 2007). This is an area of ongoing study and further implicates interactions between the genome and epigenetic control mechanisms as a determinant of gene expression in the human genome.
Genes, Genomes and Disease
(a)
■
13
(b)
1
2
6
13
19
3
7
14
20
8
4
9
10
15
5
11
16
21
22
12
17
18
X
Y
Figure 1.3 Organization of the human genome in metaphase and interphase. (a) Individual metaphase chromosomes from a normal male donor, HuRef. Chromosomes are colored differently after hybridization to chromosome-specific DNA. From Levy et al. (2007), with permission. (b) Arrangement of chromosomes in an interphase nucleus. Each chromosome maintains its own territory, with minimal mixing. From Bolzer et al. (2005), with permission.
reflects the megabase-sized domains evident from examining coordinated patterns of gene expression at the chromosome level (Gierman et al., 2007). The biophysical and/or genomic properties that facilitate or specify the orderly and dynamic packaging of each chromosome during each cell cycle without reducing the genome to a tangled mess within the nucleus remain unknown and are the subject of much investigation in genome biology. Overall, then, one should view the gene expression phenotype as the sum of several different (but interrelated) effects, including gene sequence, regulatory sequences and their epigenetic packaging, organization of the genome into domains and isochores, programmed interactions between different parts of the genome, and three-dimensional packaging in the nucleus (Fraser and Bickmore, 2007). All must coordinate in an efficient and hierarchical fashion, and disruption of any one – due either to genetic change or to disease-related processes – would be expected to alter the overall cellular phenotype.
GENES, GENOMES AND DISEASE In the context of genomic and personalized medicine, a key question is to what extent variation in the sequence and/or expression of one’s genome influences the likelihood of disease onset, determines or signals the natural history of disease, and/or provides clues relevant to the management of disease. Variation in one’s constitutional genome can have a number of different direct or indirect effects on gene expression (Table 1.3), thus contributing to the likelihood of disease. It is not, however, just the human genome whose variation is relevant to an individual’s state of health; there are thousands of microorganisms,
both symbiotic and pathogenic, whose genomes are also relevant to human phenotypes, and sequence determination of their genomes is providing new insights and approaches for the diagnosis, study and treatment of infectious disease (see Box 1.2). Contemporary frameworks for considering the impact of variation in the human genome on disease build on decades of success in establishing the role of individual, typically rare, mutations as causal determinants of now more than 2000 simple Mendelian diseases (Online Mendelian Inheritance in Man, 2008). As a result of that success, much of which paralleled the development of technologies in the early stages of the Human Genome Project (Peltonen and McKusick, 2001), attention has now turned to the genes presumed to underlie susceptibility to common complex diseases, which are the subject of many of the subsequent chapters in this book. There are two distinct, but non-exclusive models for thinking about human genetic variation and disease (Altshuler and Clark, 2005; Fearnhead et al., 2005; Florez et al., 2003; Pritchard, 2001). One – the “common allele, common disease” hypothesis – posits that variation common in the population accounts for the relatively higher or lower risk that some individuals (and their families) have for a particular condition. Under this model, the collection of 10–15 million common SNPs, CNVs and other variants in the genome underlies the range of susceptibility that one finds in the general population, modulated by the particular and often variable environmental inputs and factors that are present in that population and that may, in fact, shift or even obscure the relative impact of inherited factors. An alternative model – the rare variant hypothesis – argues that genetic susceptibility to disease is due to the accumulated risk conferred by multiple rare variants in an individual’s genome,
14
CHAPTER 1
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
BOX 1.2 The Genomes Within The human genome is not the only genome relevant to the practice of medicine. Both in states of health and disease, our own genome is vastly outnumbered by the genomes of a host of microorganisms, many living peacefully and continuously on various body surfaces, especially throughout the gastrointestinal tract, others wreaking havoc as adventitious viral, bacterial or fungal pathogens. The genomes of thousands of microorganisms have been determined and are being utilized to provide rapid diagnostic tests in clinical settings, to predict antibiotic or antifungal efficacy, to identify the source of airborne, water or soil contaminants, to monitor hospital or community environments, and to better understand the contribution of microbial ecosystems and various environmental exposures to diverse human phenotypes. The human colon contains more than 400 bacterial species comprising some 1013 to 1014 microorganisms. Each adult’s gut provides a unique environment – the microbiome – whose origins and impact on human disease are just being explored. The microbiotic gene set is significantly different from that of the human genome and thus has the capacity to alter the metabolic profile of different individuals or different populations, with clinically meaningful effects on drug metabolism, toxicity and efficacy (Gill et al., 2006; Li et al., 2008; Palmer et al., 2007). The applications to microbiomes of approaches in genomics (as well as proteomics and transcriptomics)
variants therefore not likely to be captured by study of the common variants identified by studies to date. These two hypotheses suggest different approaches that will likely be informative for delineating the genetic contribution to disease, both for designing research studies and for eventual clinical surveillance. It is worth emphasizing that these two hypotheses are not mutually exclusive and are each likely to be correct in some cases; indeed, there is evidence supporting each for different diseases. Genome-wide Association Studies The “common allele, common disease” hypothesis has been explored with notable success in a number of conditions, utilizing large cohorts of well-phenotyped patients and high throughput methods to genotype up to 500,000 or a million SNPs in the genome (Topol et al., 2007) (see Chapter 8). These genome-wide association studies report the statistical association of one or more variants in a narrow genomic region (which may or may not contain an annotated gene) with the presence or absence of the clinical condition. The reported SNPs define immediately accessible risk factors for that condition, at least in the population(s) under study, and can provide novel insights into the biology of the disease. It should be stressed, however, that in most instances, causality of the reported SNP(s) and the increased risk has not been proved; it may be that the actual causal variant is not the SNP itself, but is a currently undetected variant that lies in linkage disequilibrium with the SNP. In the most favorable cases, the associated SNP may be a non-synonymous variant, leading to a pathological amino acid change in the relevant gene (e.g., Thorleifsson et al., 2007), or
are revolutionizing clinical diagnostics, for example to identify unknown viral infections (Delwart, 2007; Long et al., 2004; Wang et al., 2003) or to diagnose antibiotic-resistance infections such as methicillin-resistant Staphylococcus aureus (MRSA) (Francois et al., 2007). The field of metagenomics explores this heterogeneous ecosystem by comprehensive sequence analysis of the collected genomes from biological specimens (such as stool, urine, sputum, water sources and air), followed by both taxonomic and bioinformatic analysis to deconvolute the many genomes contained in such specimens and to define the different organisms, their genes and genome variants. This approach is particularly informative for characterizing organisms that cannot be cultured in standard microbiology labs. A number of diseases have been associated with large-scale imbalances in the gut microbiome, including Crohn’s disease, ulcerative colitis, antibioticresistant diarrhea and obesity (Frank and Pace, 2008; Ley et al., 2006; Turnbaugh et al., 2006). Undoubtedly, the states of health and disease are determined in part by the balance of genomes both within us and external to us. The full complement of genomic information from both of these sources of genomes will provide insights into defining the states of health and disease and the basis for unsurpassed precision in both the prevention of disease and its treatment.
may be a variant in an RNA splice site, leading to a clinically meaningful change in the production of the gene’s transcript(s) (e.g., Heinzen et al., 2007). But in most instances, the functional impact of the associated SNP is obscure, notwithstanding very clear genetic evidence of a role of genome variation in susceptibility to the particular condition. As emphasized throughout this chapter, SNPs are but one of several types of genome variation that can influence gene expression and/or disease (Table 1.3). CNVs have also been associated with some disorders (reviewed in Estivill and Armengol, 2007), and it will require integrated genomic, genetic and functional studies to elucidate the precise basis for the role(s) of genome variation in different diseases. Medical Resequencing in Search of Rare Variants While genome-wide association studies certainly establish that common alleles do indeed underlie susceptibility to common disease in some instances, they do not allow one to conclude that this is always the case. Indeed, in most (but not all) cases to date, the common SNPs found to be associated with disease only explain a small fraction of the total genetic variation, implicating an as yet undiscovered (and presumably rarer) basis for most genetic variation underlying a given condition (Iles, 2008). In other words, while a positive statistical association is sufficient to conclude that a particular variant does indeed contribute to disease in the population under study, such a finding is insufficient to say anything declarative about the cause of or likelihood of disease in a particular case. This conclusion is, of course, highly relevant to the prospects of genomic and personalized medicine.
From Genome to Personalized Medicine
An alternative or complementary approach to genomewide genotyping of common variants is to resequence specific genes in a cohort of affected individuals, in an effort to uncover rare variants responsible for (or at least statistically associated with) the disease in question. To date, most efforts have focused on one or several genes that were believed to be strong candidates for the phenotype under study; the notable exception thus far is the whole-genome resequencing of the HuRef genome and correlation of novel variants detected in that study with his family and personal medical histories (Levy et al., 2007). Rare variants, including non-synonymous variants, in relevant candidate genes have been detected at a statistically significant higher frequency in the genomes of patients with colorectal adenomas (Fearnhead et al., 2004), with low plasma high-density lipoprotein cholesterol (HDL-C) (Cohen et al., 2004), and with triglyceride levels in the lowest quartile (Romeo et al., 2007). These successes point to a strategy of resequencing relevant genes in individuals at the extremes of the population distribution for measurable traits (Topol and Frazer, 2007), in which the only limiting parameters are the cost of sequencing and the quality of the phenotypic or quantitative data. In a proof-of-principle for this case-control approach, a recent study sequenced coding exons and splice junctions of 58 genes in nearly 400 obese and lean individuals, at the 95th or 10th percentile of body mass index (Ahituv et al., 2007). Of the 1000 variants detected, most were rare variants, including over 270 non-synonymous mutations, many of which were found only in the obese cohort and thus become strong functional candidates for a role in obesity. Searching for Somatic Mutations While genome-wide association studies are restricted to inherited variation, medical resequencing studies can target either inherited or somatic variants. In cancer especially, it is of interest to use medical resequencing to search for somatic mutations in tumor tissue in order to identify genes potentially relevant to cancer progression (Greenman et al., 2007; Sjoblom et al., 2006). Two important points emerge from such studies. First, the genes implicated by virtue of discovering rare somatic variants in multiple cases of a particular cancer tend to be different from those identified in previous genetic studies as inherited risk factors. This provides novel insights into the biology of human cancer and suggests candidates for further exploring mechanisms of tumorigenesis or metastasis or for developing therapeutic approaches. Second, however, the large number of mutations uncovered by this approach introduces the need for caution, as many will be “bystander” or “passenger” mutations around the genome that reflect only the increased mutation rate associated with cancer, not genes involved directly in cancer.
FROM GENOME TO PERSONALIZED MEDICINE Of all the promises of the current scientific and social revolution stemming from advances in our understanding of the human
■
15
genome and its variation, genomic and personalized medicine may be the most eagerly awaited. The prospect of examining an individual’s entire genome (or at least a significant fraction of it) in order to make individualized risk predictions and treatment decisions is an attractive, albeit challenging, one (Bentley, 2004; Willard et al., 2005). Having access to the reference human genome sequence has been transformational for the fields of human genetics and genome biology, but by itself is an insufficient prerequisite for genomic medicine. As discussed in subsequent chapters, equally important are the various complementary technologies to reliably capture information on individual genomes, their epigenetic modification, and their derivatives the transcriptome, proteome and metabolome for health and disease status (Table 1.4). Each of these technologies provides information that, in combination with clinical data and evaluation of environmental triggers, will in time contribute to assessment of individual risks and guide clinical management and decision-making (Figure 1.4). Critical enablers of this new approach have been innovations in laboratory technology (to address biologically and medically relevant questions on a scale and with a throughput hardly imaginable just a few decades ago), paired with equally transforming developments in informatics and information systems to handle the onslaught of genomic data (West et al., 2006) (see Chapter 17). Personal Genomics At the heart of the genomic approach to personalized medicine will be information from individual genomes, a fast-moving area of technological development that is spawning a social and information revolution among consumers. Dramatic improvements in sequencing technology (Bentley, 2006) have reduced the cost and time of resequencing projects to a level that invites conjecture about the long awaited “$1000 genome” (Dalton, 2006; Service, 2006; Wolinsky, 2007). The much-ballyhooed release of the genome sequences of Craig Venter (“HuRef ”) (Levy et al., 2007) and James Watson has stirred up concerns about “celebrity genomics” (Check, 2007); however, while the Venter and Watson sequences may have been the first, they will not be alone for long, as numerous additional genomes are already in various sequencing pipelines (Table 1.5). What remains unsettled for now is what degree of genome surveillance will be most useful, either for research or for clinical practice, a topic that is returned to frequently in subsequent chapters in this volume. While whole-genome sequencing is increasingly possible, it is unlikely to provide more information about established disease associations than would highdensity, genome-wide SNP genotyping. Targeted resequencing of, for example, exons and known regulatory regions would allow detection of rare variants in portions of the genome most relevant to disease at a fraction of the cost of whole-genome sequencing (Hodges et al., 2007). The availability of associated clinical data is variable among the studies announced to date, but there exists, at least among some participants, a strong sense of “health-information altruism” to contribute to the much needed large-scale correlation
16
CHAPTER 1
TABLE 1.4
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
Personalized signatures of health or disease
Dataset
“omic” approach
Technology platform or approach
Chapter number(s)
Human genome sequence
Genomics
Single nucleotide polymorphisms (SNPs); Other genome variants
Chapters 7–9
Gene expression profiles
Transcriptomics
Microarrays of 20,000 gene transcripts
Chapters 12, 13
Protein abundance
Proteomics
Protein arrays of specific protein products
Chapter 14
Metabolites
Metabolomics
Analysis of hundreds to thousands of metabolites
Chapter 15
Chromatin
Epigenomics
Array- or sequence-based assessment of chromatin modification
Chapters 5, 11
Gene networks, interactions
Systems biology
Large-scale interactions among genes or proteins
Chapter 6
Microbiome
Metagenomics
Analysis of viral, fungal and bacterial populations in human specimens
Chapter 48
Genomic and clinical data
Informatics
Integrated databases of “omic” data and electronic health records
Chapters 17–21
Individual A
Individual B
• Predisposition to disease
Genome sequence Somatic mutation Epigenetic changes Personal environment
Transcriptome
Proteome
• • • •
Diagnosis Natural history Treatment options Prognosis
Phenotype
Figure 1.4
The promise of personal genomics in the era of genomic and personalized medicine.
of genotype and phenotype (Kohane and Altman, 2005). Notwithstanding individual’s willingness to make genome sequence data (much less medical information) available moreor-less publicly, substantial concerns have been raised about privacy, since a surprisingly (to some) small number of SNPs or other genome variants are sufficient to allow identification of individuals (Lin et al., 2004; McGuire and Gibbs, 2006).
A Consumer Revolution While the genome revolution has without doubt been driven by technological improvements and by an explosion in the availability of genome data, the push for incorporation of genome information into clinical practice may come as much or more from consumers as from professionals. A half dozen or more companies are already offering genome-wide SNP profiles to
From Genome to Personalized Medicine
TABLE 1.5
■
17
Individual sequenced human genomes (as of March 2008)
Individuals
Nature of genome
Method
Year completed/ anticipated
References
“RP11”a
Partial haploid
Sanger sequencing
2003
International Human Genome Sequencing Consortium, (2001); Osoegawa et al. (2001)
J. Craig Venter
Complete diploid
Sanger sequencing
2007
Levy et al. (2007)
James D. Watson
Complete diploid
454 pyrosequencing
2007
Wheeler et al. (2008)
YanHuang researcher
Complete diploid
Illumina sequencing by synthesis
2007
Qiu and Hayden (2008); Xin (2007)
YanHuang paying customer
Complete diploid
Illumina sequencing by synthesis
2008
Qiu and Hayden (2008)
Yoruba HapMap sample
Complete diploid
Illumina sequencing by synthesis
2008
Knome customers
Complete diploid
Various next-generation technologies
2008 and beyond
Personal Genome Project; 10 individuals
“Exomes” (coding exons only)
Polony sequencing
2008
Venter institute; 10–50 genomes
Complete diploid
Various next-generation technologies
2008
ClinSeq 1000
Exons of 200–400 cardiovascular genes
To be determined
2009 or beyond
http://www.genome. gov/25521304
YanHuang 99
Complete diploid
Various next-generation technologies
2010
Xin (2007)
GATC 100
Complete diploid
Various next-generation technologies
2010
http://www.gatc-biotech.com/ en/discover_gatc/research/ Human_Genome_Sequencing_ Service.php
1000 Genomes Project
Variousb
Various next-generation technologies
2010
www.1000genomes.org; http:// www.genome.gov/26524516; Hayden (2008); Kaiser (2008)
Personal Genome – Project; 500–100,000
Exomes
Polony sequencing
?
http://www. personalgeneomes.org
http://www.personalgenomes. org; Church (2006)
Source: Courtesy of Misha Angrist, Duke Institute for Genome Sciences & Policy a An anonymous male from Buffalo, New York, whose genome was overrepresented in the publicly funded Human Genome Project b 6 people fully sequenced at 20x coverage; 180 at 2x; and exons of 1000 genes in 1000 people
the public, some with associated risk estimates for relevant clinical conditions (see Chapter 21). At least a few companies will also sequence individual genomes for a cost that, while high currently, is not out of reach for some individuals. Some in the medical community have called such genome scans premature and ill advised (Hunter et al., 2008), and similar concerns have been raised for more traditional and less comprehensive commercial tests targeted to specific genes or to specific conditions (Janssens et al., 2008). Hesitation on the part of health
professionals aside, however, it seems clear that at least some consumers, whatever their motivation, will choose to test their genomes, regardless of whether any of the findings are (yet) considered actionable by their physicians. This creates both the need and an opportunity for health professionals to stay abreast of the rapidly developing scientific foundations of genomic medicine, to place in context the dynamic and still poorly understood interplays of the genome with the environment and to communicate effectively the complexities of overall genetic risk to their patients.
18
CHAPTER 1
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
CONCLUSION In this chapter, we have described the organization, variation and expression of the human genome as a foundation for the many chapters to follow on principles of human genomics, on new technologies in genomics and informatics, on approaches in translational genomics and, finally, on the first applications of genomic and personalized medicine to specific diseases. What is now referred to as the “Genome Revolution” has its roots in a technological revolution over the past two decades that has transformed biology and human genetics as they transition from discovery-based analog sciences to digital systems based on comprehensive information from the genome and its derivatives. Advances in genome technology and the resulting
explosion in knowledge and information stemming from the Human Genome Project are now playing a major and increasingly transformational role in our understanding of human health and in the delivery of health care. While the challenges both for the health care profession and for society are significant, the development of comprehensive, cost-efficient and high-throughput technologies, combined with powerful tools in biomedical and clinical informatics for analyzing and storing vast amounts of data, will enable the growing field of genomic and personalized medicine. As the chapters to follow illustrate, the genome sciences – encompassing large-scale analyses of individual genomes, epigenomes, transcriptomes, proteomes and metabolomes – will drive profound changes in clinical medicine.
REFERENCES Ahituv, N., Kavaslar, N., Schackwitz, W., Ustaszewska, A., Martin, J., Hebert, S., Doelle, H., Ersoy, B., Kryukov, G., Schmidt, S. et al. (2007). Medical sequencing at the extremes of human body mass. Am J Hum Genet 80, 779–791. Altshuler, D. and Clark, A. (2005). Harvesting medical information from the human family tree. Science 307, 1052–1053. Bailey, J. and Eichler, E. (2006). Primate segmental duplications: Crucibles of evolution, diversity and disease. Nat Rev Genet 7, 552–564. Barrera, L. and Ren, B. (2006). The transcriptional regulatory code of eukaryotic cells – insights from genome-wide analysis of chromatin organization and transcription factor binding. Curr Opin Cell Biol 18, 291–298. Beckmann, J., Estivill, X. and Antonarakis, S. (2007). Copy number variants and genetic traits: Closer to the resolution of phenotypic variability. Nat Rev Genet 8, 639–646. Bejerano, G., Pheasant, M., Makunini, I., Stephen, S., Kent,W., Mattick, J. and Haussler, D. (2004). Ultraconserved elements in the human genome. Science 304, 1321–1325. Bentley, D.R. (2004). Genomes for medicine. Nature 429, 440–445. Bentley, D.R. (2006). Whole-genome resequencing. Curr Opin Genet Deve 16, 545–552. Bentwich, I. (2005). Identification of hundreds of conserved and nonconserved human microRNAs. Nat Genet 37, 766–770. Bernstein, B., Meissner, A. and Lander, E. (2007). The mammalian epigenome. Cell 128, 669–681. Bolnick, D., Fullwiley, D., Duster, T., Cooper, R., Fujimura, J., Kahn, J., Kaufman, J., Marks, J., Morning, A., Nelson, A. et al. (2007). The science and business of genetic ancestry training. Science 318, 399–400. Bolzer, A., Kreth, G., Solovei, I., Koehler, D., Saracoglu, K., Fauth, C., Muller, S., Eils, R., Cremer, C., Speicher, M. et al. (2005). Threedimensional maps of all chromosomes in human male fibroblast nuclei and prometaphase rosettes. PLoS Biol 3, e157. Brena, R., Huang, T. and Plass, C. (2006). Toward a human epigenome. Nat Genet 38, 1359–1360. Buckland, P. (2003). Polymorphically duplicated genes: Their relevance to phenotypic variation in humans. Ann Med 35, 308–315. Caron, H., van Schaik, B., van der Mee, M., Baas, F., Riggins, G., van Sluis, P., Hermus, M.C., van Asperen, R., Boon, K., Voute, P.A. et al. (2001). The human transcriptome map: Clustering of highly expressed genes in chromosomal domains. Science 291, 1289–1292.
Carrel, L. and Willard, H.F. (2005). X-inactivation profile reveals extensive variability in X-linked gene expression in females. Nature 434, 400–404. Chang, T. and Mendell, J. (2007). microRNAs in vertebrate physiology and human disease. Annu Rev Genomics Hum Genet 8, 215–239. Check, E. (2007). Celebrity genomes alarm researchers. Nature 447, 358–359. Cheung, V. and Spielman, R. (2002). The genetics of variation in gene expression. Nat Genet (Suppl 32), 522–525. Cheung, V., Spielman, R., Ewens, K., Weber, T., Morley, M. and Burdick, J. (2005). Mapping determinants of human gene expression by regional and genome-wide association. Nature 437, 1365–1369. Church, G. (2006). Genomes for all. Sci Am 294, 46–54. Clamp, M., Fry, B., Kamal, M., Xie, X., Cuff, J., Lin, M., Kellis, M., Lindblad-Toh, K. and Lander, E. (2007). Distinguishing proteincoding and noncoding genes in the human genome. Proc Natl Acad Sci USA 104, 19428–19433. Cohen, J., Kiss, R., Pertsemlidis, A., Marcel, Y., McPherson, R. and Hobbs, H. (2004). Multiple rare alleles contribute to low plasma levels of HDL cholesterol. Science 305, 869–871. Conrad, B. and Antonarakis, S. (2007). Gene duplication: A drive for phenotypic diversity and cause of human disease. Annu Rev Genomics Hum Genet 8, 17–35. Conrad, D., Andrews, T., Carter, N., Hurles, M. and Pritchard, J. (2006). A high-resolution survey of deletion polymorphism in the human genome. Nat Genet 38, 75–81. Dalton, R. (2006). Sequencers step up to the speed challenge. Nature 443, 258–259. Deininger, P., Moran, J., Batzer, M. and Kazazian, H., Jr. (2003). Mobile elements and mammalian genome evolution. Curr Opin Genet Dev 13, 651–658. Delwart, E.L. (2007).Viral metagenomics. Rev Med Virol 17, 115–131. Deo, R., Patterson, N.,Tandon, A., McDonald, G., Haiman, C., Ardlie, K., Henderson, B., Henderson, S. and Reich, D. (2007). A high-density admixture scan in 1,670 African Americans with hypertension. PLoS Genet 3, e196. De Vries, B., Pfundt, R., Leisink, M., Koolen, D., Vissers, L., Janssen, I., van Reijmersdal, S., Nillesen, W., Huys, E., de Leeuw, N. et al. (2005). Diagnostic genome profiling in mental retardation. Am J Hum Genet 77, 606–616.
References
ENCODE Project Consortium (2007). Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447, 799–816. Estivill, X. and Armengol, L. (2007). Copy number variants and common disorders: Filling the gaps and exploring complexity in genome-wide association studies. PLoS Genet 3, e190. Eyre-Walker, A. and Hurst, L. (2001). The evolution of isochores. Nat Rev Genet 2, 549–555. Fearnhead, N.S.,Wilding, J.,Winney, B.,Tonks, S., Bartlett, S., Bicknell, D., Tomlinson, I., Montensen, N.J. and Bodmer, W.F. (2004). Multiple rare variants in different genes account for multifactorial inherited susceptibility to colorectal adenomas. Proc Natl Acad Sci USA 101, 15992–15997. Fearnhead, N.S., Winney, B. and Bodmer, W.F. (2005). Rare variant hypothesis for multifactorial inheritance: Susceptibility to colorectal adenomas as a model. Cell Cycle 4, 521–525. Feinberg, A. (2007). Phenotype plasticity and the epigenetics of human disease. Nature 447, 433–440. Felsenfeld, G. and Groudine, M. (2003). Controlling the double helix. Nature 421, 448–453. Feuk, L., Marshall, C., Wintle, R. and Scherer, S. (2006). Structural variants: Changing the landscape of chromosomes and design of disease studies. Hum Mol Genet 15(Review Issue 1), 15–66. Filipowicz, W., Bhattacharyya, S. and Sonenberg, N. (2008). Mechanisms of post-transcriptional regulation by microRNAs: Are the answers in sight?. Nat Rev Genet 9, 102–114. Florez, J., Hirschhorn, J. and Altshuler, D. (2003). The inherited basis of diabetes mellitus: Implications for the genetic analysis of complex traits. Annu Rev Genomics Hum Genet 4, 257–291. Fraga, M., Ballestar, E., Paz, M., Ropero, S., Setien, F., Ballestar, M., Heine-Suner, D., Cigudosa, J., Urioste, M., Benitez, J. et al. (2005). Epigenetic differences arise during the lifetime of monozygotic twins. Proc Natl Acad Sci USA 102, 10604–10609. Francois, P., Scherl, A, Hochstrasser, D., and Schrenzel, J. (2007). Proteomic approach to investigate MRSA. Methods Mol Biol 391, 179–199. Frank, D.N. and Pace, N.R. (2008). Gastrointestinal microbiology enters the metagenomics era. Curr Opin Gastroenterol 24, 4–10. Fraser, P. and Bickmore, W. (2007). Nuclear organization of the genome and the potential for gene regulation. Nature 447, 413–417. Freedman, M., Haiman, C., Patterson, N., McDonald, G., Tandon, A., Waliszewska, A., Penney, K., Steen, R., Ardlie, K., John, E. et al. (2006). Admixture mapping identifies 8q24 as a prostate cancer risk locus in African-American men. Proc Natl Acad Sci USA 103, 14068–14073. Friedman, J., Baross, A., Delaney, A., Ally, A., Arbour, L., Asano, J., Bailey, D., Barber, S., Birch, P., Brown-John, M. et al. (2006). Oligonucleotide microarray analysis of genomic imbalance in children with mental retardation. Am J Hum Genet 79, 500–513. Garrod, A. (1902). The incidence of alkaptonuria: A study in chemical individuality. Mol Med 2, 274–282. Reprint (1996) Gierman, H., Indemans, M., Koster, J., Goetze, S., Seppen, J., Geerts, D., van Driel, R. and Versteeg, R. (2007). Domain-wide regulation of gene expression in the human genome. Genome Res 17, 1286–1295. Gill, S.R., Pop, M., DeBoy, R., Eckburg, P.B.,Turnbaugh, P., Samuel, B.S., Gordon, J.I., Relman, D.A., Fraser-Liggett, C.M. and Nelson, K.E. (2006). Metagenomic analysis of the human distal gut microbiome. Science 312, 1355–1359.
■
19
Gimelbrant, A., Hutchinson, J., Thompson, B. and Chess, A. (2007). Widespread monoallelic expression on human autosomes. Science 318, 1136–1140. Goetze, S., Mateos-Langerak, J., Gierman, H., de Leeuw,W., Giromus, O., Indemans, M., Koster, J., Ondrej, V., Versteeg, R. and van Driel, R. (2007). The three-dimensional structure of human interphase chromosomes is related to the transcriptome map. Mol Cell Biol 27, 4475–4487. Greenman, C., Stephens, P., Smith, R., Dalgliesh, G., Hunter, C., Bignell, G., Davies, H., Teague, J., Stevens, C., Edkins, S. et al. (2007). Patterns of somatic mutation in human cancer genomes. Nature 446, 153–158. Griffiths-Jones, S. (2007). Annotating noncoding RNA genes. Annu Rev Genomics Hum Genet 8, 279–298. Harris, H. (1980). The Principles of Human Biochemical Genetics, 3rd edition. Elsevier,Amsterdam. Harris, H. and Hopkinson, D. (1972). Average heterozygosity per locus in man: An estimate based on the incidence of enzyme polymorphisms. Ann Hum Genet 36, 9–20. Hawkins, D. and Ren, B. (2006). Genome-wide location analysis: Insights on transcriptional regulation. Hum Mol Genet 15(Review Issue 1), 1–7. Hayden, E. (2008). International genome project launched. Nature 451, 378–379. Heintzman, N., Stuart, R., Hon, G., Fu, Y., Ching, C., Hawkins, R., Barrera, L., Calcar, S., Qu, C., Ching, K. et al. (2007). Distinct and predictive chromatin signatures of transcriptional promoters and enhancers in the human genome. Nat Genet 39, 311–318. Heinzen, E., Yoon, W., Tate, S., Sen, A., Wood, N., Sisodiya, S. and Goldstein, D. (2007). Nova2 interacts with a cis-acting polymorphism to influence the proportions of drug-responsive slice variants of SCN1A. Am J Hum Genet 80, 876–883. Hinds, D., Stuve, L., Nilsen, G., Halperin, E., Eskin, E., Ballinger, D., Frazer, K. and Cox, D. (2005). Whole-genome patterns of common DNA variation in three human populations. Science 307, 1072–1079. Hodges, E., Xuan, Z., Balija, V., Kramer, M., Molla, M., Smith, S., Middle, C., Rosesch, M., Albert, T., Hannon, G. et al. (2007). Genome-wide in situ exon capture for selective resequencing. Nat Genet 39, 1522–1527. Hunter, D., Khoury, M. and Drazen, J. (2008). Letting the genome out of the bottle – will we get our wish?. New Engl J Med 358, 105–107. Hurst, L., Pal, C. and Lercher, M. (2004). The evolutionary dynamics of eukaryotic gene order. Nat Rev Genet 5, 299–310. Iles, M. (2008). What can genome-wide association studies tell us about the genetics of common disease?. PLoS Genet 4, e33. International HapMap Consortium (2003). The International HapMap Project. Nature 426, 789–796. International HapMap Consortium (2007). A second generation human haplotype map of over 3.1 million SNPs. Nature 449, 851–862. International Human Genome Sequencing Consortium (2001). Initial sequencing and analysis of the human genome. Nature 409, 860–921. International Human Genome Sequencing Consortium (2004). Finishing the euchromatic sequence of the human genome. Nature 431, 931–945. Janssens, A., Gwinn, M., Bradley, L., Oostra, B., van Duijn, C. and Khoury, M. (2008). A critical appraisal of the scientific basis of commercial genomic profiles used to assess health risks and personalize health interventions. Am J Hum Genet 82, 593–599.
20
CHAPTER 1
■
Organization,Variation and Expression of the Human Genome as a Foundation of Genomic and Personalized Medicine
Kaiser, J. (2008). A plan to capture human diversity in 1000 genomes. Science 319, 395. Kazazian, H.H. and Moran, J.V. (1998). The impact of L1 retrotransposons on the human genome. Nat Genet 19, 19–24. Kim,T., Barrera, L., Zheng, M., Qu, C., Singer, M., Richmond,T.,Wu,Y., Green, R. and Ren, B. (2005). A high-resolution map of active promoters in the human genome. Nature 436, 876–880. Kittles, R. and Weiss, K. (2003). Race, ancestry, and genes: Implications for defining disease risk. Annu Rev Genomics Hum Genet 4, 33–67. Kohane, I. and Altman, R. (2005). Health-information altruists – a potentially critical resource. New Engl J Med 353, 2074–2077. Korbel, J., Urban, A., Affourtit, J., Godwin, B., Grubert, F., Simons, J., Kim, P., Palejev, D., Carriero, N., Du, L. et al. (2007). Pairedend mapping reveals extensive structural variation in the human genome. Science 318, 420–426. Lanctot, C., Cheutin, T., Cremer, M., Cavalli, G. and Cremer, T. (2007). Dynamic genomic architecture in the nuclear space: Regulation of gene expression in three dimensions. Nat Rev Genet 8, 104–115. Lee, C., Iafrate, A. and Brothman, A. (2007). Copy number variations and clinical cytogenetic diagnosis of constitutional disorders. Nat Genet Suppl. 39, 48–54. Levy, S., Sutton, G., Ng, P., Feuk, L., Halpern, A., Walenz, B., Axelrod, N., Huang, J., Kirkness, E., Denisov, G. et al. (2007). The diploid genome sequence of an individual human. PLoS Biol 5(10), e254. Lewontin, R. (1967). An estimate of average heterozygosity in man. Am J Hum Genet 19, 681–685. Ley, R.E., Turnbaugh, P., Klein, S. and Gordon, J.I. (2006). Human gut microbes associated with obesity. Nature 444, 1022–1023. Li, M., Wang, B., Zhang, M., Rantalainen, M., Wang, S., Zhou, H., Zhang, Y., Shen, J., Pang, X., Zhang, M. et al. (2008). Symbiotic gut microbes modulate human metabolic phenotypes. Proc Natl Acad Sci USA 105, 2117–2122. Lim, L., Glasner, M.,Yekta, S., Burge, C. and Bartel, D. (2003).Vertebrate microRNA genes. Science 299, 1540. Lim, L., Lau, N., Garrett-Engele, P., Grimson, A., Schelter, J., Castle, J., Bartel, D., Linsley, P. and Johnson, J. (2005). Microarray analysis shows that some microRNAs downregulate large numbers of target mRNAs. Nature 433, 769–773. Lin, Z., Owen, A. and Altman, R. (2004). Genomic research and human subject privacy. Science 305, 183. Linardopoulou, E.,Williams, E., Fan, Y., Friedman, C.,Young, J. and Trask, B. (2005). Human subtelomeres are hot spots of interchromosomal recombination and segmental duplication. Nature 437, 94–100. Long,W.H., Xiao, H., Gu, X., Zhang, Q.,Yang, H., Zhao, G. and Liu, J.H. (2004). A universal microarray for detection of SARS coronavirus. J Virol Meth 121, 57–63. Margulies, E., Blanchette, M., Haussler, D. and Green, E.NISC Comparative Sequencing Program (2003). Identification and characterization of multi-species conserved sequences. Genome Res 13, 2507–2518. Mattick, J. and Makunini, I. (2006). Non-coding RNA. Hum Mol Genet 15(Review Issue 1), 17–29. McEvoy, B., Beleza, S. and Shriver, M. (2006). The genetic architecture of normal variation in human pigmentation: An evolutionary perspective and model. Hum Mol Genet Suppl. 15, 176–181. McGuire, A. and Gibbs, R. (2006). Genetics. No longer de-identified. Science 312, 370–371. Morley, M., Molony, C., Weber, T., Devlin, J., Ewins, K., Spielman, R. and Cheung, V. (2004). Genetic analysis of genome-wide variation in human gene expression. Nature 230, 743–747.
Nalls, M., Wilson, J., Patterson, N., Tandon, A., Zmuda, J., Huntsman, S., Garcia, M., Hu, D., Li, R., Beamer, B. et al. (2008). Admixture mapping of white cell count: Genetic locus responsible for lower white blood cell count in the health ABC and Jackson Heart Studies. Am J Hum Genet 82, 81–87. Nussbaum, R., McInnes, R. and Willard, H.F. (2007). Genetics in Medicine, 7th edition.WB Saunders Co, Philadelphia. Ogawa,Y. and Lee, J. (2002). Antisense regulation in X inactiviation and autosomal imprinting. Cytogenet genome res 99, 59–65. Online Mendelian Inheritance in Man (2008). http://www.ncbi.nlm. nih.gov/sites/entrez?dbOMIM&itooltoolbar Osoegawa, K., Mammoser, A.,Wu, C., Frengen, E., Zeng, C., Catanese, J. and de Jong, P. (2001). A bacterial artificial chromosome library for sequencing the complete human genome. Genome Res 11, 483–496. Palmer, C., Bik, E.M., DiGiulio, D., Rleman, D.A. and Brown, P.O. (2007). Development of the human infant intestinal microbiota. PLoS Biol 5, e177. Paschou, P., Ziv, E., Burchard, E., Choudhry, S., Rodriguez-Cintron, W., Mahoney, M. and Drineas, P. (2007). PCA-correlated SNPs for structure identification in worldwide human populations. PLoS Genet 3, e160. Pauling, L., Itano, H., Singer, S.J. and Wells, I.G. (1949). Sickle cell anemia a molecular disease. Science 110(2865), 543–548. Peltonen, L. and McKusick, V. (2001). Genomics and medicine. Dissecting human disease in the postgenomic era. Science 291, 1224–1229. Pennacchio, L., Ahituv, N., Moses, A., Prabhakar, S., Nobrega, M., Shoukry, M., Minovitsky, S., Dubchak, I., Holt, A., Lewis, K. et al. (2006). In vivo enhancer analysis of human conserved non-coding sequences. Nature 444, 499–502. Pheasant, M. and Mattick, J. (2007). Raising the estimate of functional human sequences. Genome Res 17, 1245–1253. Price, A., Patterson, N.,Yu, F., Cox, D., Waliszewska, A., McDonald, G., Tandon, A., Schirmer, C., Neubauer, J., Bedoya, et al. (2007). A genomewide admixture map for Latino populations. Am J Hum Genet 80, 1024–1036. Pritchard, J. (2001). Are rare variants responsible for susceptibility to complex diseases?. Amer J Hum Genet 69, 124–137. Qiu, J. and Hayden, E. (2008). Genomics sizes up. Nature 451, 234. Redon, R., Ishikawa, S., Fitch, K., Feuk, L., Perry, G., Andrews, T., Fiegler, H., Shapero, M., Carson, A., Chen, W. et al. (2006). Global variation in copy number in the human genome. Nature 444, 444–454. Riethman, H., Ambrosini, A. and Paul, S. (2005). Human subtelomere structure and variation. Chromosome Res 13, 505–515. Romeo, S., Pennacchio, L., Fu, Y., Boerwinkle, E., Tybjaerg-Hansen, A., Hobbs, H. and Cohen, J. (2007). Population-based resequencing of ANGPTL4 uncovers variations that reduce triglycerides and increase HDL. Nat Genet 39, 513–516. Rudd, M.K. and Willard, H.F. (2004). Analysis of the centromeric regions of the human genome assembly. Trends Genet 20, 529–533. Scherer, S., Lee, C., Birney, E., Altshuler, D., Eichler, E., Carter, N., Hurles, M. and Feuk, L. (2007). Challenges and standards in integrating surveys of structural variation. Nat Genet Suppl. 39, 7–15. Schueler, M. and Sullivan, B. (2006). Structural and functional dynamics of human centromeric chromatin. Annu Rev Genomics Hum Genet 7, 301–313. Sebat, J., Lakshmi, B., Troge, J., Alexander, J., Young, J., Lundin, P., Maner, S., Massa, H., Walker, M., Chi, M. et al. (2004). Large-scale copy number polymorphism in the human genome. Science 305, 525–528.
Recommended Resources
Serre, D., Gurd, S., Ge, B., Sladek, R., Sinnett, D., Harmsen, E., Bibikova, M., Chudin, E., Barker, D., Dickinson, T. et al. (2008). Differential allelic expression in the human genome: A robust approach to identify genetic and epigenetic cis-acting mechanisms regulating gene expression. PLoS Genet 4, e1000006. Service, R. (2006).The race for the $1000 genome. Science 311, 1544–1546. Sharp, A., Cheng, Z. and Eichler, E. (2006). Structural variation of the human genome. Annu Rev Genomics Hum Genet 7, 407–442. Sharp, A., Locke, D., McGrath, S., Cheng, Z., Bailey, J., Vallente, R., Pertz, L., Clark, R., Schwartz, S., Segraves, R. et al. (2005). Segmental duplications and copy-number variation in the human genome. Am J Hum Genet 77, 78–88. Shianna, K. and Willard, H.F. (2006). Human genomics: In search of normalcy. Nature 444, 428–429. Shriver, M. and Kittles, R. (2004). Genetic ancestry and the search for personalized genetic histories. Nat Rev Genet 5, 611–618. Sjoblom, T., Jones, S., Wood, L., Parsons, D., Lin, J., Barber, T., Mandelker, D., Leary, R., Ptak, J., Silliman, N. et al. (2006). The consensus coding sequences of human breast and colorectal cancers. Science 314, 268–274. Skaletsky, H., Kuroda-Kawaguchi, T., Minx, P., Cordum, H., Hillier, L., Brown, L., Repping, S., Pyntikova, T., Ali, J., Bieri, T., et al. (2003). The male-specific region of the human Y chromosome is a mosaic of discrete sequence classes. Nature. 423, 825–837. Smith, M., Patterson, N., Lautenberger, J., Truelove, A., McDonald, G., Waliszewska, A., Kessing, B., Malasky, M., Scafe, C., Le, E. et al. (2004). A high-density admixture map for disease gene discovery in African Americans. Am J Hum Genet 74, 1001–1013. Spielman, R., Bastone, L., Burdick, J., Morley, M., Ewens,W. and Cheung, V. (2007). Common genetic variants account for differences in gene expression among ethnic groups. Nat Genet 39, 226–231. Stefansson, H., Helgason, A., Thorleifsson, G., Steinthorsdottir, V., Masson, G., Barnard, J., Baker, A., Jonasdottir, A., Ingason, A., Gudnadottir, V. et al. (2005). A common inversion under selection in Europeans. Nat Genet 37, 129–137. Stranger, B., Forrest, M., Dunning, M., Ingle, C., Beazley, C., Thorne, N., Redon, R., Bird, C., de Grassi, A.., Lee, C. et al. (2007a). Relative impact of nucleotide and copy number variation on gene expression phenotypes. Science 315, 848–853. Stranger, B., Nica, A., Forrest, M., Dimas, A., Bird, C., Beazley, C., Ingle, C., Dunning, M., Flicek, P., Koller, D. et al. (2007b). Population genomics of human gene expression. Nature Genet 39, 1217–1224. Thorleifsson, G., Magnusson, K., Sulem, P., Bragi Walters, G., Gudbjartsson, D., Stefansson, H., Jonsson, T., Jonasdottir, A., Jonasdottir, A. and Stefansdottir, G. (2007). Common sequence variants in the LOXL1 gene confer susceptibility to exfoliation glaucoma. Science 317, 1397–1400.
21
Tian, C., Plenge, R., Ransom, M., Lee, A., Villoslada, P., Selmi, C., Klareskog, L., Pulver, A., Qi, L., Gregersen, P. et al. (2008). Analysis and application of European genetic substructure using 300K SNP information. PLoS Genet 4, e4. Topol, E. and Frazer, K. (2007). The resequencing imperative. Nat Genet 39, 439–440. Topol, E., Murray, S. and Frazer, K. (2007). The genomics gold rush. JAMA 298, 218–221. Trask, B. (2002). Human cytogenetics: 46 chromosomes, 46 years and counting. Nat Rev Genet 3, 769–778. Turnbaugh, P.J., Ley, R.E., Mahowald, M.A., Magrini, V., Mardis, E. and Gordon, J.I. (2006). An obesity-associated gut microbiome with increased capacity for energy harvest. Nature 444, 1027–1031. Tuzun, E., Sharp, AJ., Bailey, JA., Kaul, R., Morrison, VA., Pertz, LM., Haugen, E., Hayden, H., Albertson, D., Pinkel, D. et al. (2005). Fine-scale structural variation of the human genome. Nature Genet 37(7), 727–732. van Rooij, E., Liu, N. and Olson, E.N. (2008). MicroRNAs flex their muscles. Trends Genet, 24, 159–166. Venter, J., Adams, M., Myers, E., Li, P., Mural, R., Sutton, G., Smith, H., Yandell, M., Evans, C., Holt, R. et al. (2001). The sequence of the human genome. Science 291, 1304–1351. Wang, D., Urisman, A., Liu,Y.T., Springer, M., Ksiazek, T.G., Erdman, D., Mardis, E., Hickenbotham, M., Magrini, V., Eldred, J. et al. (2003). Viral discovery and sequence recovery using DNA microarrays. PLoS Biol 1, e2. Weiss, L.A., Shen, Y., Korn, J., Arking, D., Miller, D., Fossdal, R., Saemundsen, E., Stefansson, H., Ferreira, M., Green, T. et al. (2008). Association between microdeletion and microduplication at 16p11.2 and autism. New Engl J Med 358, 667–675. West, M., Ginsburg, G., Huang, A. and Nevins, J. (2006). Embracing the complexity of genomic data for personalized medicine. Genome Res 16, 559–566. Wheeler, D.A., Srinivasan M., Egholm, M., Shen, Y.., Chen, L., McGuire, A., He, W., Chen, Y.-J., Makhijani, V., Roth, G.T. et al. (2008). The complete genome of an individual by massively parallel DNA Sequencing. Nature 452, 872–876. Willard, H.F., Angrist, M. and Ginsburg, G. (2005). Genomic medicine: Genetic variation and its impact on the future of health care. Philos Trans Roy Soc Lond B Biol Sci 360, 1543–1550. Wolinsky, H. (2007). The thousand-dollar genome: Genetic brinksmanship or personalized medicine?. EMBO Rep 8, 900–903. Wong, K., deLeeuw, R., Dosanjh, N., Kim, L., Cheng, Z., Horsman, D., MacAulay, C., Ng, R., Brown, C., Eichler, E. et al. (2007). A comprehensive analysis of common copy-number variataions in the human genome. Am J Hum Genet 80, 91–104. Xin, H. (2007). The million-dollar genome. Science 318, 550.
RECOMMENDED RESOURCES Brown, T.A. (2006). Genomes 3, 3rd edition. Garland Science Publ., New York. Nussbaum, R., McInnes, R. and Willard, H.F. (2007). Genetics in Medicine, 7th edition. WB Saunders Co., Philadelphia. Strachan, T. and Read, A.P. (2004). Human Molecular Genetics 3, 3rd edition. Garland Science Publ., New York. NCBI Human Genome Resources http://www.ncbi.nlm.nih.gov/Genome/guide/human/ Online Mendelian Inheritance in Man http://www.ncbi.nlm.nih.gov/sites/entrez?dbOMIM& itooltoolbar
■
UCSC Genome Bioinformatics http://genome.ucsc.edu/ Ensembl Human Genome Browser http://www.ensembl.org/Homo_sapiens/index.html Genome Sequence of J. Craig Venter (“HuRef ”) http://www.jcvi.org/cms/research/projects/huref/overview/ Genome Sequence of James D. Watson http://jimwatsonsequence.cshl.edu/about.html
CHAPTER
2 Concepts of Population Genomics Mike E. Weale and David B. Goldstein
INTRODUCTION Over the past decade, population genetics – the study of processes affecting the evolution, distribution and correlation of genetic variants within populations – has truly “gone genomic” (Goldstein and Weale 2001; Jorde et al., 2001). The Human Genome Project provided a blueprint for the “average” human genome. Subsequent projects, and in particular the arrival of affordable high-density genome-wide genotyping, has expanded this picture to provide detailed information on the genomes of many thousands of individuals from different parts of the world, focusing on the genetic differences that distinguish us all. Population genetics has thus been transformed into population genomics. This revolution has resulted in fundamental advances in our understanding of the evolution of the human genome and the history of human populations. But this information is not just of academic interest: it has had a direct bearing on how medical genomic research is conducted today. Indeed, this has been the primary motivation for collecting much of this information in the first place. As medical genomics has moved from the description of simple Mendelian traits to the investigation of complex multifactorial traits, so population genomics has assumed pivotal importance. We need population genomic theory in order to know how to look for variants that affect these traits, how to understand their genealogical history and their geographic distribution, and how to use this information effectively in diagnosis Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 22
and treatment (Box 2.1). Understanding Mendelian traits generally requires only an understanding of variation within families; understanding complex traits requires an understanding of the population within which the complex trait occurs. The first section of this chapter reviews basic concepts in population genomics. The second section reviews what we know about human population genomics, while the third section describes the application of this to medical genomics.
IMPORTANT CONCEPTS IN POPULATION GENOMICS Population genomics is the study, on a genomic scale, of the processes affecting the evolution, distribution and correlation of genetic variants within populations. Population genomics has little interest in the parts of the genome that are the same among all individuals and focuses instead on the genomic differences between us. Roughly 0.1% of our DNA varies appreciably among individuals at the single nucleotide level, and considerably more, we have recently learned (Redon et al., 2006), varies at the larger-scale structural level. The raw engine for variation is mutation (see previous Chapter 1). Once created by a mutation event, two forces – chance (modified by population structure) and selection (only relevant if the variant has an effect on viability or reproductive success) – act on the new variant as it passes from one generation to the next. Many are soon lost from Copyright © 2009, Elsevier Inc. All rights reserved.
Important Concepts in Population Genomics
BOX 2.1
Population Genomics in Action
■
23
Generation t-6 Generation t-5
While many aspects of population genomics can appear technical and somewhat distant from both application and real-world problems and interests in biology and medicine, in fact population genomics is transforming fields as diverse as genetic anthropology and medical genomics. Among the more surprising insights won with our increasing knowledge of the pattern of human genetic variation are a string of observations about relationships among certain populations. Patterns of Y chromosome variation have been used to support the claims of Jewish ancestry of a Bantu-speaking population in South Africa and for a long-term patrilineal pattern of inheritance among Jewish men that consider themselves to be “priests” or Cohanim in Hebrew (see Goldstein, 2008). More recently, Joel Hirschhorn, David Reich and their colleagues have identified a genome-wide signature of Jewish ancestry (Price et al., 2007), using PCA. Similar methodologies have also been developed by Reich and others to study the detailed genomic history of populations that have mixed ancestry. For example, looking at the whole genomes of African Americans with prostate cancer, Reich and his colleagues were able to determine which part of each person’s genome came originally from Africa and which from Europe with a high degree of statistical accuracy (Haiman et al., 2007). Most of the genome reflected the overall proportion of European ancestry in African Americans, but a section of chromosome 8 (within band 8q24) stood out strikingly. African Americans with prostate cancer had a much higher chance of having that region of the genome be of African ancestry compared to the rest of the genome. This result provides overwhelming evidence that there is a major risk factor for prostate cancer in this part of the genome and that this risk factor is likely to contribute to the observed differences in risk of prostate cancer between African Americans and Americans of European ancestry.
the gene pool, but some rise in frequency and may, eventually, come to replace the original ancestral allele. This is how evolution works, and population genomics provides the theory that describes it (Graur and Li, 1999; Hartl and Clark, 2007). Here we briefly review the important population genomic concepts of genealogy, the standard neutral model, population structure and linkage disequilibrium. Our aim is to provide enough information to understand the medically related issues discussed later. For a more thorough coverage of population genetic theory, the reader is referred to one of many standard texts (Gillespie, 2004; Halliburton, 2003; Hartl and Clark, 2007; Hedrick, 2004). Genealogy Imagine you have taken a sample of n individuals and typed them for one single nucleotide polymorphism (SNP) – perhaps a coding polymorphism in a medically relevant gene. The 2n allele copies (two for each diploid individual) are connected by an underlying genealogy, a tree that describes the ancestor– descendent relationships linking all the copies in that population. A simple illustration of how a genealogy might emerge in a population of only four individuals is shown in Figure 2.1. Here the sample size is just n 2 (thus four allele copies, sampled in
Generation t-4
!
Generation t-3 Generation t-2 Generation t-1 Generation t Indiv 1
Indiv 2
Indiv 3
Indiv 4
Figure 2.1 A simple example of a genealogy. Here the population in any one generation consists of just 4 individuals (represented by boxes) and 8 allele copies (represented as circles within the boxes). The genealogy connecting a sample of 2 individuals (and 4 allele copies) is shown. A mutation event in Generation t-3 (indicated by a red exclamation mark) has led to two allele types, coded green and blue.
Generation t) and the population from which they are sampled is fixed in each generation at N 4 individuals. For simplicity, we assume discrete non-overlapping generations, so that the set of parents for individuals in Generation t comes exclusively from the previous generation, Generation t-1. The total set of 2N allele copies in any one generation makes up the gene pool. Consider the two allele copies possessed by Individual 1 in Generation t, one inherited from her mother and the other from her father. A separate lineage of ancestor–descendent transmissions can be traced back in time for both copies (note the same thing can be done for any two allele copies sampled from the gene pool, regardless of whether they happen to belong to the same individual or not). Provided that N is finite, eventually one will reach a previous generation where the two lineages coalesce to a common ancestral copy (in Generation t-6 in the example in Figure 2.1). Note that coalescence is inevitable, given enough time, since in each generation there is a non-zero probability that coalescence will occur. However, the timing of coalescence is typically subject to much statistical uncertainty and, in human populations, is usually much further back in time than the six generations depicted here. Superimposed on the genealogical tree are mutation events, here exemplified by a single event in Generation t-3 that results in two distinct allele types (coded as different colors). This overtly genealogical approach to population genomics, focused on an observed sample of allele copies in the present day and tracing a tree backwards in time from these, has given rise to much new work in population genomics over the past 25 years, collectively known as coalescent theory (Donnelly and Tavaré, 1995; Hein et al., 2005; Kingman, 1982; Nordborg, 2001; Rosenberg and Nordborg, 2002; Wakeley, 2007). This contrasts with a more traditional approach that focuses on the entire population and typically arrives at solutions for tracing changes in allele frequencies forward in time. Perhaps the key computational and
24
CHAPTER 2
■
Concepts of Population Genomics
conceptual simplification afforded by coalescent approaches is that they are able to ignore alleles (and the individuals carrying them) that do not contribute in any way to the genetic diversity that is sampled in the current generation. Hence in this example three individuals are irrelevant in Generation t-6, two in Generation t-5, on so on. Note that while these individuals have not contributed to this sample for this locus, they may have contributed to other loci elsewhere in the genome. Thus, for example, the genealogy for a sample of men typed for their mitochondrial DNA is entirely separate from the genealogy for the same sample of men typed for their Y chromosome, and indeed in this case the individuals lying at the root of these genealogies (corresponding to so-called “mitochondrial Eve” and “Y-chromosome Adam”, provided a large enough sample is taken) are thought to have lived at entirely separate times. Both coalescent theory and the more traditional “forward” approach that models all members of a population have their uses, but we have concentrated here on the coalescent approach because it often helps one’s conceptual understanding of population genomics problems if one thinks in terms of the underlying genealogy. The Standard Neutral Model The standard neutral model (or Fisher-Wright model) provides a convenient null case against which other more complex models can be compared. The two basic assumptions are that (1) mating is random, so any allele copy in Generation t has the same probability of descending from any individual in Generation t-1; and (2) reproductive success is random, so a second allele copy picked from Generation t has the same probabilities of descent as the first one. Under these conditions, the probability that two allele copies in Generation t coalesce in Generation t-1 is 1/(2N). This is simply the probability of picking the same parent again (and the same allele copy within this parent) for the second copy, having picked a parent at random for the first allele copy. Much elegant theory flows from this simple rule, and we give two examples here to give a flavor of this. The first is the derivation of the famous Hardy-Weinberg proportions for genotype frequencies based on allele frequencies. Since the two allele copies possessed by an individual represent random samples from the gene pool under the standard neutral model, if Allele Type i has a population frequency of pi then the probability of seeing two Type i copies in the same individual (i.e., a homozygous ii genotype) is pi2 (due to the multiplicative law governing independent statistical events). Expected frequencies for other homozygous genotypes are similarly derived, and the expected frequency of heterozygotes H, known as the heterozygosity, is what is left over after taking away all the other homozym gotes. Thus H 1 ∑ i1 pi2 , where m is the total number of allele types. Note that in the simple case of just two allele types, these relationships reduce to the more familiar p2, q2 and 2pq for the frequencies of the two homozygotes and the heterozygote respectively, where q 1 p. The second example involves the expected value of H under mutation-drift equilibrium, in a model in which mutations
occur with probability to any one allele copy in any one generation, and in which all mutations lead to a new distinguishable allele type (a reasonable assumption for longer DNA sequences where each new point mutation is likely to occur in a different nucleotide). Consider two allele copies chosen at random in Generation t. Three things can happen to their respective lineages in Generation t-1: (1) they coalesce, with probability 1/(2N); (2) a mutation occurs in one or other ancestor, with probability 2 since there are two ancestors; or (3) nothing interesting happens – just two separate ancestors with no mutation events. If Event (3) happens, one simply asks the same question for Generation t-2, with the probabilities of Events (1), (2) and (3) unchanged. Thus the question of heterozygosity boils down to whether (2) happens before (1) in the genealogy connecting the two allele copies, which in turn depends only on the relative probabilities of (2) versus (1). Thus H 2/(2 1/(2N)) 4N/(1 4N). The same relationship can also be derived, somewhat more laboriously, using more traditional forward-in-time population genetics modeling (see, e.g., Crow, 1986). The relationship predicts that heterozygosity (and hence the amount of genetic variation) at neutral loci should be greater in larger populations and with high-mutation rates, and this pattern is indeed generally seen in both human and non-human populations (Graur and Li, 1999). Population Structure The concept of a completely randomly mating population is patently false for humans. Instead, individuals are more likely to have children with some individuals rather than others. One obvious constraint is geographic distance – you are more likely to marry (or otherwise pair with) someone next door than someone on the other side of the world. National boundaries are important barriers to gene flow (the movement of allele copies from one subpopulation to another), but even within national boundaries marriages are more likely to take place within socioeconomic, cultural and ethnic groups than between them (a phenomenon known as endogamy). For humans, however, none of these barriers is absolute; instead there are simply varying probabilities of having children with anyone else. In population genomics, this complexity is often simplified by compartmentalizing people into distinct groups and then by thinking about the different rates of exchange of allele copies between and within these groups. A useful metric that is often used to describe the degree of population differentiation between these groups is the FST statistic. This can be derived in many ways, and, when these result in slightly different answers, there are some differences of opinion on which derivation is best (Balding, 2003; Excoffier, 2001; Nei, 1987). However, a commonly applied derivation of FST is based on contrasting the amounts of genetic variance attributable to within- and between-population differences. For various reasons (Excoffier, 2001; Nei, 1987), the heterozygosity H that we defined earlier is an excellent measure of the genetic variance of a population. Intuitively, one can see that H approaches zero as genetic variability is reduced (i.e., the population becomes dominated by a single allele type) and conversely approaches unity as genetic
Important Concepts in Population Genomics
variability is increased (i.e., the population becomes dominated by many different allele types, such that the probability of picking two of the same type from the gene pool becomes vanishingly small). One interpretation of FST, then, is that it represents the proportion of total heterozygosity, HT (obtained by pooling all populations together), that is attributable to between-population differences. Since it’s easier to calculate the average withinpopulation contribution (simply the mean heterozygosity H calculated over all within-population H values), and since the between- and within-population contributions together sum to HT, the definition of FST is then (H T H )/H T ). For most of the genome, FST values range from a maximum of 15% for intercontinental differences, to typically 1–3% for international differences within a continent, to less than 0.1% for differences within countries in Europe (Cavalli-Sforza et al., 1994; Steffens et al., 2006). Genetic markers with high FST are used for a number of purposes in genomic medicine, including ancestry testing and tracing the likely origin(s) of all or portions of one’s genome (Box 2.1). Linkage Disequilibrium Linkage Disequilibrium, or LD, is a measure of the degree of nonrandom association between allelic forms at two different sites (or loci) in the genome. Two factors influence this: mutation and recombination. Since the chances of recombination increase the further apart two loci are, LD tends to be highest between loci that lie very close to each other in the genome. However, there is a lot of variability around this general rule, partly because the rate of recombination varies throughout the human genome (Myers et al., 2005) and partly because chance factors in the placement of mutations in the underlying genealogy also play a role. These points are illustrated in Figure 2.2. In the absence of recombination (i.e., if the loci are very close to each other), one still sees variation in the degree of LD between two loci. If the polymorphism-generating mutations happen to occur along the same branch of the genealogy (Figure 2.2a), then only two out of the possible four 2-locus haplotypes are formed, and in this case knowledge of the allele type at one locus allows us to perfectly predict the allele type at the other. But if instead the mutations occur at widely separated points on the genealogical tree (Figure 2.2b), then knowledge of the allele type at one locus is only partially predictive of the allele type at the other. This breakdown of association, due simply to mutation, may or may not be of interest to the population geneticist depending on the application. If one is interested in the prediction of allele type at one locus depending on the allele type at the other, then mutation-led breakdown of LD is of interest and a useful LD metric in this case is r 2, which is the amount of variation at locus A that is explained by knowledge of allele states at locus B. If locus A and B are binary SNPs, then r 2 can be found simply by converting the allelic states into 0/1 variables, then calculating the square of the standard Pearson correlation coefficient between them. If instead one is interested only in the supposed amount of recombination between locus A and B, then a metric that
■
25
attempts to factor out the mutation-led breakdown of association is preferable. One such metric is D, which for SNP loci takes its maximum value of 1 if and only if one observes just three out of the possible four haplotypes in one’s sample (for (a) Generation t-6 Generation t-5 Generation t-4
!!
Generation t-3 Generation t-2 Generation t-1 Generation t Indiv 1
Indiv 2
Indiv 3
Indiv 4
Indiv 3
Indiv 4
Indiv 3
Indiv 4
(b) Generation t-6 Generation t-5 Generation t-4
! !
Generation t-3 Generation t-2 Generation t-1 Generation t Indiv 1
Indiv 2
(c) Generation t-6 Generation t-5 Generation t-4
! !
Generation t-3 Generation t-2
X
Generation t-1 Generation t Indiv 1
Indiv 2
Figure 2.2 Example genealogies involving 2 linked loci. (a) Both mutations happen on the same genealogical branch, leading to just two haplotypes – green red and blue yellow. Thus knowledge of the allele type at one locus provides perfect prediction of the allele type at the other. LD metrics: r2 1, D 1. (b) The two mutations occur separately on the genealogy tree, leading to three haplotypes and imperfect prediction. LD metrics: r2 0.33, D 1. (c) A recombination event in Generation t-2 (indicated by a red X) has created a recombinant haplotype with contributions from two separate lineages, leading to four different haplotypes and further reducing LD (to zero in the toy example, but this is not a general rule). LD metrics: r2 0, D 0.
26
CHAPTER 2
■
Concepts of Population Genomics
reviews of LD metrics see Devlin and Risch, 1995; Hudson, 2001; Nordborg and Tavaré, 2002; Mueller, 2004). Looking at Figure 2.2b, one should note that wherever one places the two mutations one can never obtain more than three haplotypes in the eventual sample, so D is always 1. Assuming no further mutations, the only way to obtain all four haplotypes in the eventual sample is via one or more recombination events between the two loci, and this is illustrated in Figure 2.2c. Note that recombination events alter the underlying genealogy, which is no longer a bifurcating tree but rather a more complicated web (technically, a “graph”). Linkage disequilibrium is both a boon and a bane for medical genomics, and this is discussed in more detail later.
HUMAN POPULATION GENOMICS We start by describing important sources of genomic data that have emerged recently. Then, we describe what these have told us about the history and evolution of human populations and about how human populations are structured today. Important Sources of Data The HapMap dataset (http://www.hapmap.org/) today constitutes the most important source of information for population genomics (The International HapMap Consortium, 2005, 2007). Instigated in 2002 with funding from the US NIH and other international sources, the project set out to describe and type common SNP variation, evenly spread throughout the human genome, in a collection of 270 individuals sampled from four populations spanning three major continents of origin – Africa, Asia and Europe. This was accomplished in two main phases – an initial Phase I completed in early 2005, featuring around 1.3 million common SNPs at an average density of 0.4 SNPs per kilobase (The International HapMap Consortium, 2005), and a subsequent Phase II completed in late 2005, featuring around 3.1 million common SNPs at an average density of 1 SNP per kilobase (The International HapMap Consortium, 2007). In addition, 10 separate 500 kb regions of the genome (the HapMap-ENCODE regions) were targeted for exhaustive discovery and typing of all but the rarest SNPs, in order to provide a valuable reference source to compare the projected performance of the rest of the HapMap data in representing all common SNP variation. It is estimated that there are some 10 million common SNPs in the human genome (Carlson et al., 2005). HapMap Phase II typed only one third of these; however, comparisons using the HapMap-ENCODE regions show that, thanks to linkage disequilibrium, these 3.1 million SNPs represent or “tag” the other 7 million to a high degree of accuracy (The International HapMap Consortium, 2007). The HapMap project, along with earlier SNP discovery projects such as The SNP Consortium (The International SNP Map Working Group, 2001), played an important role in the development of high-density genome-wide SNP panels by greatly increasing our knowledge of the location of SNPs in the human genome. In addition, the HapMap project provided the
LD data that allowed tag-based SNP panels to be designed (see Chapter 8). Genome-wide SNP panels are now being applied to population samples from around the world, providing valuable extensions of the information available from HapMap in two important ways. Firstly, new datasets have increased the sample sizes for certain populations to thousands of individuals; see, for example, the iControl database (US individuals, http:// www.illumina.com/pages.ilmn?ID231), the WTCCC database (UK individuals, http://www.wtccc.org.uk/) and the dbGaP database (http://www.ncbi.nlm.nih.gov/sites/entrez?dbgap). Secondly, new datasets have expanded the set of populations beyond the four studied for HapMap; see, for example, the Stanford University HGDP-CEPH database (51 distinct worldwide population samples, http://www.cephb.fr/hgdp-cephdb/). A key element in all these datasets is their free availability to the scientific community, reflecting a sea-change in attitudes to data access that has beneficially affected almost all genomic research since the Human Genome Project (see also The GAIN Collaborative Research Group, 2007). Currently, most databases characterize just the SNPs in the human genome. However, much interest is now also focused on Copy Number Variation (CNVs), and this is likely to widen soon to other types of structural variation such as inversions (see previous Chapter 1). The indications are that SNPs have a fair, but not perfect, ability to tag CNVs (Locke, 2006; Redon et al., 2006), and thus the development of genomic databases specifically targeting structural variation remains a priority. A Brief History of Homo sapiens The new genomic data have broadly confirmed the picture of our genetic history that had emerged from previous genetic studies and are also consistent with what we know from paleoanthropological data. This picture is of a relatively small population of ancestral humans living in East Africa between 60–100 thousand years ago, spreading out-of-Africa to central Asia and thence to Europe in one direction and South and East Asia and the Americas in the other, replacing other extant human populations as they went (Cavalli-Sforza et al., 1994; Garrigan and Hammer, 2006; Goldstein and Chikhi, 2002; Jobling et al., 2003). This picture explains why (1) compared to those of other primate species, our genomes are remarkably low in genetic diversity within populations and in differentiation between populations; (2) this diversity steadily decreases in populations further removed from Africa (Prugnolle et al., 2005; Ramachandran et al., 2005); (3) non-African diversity tends to be a subset of African diversity or, to put it in genealogical terms, most gene trees have their root in Africa; and (4) LD in the human genome is high in non-African populations. Not enough time has passed since our recent, small origins to have allowed mutations and recombinations to break down the haplotypic patterns present in the genomes that emerged from Africa. There is still room, and indeed eagerness, for debate regarding this picture. Debate continues over the degree to which there was interbreeding with extant populations either before or after the out-of-Africa expansion (Garrigan and Hammer, 2006;
Human Population Genomics
AA C EA M SA
0.05
0.10
2007; Rosenberg et al., 2005; Serre and Pääbo, 2004). Thus, the genome of any one single individual is not composed of a contribution from one single discrete ancestral group, nor even of a series of contributions from different discrete ancestral groups. Just as there are no discrete groups today, so there were none in the past. For these reasons, a more useful paradigm for human population genomic structure is to see it as comprised of a series of quantitative axes, each one reflecting one aspect of a multidimensional structure. We illustrate some of these points in Figure 2.3. The figure displays the principal aspects of the population genomic structure of a sample of approximately 500 individuals collected in North Carolina, USA. Each individual was typed using the Illumina HumanHap 550 k genome-wide SNP panel (see Chapter 8), and these data were then subjected to Principal Components Analysis (PCA), a statistical method that ranks and dissects the correlation structure of multivariate datasets (Patterson et al., 2006; Price et al., 2006). Among the population groups contained within the cohort, some of the allele frequencies change in correlated ways that distinguish between population groups. Thus, in a simple and extreme case, you might have 10,000 SNPs that are all high frequency in group A and low frequency in group B; these would have high FST values, as described earlier.These SNPs would contribute to a single axis (e.g., PC1 in Figure 2.3), and the position of an individual along that axis would be dependent on which of
PC 2 0.00
Population Genomic Structure and Race The new genomics has allowed us, at last, to separate the population genetic concept of biogeographical shared ancestry from the socio-cultural concept of race. This is because the quantity of genetic data that one can routinely collect on a single individual is enough to provide all one needs to draw up a remarkably accurate picture of the population structure of a sample of individuals. In the past, with less genetic data available, other information such as the individual’s own self-declared racial identity was needed to shore up uncertainties in the underlying population genomic structure, and much debate ensued over the effectiveness of using one to predict the other (Barnholtz-Sloan et al., 2005; Burnett et al., 2006; Liu et al., 2006b; Tang et al., 2005; Wilson et al., 2001). Now that debate is largely irrelevant – we can get all the information we need regarding patterns of ancestral relatedness from the genetic data alone. The fact that confusion still exists over conflating population genomic structure with sociocultural race stems from many reasons, but the research community is at least partly to blame. Recent surveys suggest that the term “race” continues to be used ambiguously both in the scientific literature (Sankar et al., 2007; Shanawani et al., 2006) and in dealings with the press (Condit, 2007; Lynch and Condit, 2006). Genetically, there are no such things as human “races”, in the original strict sense of discrete unequivocal groupings. Because of our recent shared roots and our considerable mobility, we are only marginally more similar to a person sharing common biogeographic ancestry and living locally than we are to a person with different roots living on the other side of the world. Leaving aside sex-linked loci, which one must treat separately, average FST never rises much above 15% even for intercontinental comparisons (Cavalli-Sforza et al., 1994), so at least 85% of all genetic differences are accounted for by “within population” variance. There are no absolute barriers to gene flow between any two parts of the world, and, while partial barriers exist, the location and extent of these are difficult to define unequivocally (Barbujani and Belle, 2006; Handley et al.,
27
0.05
Harding and McVean, 2004; Liu et al., 2006a; Ray et al., 2005), in particular regarding whether there was interbreeding with Neanderthals in Europe (Currat and Excoffier, 2004; Trinkaus, 2007). This debate revolves not only around the interpretation of genetic data, but also around morphological data such as cranial features, again with contrasting conclusions (Manica et al., 2007; Trinkaus, 2005). Likewise there is debate over the mode of the out-of-Africa expansion, for example over the use of coastal routes as preferred routes for expansion (Amos and Manica, 2006; Macaulay et al., 2005) and over the number and extent of population bottlenecks that may have occurred along the way (Liu et al., 2006a; Ramachandran et al., 2005). While our knowledge of genetic variation in the present day is excellent, the use of this to make inferences about the past is not an exact science. It may be that ancient DNA techniques will help to resolve these issues, by providing more direct information on the population genetics of these ancient populations (Caramelli et al., 2007; Noonan et al., 2006;Vernesi et al., 2004).
■
0.00
0.05 PC 1
0.10
Figure 2.3 Population structure in a sample collected in North Carolina, USA, revealed by Principal Components Analysis applied to genome-wide SNP data (Duke Genetics of Memory Cohort typed on Illumina HumanHap 550 panel). Scores for the first two principal component axes are plotted here. Self-declared racial identity labels are: AA African/AfricanAmerican; C Caucasian; EA East Asian; M Hispanic/ Mixed; SA South Asian. See text for further details. Figure based on data provided by D. Goldstein and A. Need.
28
CHAPTER 2
■
Concepts of Population Genomics
the two groups that individual came from. Different axes would reflect other correlations in the data, involving, for example, a different subset of SNPs that also distinguish different population groups (e.g., PC2 in Figure 2.3). In this way, PCA uses the correlation patterns in the data to reduce the dimensionality of the dataset. Formal methods exist to determine how much information is reflected in each of the axes. In real datasets, the correlation patterns are much less extreme and simple than the group A and group B case described above, but the idea is the same. The first point the figure illustrates is that, even though PCA is based solely on genotype information and is thus blind to non-genetic information, there is a remarkably good correspondence with self-declared racial identity. Most individuals belonging to three population subgroups (Caucasian in red, South Asian in turquoise, and East Asian in black) are clearly identified and separately from each other by analysis of their genotypes (Figure 2.3). However, a second point, and just as important as the first, is that there is additional population structure that is not picked up by the information on racial identity. For example, there is a large spread in the African American group, reflecting individual differences in the degree of European admixture in their genomes. Thirdly, there is more than one axis along which co-ancestry relationships can be defined, and in fact the PCA reveals many more significant axes for this dataset than the two used to plot Figure 2.3. This underlines the multidimensional nature of human population genomic structure. Finally, the degree of resolution in elucidating such structure, as evidenced by the sharpness in the clustering of some of the groups in Figure 2.3, is something altogether new, and is a largely unforeseen benefit of the step-increase in genetic data afforded by the latest genome-wide SNP chips.
APPLICATION OF POPULATION GENOMICS TO GENOMIC MEDICINE Here we show how the concepts we have discussed so far can be applied to genomic medicine. The first two sections deal with the important role that LD and population structure have in the search for medically relevant causal variants, while the last deals with the insights beginning to emerge on the common population genomic patterns describing these variants (Box 2.1). LD in the Search for Medically Relevant Variants Today the hunt for medically relevant variants is taking place in large population-based samples, using genotyping panels that scan the entire genome. This creates both challenges and benefits, and both require an understanding of population genomics. In essence, the reason why association studies are so much more powerful than linkage studies at localizing a causal genetic variant is that linkage studies only make use of recombination events taking place over one or two generations in the pedigree under study, whereas association studies make use of the much larger number of recombination events that have taken place over the entire genealogy linking the individuals in a sample at a given
locus. Thus the breakdown in association with a nearby marker due to LD happens over a much shorter distance than the breakdown in linkage due to observed recombination events within pedigrees. The cost is that you need to cover the genome with a much greater density of markers. Using a simple theoretical model, Kruglyak estimated that this breakdown might happen over a 3 kb distance (Kruglyak, 1999). We now know that breakdown of LD is both region-specific (i.e., varies between different parts of the genome) and population-specific. The former is due at least in part to variation in the rate of recombination throughout the genome as a reflection of genome organization (Myers et al., 2005), while the latter is due to differences in the demographic histories of different populations. For example, levels of LD in European populations are higher than those in African populations and higher than the level estimated by Kruglyak (1999). This is presumed to be due to one or more population bottlenecks that have occurred since the out-of-Africa expansion (bottlenecks generate higher LD by shortening the average genealogical tree linking a sample of individuals). In a practical sense, this means that fewer SNPs are needed to cover the genome in Europeans than in Africans. Still greater efficiency can be achieved, however, by leveraging what we know about the specific patterns of LD to pick out a special set of SNPs that still largely represents all the common variation in a given genomic region – a procedure known as tagging (de Bakker et al., 2005; Weale et al., 2003). This idea is now applicable to the entire genome, with the availability of genomewide genotyping arrays containing hundreds of thousands of SNPs. Thus either through simply the sheer density of SNPs (e.g., Affymetrix platforms) or through explicit use of tagging methods to guide the choice of SNPs in the first place (e.g., Illumina platforms), these modern genome-wide platforms are able to represent all common variation in the human genome to an extremely high level of accuracy (see Chapter 8). Recent studies have shown that the ability to port tagging SNP sets inferred from one population to another related one (e.g., from the same continent) is generally good (Conrad et al., 2006; de Bakker et al., 2006). Since LD varies in extent among different populations, this presents the possibility of conducting different studies at different stages in different populations in order to maximize efficiency. Two types of special populations have been used to capitalize on larger levels of LD in the initial scan for causal variants. One of these is the admixed population, one in which individuals are descendents of a recent period of admixture between two previously separated ancestral populations. The best-studied example of this is the African American population, as on average their genome is comprised of an 80:20 mixture of West African and European DNA, although there is quite a high degree of variation among different individuals who self-identify as African American in terms of the proportion of the genome that is of West African or European origin (Parra et al., 1998). The other type is the isolated population, one in which the individuals descend from a small number of founders and have, through endogamy, remained relatively free from subsequent admixture (Service et al., 2006). One well-studied example is the Dutch “GRIP” population (Liu et al., 2007). Despite their differences
Conclusions
in origin, both admixed and isolated populations have genomes with elevated levels of LD, and this can be used to positive effect in the early stages of the search for new causal variants (Angius et al., 2008; Seldin, 2007; Smith and O’Brien, 2005). Ultimately, though, the very same LD that allows us to economize in the number of SNPs needed to perform effective genome-wide association studies also acts as a barrier, limiting one’s ability to localize a causal variant based on association alone. Instead, a signal pointing to causal involvement of locus A could equally be due to the actual influence of locus B, if locus B is in high LD with locus A. In such cases, other types of information (e.g., based on the location of the respective polymorphisms and their predicted functional influence) are needed to tease apart the causal relationships. Correcting for Population Structure We have already noted that human populations have a complex structure (also known as stratification) that neither conforms to the notion of distinct races nor to that of one single randomly mating gene pool but rather falls somewhere in between. In practice, what this means is that even within a carefully designed study that attempts to match cases and controls for biogeographical shared ancestry, there may be hidden structure within the study. This creates a problem if cases and controls are nonrandomly distributed with respect to this structure. To illustrate this, consider again the structure revealed in Figure 2.3. Imagine a case-control study conducted within the African American group, in which cases are more likely to come from the bottom right and controls from the top left of the cluster. This sets up a situation in which SNPs that differentiate strongly between these two sides would also differentiate cases and controls, resulting in many false signals of association. Luckily, we can use the fact that population structure affects all SNPs in the genome to affect a correction. Older genomic control methods applied a uniform correction to all SNPs (Devlin et al., 2004; Devlin and Roeder, 1999). Newer methods allow a SNP-specific correction to be applied, tailored to the observed degree to which the SNP in question is seen to be affected by population structure (some SNPs vary more simply by evolutionary chance (Balding and Nichols, 1995), others because of presumed selective gradients (Campbell et al., 2005)). Of these newer methods, the EIGENSTRAT method of Price et al. (2006) is particularly convenient and flexible because it (a) makes use of PCA to efficiently dissect the population structure; (b) uses as many co-ancestry related principal component axes as are deemed necessary to effect the correction (guided by theory developed in Patterson et al., 2006); and (c) automatically recognizes, since principal component axes are quantitative, that human population structure is continuous rather than discrete. Understanding the History and Population Distribution of Variants In this final section, we anticipate the forthcoming flood of data on causal variants that will be discovered as a result of genome-wide association studies on common complex diseases
■
29
(Frayling, 2007). These will soon allow the resolution of questions that have been debated for decades regarding the population genomic properties of causal variation relative to other variation in the human genome (Mitchell-Olds et al., 2007). An early example of this is a meta-analysis by Ioannidis et al. (2004) on the question of whether causal variants tend to exert consistent effects in different parts of the world – in other words, does possession of an AA genotype for locus Q always increase your risk of a given disease relative to genotype aa, or is it different for people living in Africa versus people living in Asia? Their study suggested that the general rule was the former and that differences in population risk were more due to differences in allele frequencies among populations than to differences in the effect of possessing a given genotype. The take-home message is that a new causal variant discovered in Population A because it is at high frequency can still lead to new therapies of benefit to people living in Population B, where the causal variant is rare, because the underlying metabolic pathway being affected by that causal variant is the same. As one example of this, a new class of antiretroviral drugs – the HIV entry inhibitors – was developed following the discovery of the importance of the CCR5-32 polymorphism in progression to HIV-AIDS, and these drugs appear to be of universal benefit despite the fact that the CCR5-32 mutation is almost wholly absent in subSaharan Africa (Esté and Telenti, 2007). Looking ahead, large-scale resequencing technologies will help us to build up a picture of the “allelic architecture” of causal variants for common complex diseases (Wang et al., 2005). A null hypothesis would be that the allele frequency distributions of causal variants are no different from those observed for neutral genetic variants in the human genome. Much opinion has been offered on the number of variants open to detection by association studies (which are only powered to detect common variants) (Blangero, 2004; Pritchard, 2001; Pritchard and Cox, 2002; Reich and Lander, 2001; Smith and Lusis, 2002). One way in which this debate has been couched is over the degree to which causal variants have been under selection, and whether this has been large enough to alter their allele frequency and LD properties.Various methods have been proposed recently to look for signals of selection on a genome-wide scale (Kelley et al., 2006; McVean and Spencer, 2006; Sabeti et al., 2006; Sabeti et al., 2007; Voight et al., 2006; Thornton et al., 2007), but so far, while individual examples exist of what appear to be strong signals of selection around genes affecting Mendelian traits (e.g., the lactase gene and adult lactose intolerance (Bersaglieri et al., 2004; Burger et al., 2007)), a systematic treatment of this question for variants affecting common complex disorders awaits further data.
CONCLUSIONS In this chapter we have given what we hope is a concise and accessible account of the main concepts of population genomics and their importance for medical genomics. The field has been
30
CHAPTER 2
■
Concepts of Population Genomics
transformed over the past 5 or so years by the move away from a single “Human Genome” to the diverse human genomes of many thousands of individuals. The new data have allowed us to quantify with unprecedented accuracy the patterns in different human populations of allele frequency and of linkage disequilibrium throughout the genome. This has allowed us to estimate the variable rates of recombination and mutation that apply to
different part of the genome and to reaffirm the remarkably recent and unified origins of our species in Africa. Most importantly, it has provided us with the necessary tools to carry out association-based genome-wide scans to look for novel causal variant of common complex disease, and the necessary framework for interpreting these in the translation of this knowledge into new effective therapies.
REFERENCES Amos, W. and Manica, A. (2006). Global genetic positioning: Evidence for early human population centers in coastal habitats. Proc Nat Acad Sci USA 103, 820–824. Angius, A., Hyland, F.C., Persico, I., Pirastu, N., Woodage, T., Pirastu, M. and de la Vega, F.M. (2008). Patterns of linkage disequilibrium between SNPs in a Sardinian population isolate and the selection of markers for association studies. Hum Hered 65, 9–22. Balding, D.J. (2003). Likelihood-based inference for genetic correlation coefficients. Theor Popul Biol 63, 221–230. Balding, D.J. and Nichols, R.A. (1995). A method for quantifying differentiation between populations at multi-allelic loci and its implications for investigating identity and paternity. Genetica 96, 3–12. Barbujani, G. and Belle, E.M. (2006). Genomic boundaries between human populations. Hum Hered 61, 15–21. Barnholtz-Sloan, J.S., Chakraborty, R., Sellers, T.A. and Schwartz, A.G. (2005). Examining population stratification via individual ancestry estimates versus self-reported race. Cancer Epidemiol Biomarkers Prev 14, 1545–1551. Bersaglieri, T., Sabeti, P.C., Patterson, N.,Vanderploeg, T., Schaffner, S.F., Drake, J.A., Rhodes, M., Reich, D.E. and Hirschhorn, J.N. (2004). Genetic signatures of strong recent positive selection at the lactase gene. Am J Hum Genet 74, 1111–1120. Blangero, J. (2004). Localization and identification of human quantitative trait loci: King harvest has surely come. Curr Opin Genet Dev 14, 233–240. Burger, J., Kirchner, M., Bramanti, B., Haak, W. and Thomas, M.G. (2007). Absence of the lactase-persistence-associated allele in early Neolithic Europeans. Proc Nat Acad Sci USA 104, 3736–3741. Burnett, M.S., Strain, K.J., Lesnick, T.G., de Andrade, M., Rocca, W.A. and Maraganore, D.M. (2006). Reliability of self-reported ancestry among siblings: Implications for genetic association studies. Am J Epidemiol 163, 486–492. Campbell, C.D., Ogburn, E.L., Lunetta, K.L., Lyon, H.N., Freedman, M.L., Groop, L.C., Altshuler, D., Ardlie, K.G. and Hirschhorn, J.N. (2005). Demonstrating stratification in a European American population. Nat Genet 37, 868–872. Caramelli, D., Vernesi, C., Sanna, S., Sampietro, L., Lari, M., Castrì, L., Vona, G., Floris, R., Francalacci, P., Tykot, R. et al. (2007). Genetic variation in prehistoric Sardinia. Hum Genet 122, 327–336. Carlson, C.S., Eberle, M.A., Rieder, M.J., Smith, J.D., Kruglyak, L. and Nickerson, D.A. (2005). Additional SNPs and linkagedisequilibrium analyses are necessary for whole-genome association studies in humans. Nat Genet 33, 518–521. Cavalli-Sforza, L.L., Menozzi, P. and Piazza, A. (1994). The History and Geography of Human Genes. Princeton University Press, Princeton. Condit, C.M. (2007). How geneticists can help reporters to get their story right. Nat Rev Genet 8, 815–820.
Conrad, D.F., Jakobsson, M., Coop, G., Wen, X., Wall, J.D., Rosenberg, N.A. and Pritchard, J.K. (2006). A worldwide survey of haplotype variation and linkage disequilibrium in the human genome. Nat Genet 38, 1251–1260. Crow, J. (1986). Basic Concepts in Population, Quantitative and Evolutionary Genetics.W.H.Freeman & Co Ltd. Currat, M. and Excoffier, L. (2004). Modern humans did not admix with Neanderthals during their range expansion into Europe. PLoS Biol 2, e421. de Bakker, P.I., Burtt, N.P., Graham, R.R., Guiducci, C., Yelensky, R., Drake, J.A., Bersaglieri, T., Penney, K.L., Butler, J., Young, S. et al. (2006). Transferability of tag SNPs in genetic association studies in multiple populations. Nat Genet 38, 1298–1303. de Bakker, P.I., Yelensky, R., Pe’er, I., Gabriel, S.B., Daly, M.J. and Altshuler, D. (2005). Efficiency and power in genetic association studies. Nat Genet 37, 1217–1223. Devlin, B., Bacanu, S.A. and Roeder, K. (2004). Genomic control to the extreme. Nat Genet 36, 1129–1130. Devlin, B. and Risch, N. (1995). A comparison of linkage disequilibrium measures for fine-scale mapping. Genomics 29, 311–322. Devlin, B. and Roeder, K. (1999). Genomic control for association studies. Biometrics 55, 997–1004. Donnelly, P. and Tavaré, S. (1995). Coalescents and genealogical structure under neutrality. Annu Rev Genet 29, 401–421. Esté, J.A. and Telenti, A. (2007). HIV entry inhibitors. Lancet 370, 81–88. Excoffier, L. (2001). Analysis of population subdivision. In Handbook of Statistical Genetics (D.J. Balding, M. Bishop and C. Cannings, eds), Wiley, Chichester, UK, p. 309. Frayling, T.M. (2007). Genome-wide association studies provide new insights into type 2 diabetes aetiology. Nat Rev Genet 8, 657–662. Garrigan, D. and Hammer, M.F. (2006). Reconstructing human origins in the genomic era. Nat Rev Genet 7, 669–680. Gillespie, J. (2004). Population Genetics: A Concise Guide. John Hopkins University Press. Goldstein, D.B. (2008). Jacob’s Legacy: A Genetic View of Jewish History. Yale University Press, New Haven & London. Goldstein, D.B. and Chikhi, L. (2002). Human migrations and population structure: What we know and why it matters. Annu Rev Genom Hum Genet 3, 129–152. Goldstein, D.B. and Weale, M.E. (2001). Population genomics: Linkage disequilibrium holds the key. Curr Biol 11, R576–R579. Graur, D. and Li, W.-H. (1999). Fundamentals of Molecular Evolution. Sinauer Associates Inc. Haiman, C.A., Patterson, N., Freedman, M.L., Myers, S.R., Pike, M.C., Waliszewska, A., Neubauer, J., Tandon, A., Schirmer, C., McDonald, G.J. et al. (2007). Multiple regions within 8q24 independently affect risk for prostate cancer. Nat Genet 39, 638–644. Halliburton, R. (2003). Introduction to Population Genetics. Addison Wesley.
References
Handley, L.J., Manica, A., Goudet, J. and Balloux, F. (2007). Going the distance: Human population genetics in a clinal world. Trends Genet 23, 432–439. Harding, R.M. and McVean, G. (2004). A structured ancestral population for the evolution of modern humans. Curr Opin Genet Dev 14, 667–674. Hartl, D.L. and Clark, A.G. (2007). Principles of Population Genetics. Sinauer Associates Inc. Hedrick, P.W. (2004). Genetics of Populations. Jones & Bartlett Publishing Co. Hein, J., Schierup, M.H. and Wiuf, C. (2005). Gene Genealogies, Variation and Evolution: A Primer in Coalescent Theory. Oxford University Press. Hudson, R.R. (2001). Linkage disequilibrium and recombination. In Handbook of Statistical Genetics (D.J. Balding, M. Bishop and C. Cannings, eds),Wiley, Chichester, UK, p. 309. Ioannidis, J.P., Ntzani, E.E. and Trikalinos, T.A. (2004). “Racial” differences in genetic effects for complex diseases. Nat Genet 36, 1312–1318. Jobling, M.A., Tyler-Smith, C. and Hurles, M. (2003). Human Evolutionary Genetics: Origins, Peoples and Disease. Garland Publishing Inc. Jorde, L.B., Watkins,W.S. and Bamshad, M.J. (2001). Population genomics: A bridge from evolutionary history to genetic medicine. Hum Mol Genet 10, 2199–2207. Kelley, J.L., Madeoy, J., Calhoun, J.C., Swanson, W. and Akey, J.M. (2006). Genomic signatures of positive selection in humans and the limits of outlier approaches. Genome Res 16, 980–989. Kingman, J.F.C. (1982). On the genealogy of large populations. J Appl Probab 19A, 27–43. Kruglyak, L. (1999). Prospects for whole-genome linkage disequilibrium mapping of common disease genes. Nat Genet 22, 139–144. Liu, F., Arias-Vásquez, A., Sleegers, K., Aulchenko, Y.S., Kayser, M., Sanchez-Juan, P., Feng, B.J., Bertoli-Avella, A.M., van Swieten, J., Axenovich, T.I. et al. (2007). A genomewide screen for late-onset Alzheimer disease in a genetically isolated Dutch population. Am J Hum Genet 81, 17–31. Liu, H., Prugnolle, F., Manica, A. and Balloux, F. (2006a). A geographically explicit genetic model of worldwide human-settlement history. Am J Hum Genet 79, 230–237. Liu, X.Q., Paterson, A.D., John, E.M. and Knight, J.A. (2006b). The role of self-defined race/ethnicity in population structure control. Ann Hum Genet 70, 496–505. Locke, D.P. (2006). Linkage disequilibrium and heritability of copynumber polymorphisms within duplicated regions of the human genome. Am J Hum Genet 79, 275–290. Lynch, J. and Condit, C.M. (2006). Genes and race in the news: A test of competing theories of news coverage. Am J Health Behav 30, 125–135. Macaulay, V., Hill, C., Achilli, A., Rengo, C., Clarke, D., Meehan, W., Blackburn, J., Semino, O., Scozzari, R., Cruciani, F. et al. (2005). Single, rapid coastal settlement of Asia revealed by analysis of complete mitochondrial genomes. Science 308, 1034–1036. Manica, A., Amos, W., Balloux, F. and Hanihara, T. (2007). The effect of ancient population bottlenecks on human phenotypic variation. Nature 448, 346–349. McVean, G. and Spencer, C.C. (2006). Scanning the human genome for signals of selection. Curr Opin Genet Dev 16, 624–629. Mitchell-Olds, T., Willis, J.H. and Goldstein, D.B. (2007). Which evolutionary processes influence natural genetic variation for phenotypic traits?. Nat Rev Genet 8, 845–856.
■
31
Mueller, J.C. (2004). Linkage disequilibrium for different scales and applications. Brief Bioinform 5, 355–364. Myers, S., Bottolo, L., Freeman, C., McVean, G. and Donnelly, P. (2005). A fine-scale map of recombination rates and hotspots across the human genome. Science 310, 321–324. Nei, M. (1987). Molecular Evolutionary Genetics. Columbia University Press, New York. Noonan, J.P., Coop, G., Kudaravalli, S., Smith, D., Krause, J., Alessi, J., Chen, F., Platt, D., Pääbo, S., Pritchard, J.K. et al. (2006). Sequencing and analysis of Neanderthal genomic DNA. Science 314, 1113–1118. Nordborg, M. (2001). Coalescent theory. In Handbook of Statistical Genetics (D.J. Balding, M. Bishop and C. Cannings, eds), Wiley, Chichester, UK, p. 309. Nordborg, M. and Tavaré, S. (2002). Linkage disequilibrium: What history has to tell us. Trends Genet 18, 83–90. Parra, E.J., Marcini, A., Akey, J., Martinson, J., Batzer, M.A., Cooper, R., Forrester, T., Allison, D.B., Deka, R., Ferrell, R.E. et al. (1998). Estimating African American admixture proportions by use of population-specific alleles. Am J. Hum Genet 63, 1839–1851. Patterson, N., Price, A.L. and Reich, D. (2006). Population structure and eigenanalysis. PLoS Genet 2, e190. Price, A.L., Patterson, N.J., Plenge, R.M., Weinblatt, M.E., Shadick, N. A. and Reich, D. (2006). Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet 38, 904–909. Price, A.L., Butler, J., Patterson, N., Capelli, C., Pascali, V.L., Scarnicci, F., Ruiz-Linares, A., Groop, L., Saetta, A.A., Korkolopoulou, P. et al. (2007). Discerning the ancestry of European Americans in genetic association studies. PLoS Genet. doi:10.1371/journal.pgen.0030236.eor. Pritchard, J.K. (2001). Are rare variants responsible for susceptibility to complex diseases?. Am J Hum Genet 69, 124–137. Pritchard, J.K. and Cox, N.J. (2002). The allelic architecture of human disease genes: Common disease-common variant ... or not?. Hum Mol Genet 11, 2417–2423. Prugnolle, F., Manica, A. and Balloux, F. (2005). Geography predicts neutral genetic diversity of human populations. Curr Biol 15, R159–R160. Ramachandran, S., Deshpande, O., Roseman, C.C., Rosenberg, N.A., Feldman, M.W. and Cavalli-Sforza, L.L. (2005). Support from the relationship of genetic and geographic distance in human populations for a serial founder effect originating in Africa. Proc Nat Acad Sci USA 102, 15942–15947. Ray, N., Currat, M., Berthier, P. and Excoffier, L. (2005). Recovering the geographic origin of early modern humans by realistic and spatially explicit simulations. Genome Res 15, 1161–1167. Redon, R., Ishikawa, S., Fitch, K.R., Feuk, L., Perry, G.H., Andrews, T.D., Fiegler, H., Shapero, M.H., Carson, A.R., Chen, W. et al. (2006). Global variation in copy number in the human genome. Nature 444, 445–454. Reich, D.E. and Lander, E.S. (2001). On the allelic spectrum of human disease. Trends Genet 17, 502–510. Rosenberg, N.A., Mahajan, S., Ramachandran, S., Zhao, C., Pritchard, J.K. and Feldman, M.W. (2005). Clines, clusters, and the effect of study design on the inference of human population structure. PLoS Genet 6, e70. Rosenberg, N.A. and Nordborg, M. (2002). Genealogical trees, coalescent theory and the analysis of genetic polymorphisms. Nat Rev Genet 3, 380–390.
32
CHAPTER 2
■
Concepts of Population Genomics
Sabeti, P.C., Schaffner, S.F., Fry, B., Lohmueller, J., Varilly, P., Shamovsky, O., Palma, A., Mikkelsen, T.S., Altshuler, D. and Lander, E.S. (2006). Positive natural selection in the human lineage. Science 312, 1614–1620. Sabeti, P.C., Varilly, P., Fry, B., Lohmueller, J., Hostetter, E., Cotsapas, C., Xie, X., Byrne, E.H., McCarroll, S.A., Gaudet, R. et al. (2007). Genome-wide detection and characterization of positive selection in human populations. Nature 449, 913–918. Sankar, P., Cho, M.K. and Mountain, J. (2007). Race and ethnicity in genetic research. Am J Med Genet A 143A, 961–970. Seldin, M.F. (2007). Admixture mapping as a tool in gene discovery. Curr Opin Genet Dev 17, 177–181. Serre, D. and Pääbo, S. (2004). Evidence for gradients of human genetic diversity within and among continents. Genome Res 14, 1679–1685. Service, S., DeYoung, J., Karayiorgou, M., Roos, J.L., Pretorious, H., Bedoya, G., Ospina, J., Ruiz-Linares, A., Macedo, A., Palha, J.A. et al. (2006). Magnitude and distribution of linkage disequilibrium in population isolates and implications for genome-wide association studies. Nat Genet 38, 556–560. Shanawani, H., Dame, L., Schwartz, D.A. and Cook-Deegan, R. (2006). Non-reporting and inconsistent reporting of race and ethnicity in articles that claim associations among genotype, outcome, and race or ethnicity. J Med Ethics 32, 724–728. Smith, D.J. and Lusis, A.J. (2002). The allelic structure of common disease. Hum Mol Genet 11, 2455–2461. Smith, M.W. and O’Brien, S.J. (2005). Mapping by admixture linkage disequilibrium: Advances, limitations and guidelines. Nat Rev Genet 6, 623–632. Steffens, M., Lamina, C., Illig, T., Bettecken, T., Vogler, R., Entz, P., Suk, E.-K., Toliat, M.R., Klopp, N., Caliebe, A. et al. (2006). SNPbased analysis of genetic substructure in the German population. Hum Hered 62, 20–29. Tang, H., Quertermous,T., Rodriguez, B., Kardia, S.L., Zhu, X., Brown, A., Pankow, J.S., Province, M.A., Hunt, S.C., Boerwinkle, E. et al. (2005). Genetic structure, self-identified race/ethnicity, and confounding in case-control association studies. Am J Hum Genet 76, 268–275.
The GAIN Collaborative Research Group (2007). New models of collaboration in genomewide association studies: The Genetic Association Information Network. Nat Genet 39, 1045–1051. The International HapMap Consortium (2005). A haplotype map of the human genome. Nature 437, 1299–1320. The International HapMap Consortium (2007). A second generation human haplotype map of over 3.1 million SNPs. Nature 449, 851–862. The International SNP Map Working Group (2001). A map of human genome sequence variation containing 1.42 million single nucleotide polymorphisms. Nature 409, 928–933. Thornton, K.R., Jensen, J.D., Becquet, C. and Andolfatto, P. (2007). Progress and prospects in mapping recent selection in the genome. Heredity 98, 340–348. Trinkaus, E. (2005). Early modern humans. Annu Rev Anthropol 34, 207–230. Trinkaus, E. (2007). European early modern humans and the fate of the Neandertals. Proc Nat Acad Sci USA 104, 7367–7372. Vernesi, C., Caramelli, D., Dupanloup, I., Bertorelle, G., Lari, M., Cappellini, E., Moggi-Cecchi, J., Chiarelli, B., Castrì, L., Casoli, A. et al. (2004). The Etruscans: A population-genetic study. Am J Hum Genet 74, 694–704. Voight, B.F., Kudaravalli, S., Wen, X. and Pritchard, J.K. (2006). A map of recent positive selection in the human genome. PLoS Biol 4, e72. Wakeley, J. (2007). Coalescent Theory: An Introduction. Roberts & Co. Wang, W.Y., Barratt, B.J., Clayton, D.G. and Todd, J.A. (2005). Genomewide association studies: Theoretical and practical concerns. Nat Rev Genet 6, 109–118. Weale, M.E., Depondt, C., Macdonald, S.J., Smith, A., Lai, P.S., Shorvon, S.D., Wood, N.W. and Goldstein, D.B. (2003). Selection and evaluation of tagging SNPs in the neuronal-sodium-channel gene SCN1A: Implications for linkage-disequilibrium gene mapping. Am J Hum Genet 73, 551–565. Wilson, J.F., Weale, M.E., Smith, A.C., Gratrix, F., Fletcher, B., Thomas, M.G., Bradman, N. and Goldstein, D.B. (2001). Population genetic structure of variable drug response. Nat Genet 29, 265–269.
RECOMMENDED RESOURCES The seminal work by Cavalli-Sforza et al. (1994) is still a standard reference on the use of genetics to infer aspects of human history and demography. A useful, more up-to-date work with an emphasis on human evolution is the textbook by Jobling et al. (2003).
The HapMap website (http://www.hapmap.org/tutorials.html.en) has some nice tutorial material on the rationale and use of this important population genomic resource.
CHAPTER
3 Genomic Approaches to Complex Disease Desmond J. Smith and Aldons J. Lusis
INTRODUCTION Most of the socioeconomic burden of disease in the industrialized nations is due to complex disorders, in which multiple genes interact with the environment to produce the final disease phenotype (Smith and Lusis, 2002). Typical examples include atherosclerosis, diabetes, cancer, multiple sclerosis, autism, alcoholism, and drug abuse (Diabetes Genetics Initiative et al., 2007; Easton et al., 2007; Moffatt et al., 2007; Samani et al., 2007; Scott et al., 2007; Steinthorsdottir et al., 2007; Wellcome Trust Case Control Consortium, 2007). An understanding of complex disease involves: (1) identifying the genetic variations, whether common or rare, that contribute to the disease; (2) determining how the variations influence function and which pathways they perturb; and (3) determining how the variations interact with each other and with the environment to contribute to the disease. Even if the total genetic contribution to a complex trait is substantial, typically each gene contributes only a small amount to the final phenotype, making detection of the responsible genes an arduous process (Wellcome Trust Case Control Consortium, 2007). Further complicating such analyses are allelic heterogeneity and locus heterogeneity, in which a disease can be due to different alleles of the same gene or to alleles in different genes, respectively. Genetic interactions can also occur, in which multiple genes enhance each others’ effects on a disease (synergy) or
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
block each others’ effects (epistasis). In addition, there is much variation of the environment in human studies, further contributing to “noise” in an analysis (Smith and Lusis, 2002). The advent of genomics, in which the whole genome is the unit of study rather than individual genes, is revolutionizing the study of complex disease (Olson, 2003) and is rapidly helping establish genomic medicine as a new and exciting field. This chapter focuses on the use of genomic technologies to understand complex disease and is organized as follows: First, we discuss methods for the identification of DNA variations, both common and rare (Figure 3.1). Second, we discuss linkage and association methods for relating DNA variation to phenotypes. And lastly, we discuss how the “omic” technologies can help identify the underlying genes and organize them into pathways and networks (Figure 3.2).
IDENTIFYING COMMON AND RARE GENOMIC VARIATIONS IN THE POPULATION Varieties of Polymorphism Mapping genes became practicable in humans with the realization that polymorphic DNA sequences could be used directly as genetic markers (Botstein et al., 1980). Initially, single nucleotide polymorphisms (SNPs), assayed as restriction fragment length polymorphisms (RFLPs) on gels, were employed as markers in
Copyright © 2009, Elsevier Inc. All rights reserved. 33
34
CHAPTER 3
■
Genomic Approaches to Complex Disease
ELUCIDATING COMPLEX DISEASE Phenotype of interest Identify variants Linkage or association Mixed or isolated populations Common Allele Detection Technology (e.g., TaqMan, MS, Affymetrix, Illumina) Rare Allele Discovery Technology (e.g., 454 Life science, polony sequencing, Solexa, MPSS) Gene/allele discovery interaction variants
Proteomics
Systems biology
Genetical genomics
Fxns/pathways Animal models
Molecular subtyping
Transcript profiling
Figure 3.1 Elucidating complex disease. This flow chart parallels the discussion in the text. Genetic dissection of a disease phenotype of interest requires identification of appropriate variants for mapping. These variants will usually be common and chosen from publicly available databases. The strategy for gene identification will require decisions as to whether linkage or association should be used, which type of population should be employed, and the nature of the allele detection technology. Some phenotypes may require the use of recent approaches to rare allele identification. Once a gene responsible for a phenotype has been identified, the methods of systems biology can be used to understand the relation of the causative allele to disease.
combination with Southern blotting. However, this approach is laborious. The strategy became much more feasible with the discovery of microsatellite repeats (Weber and May, 1989). These densely spaced markers (roughly one per every 2 kb) are polymorphic because of differences in lengths of short oligonucleotide repeats. They have the advantage of being easily scored using the polymerase chain reaction (PCR) and gel electrophoresis and also have multiple alleles, which greatly improves their informativeness compared to biallelic markers such as SNPs. The use of microsatellites dominated mammalian genetics until the last several years, at which time the utility of SNPs reemerged with the invention of medium- and high-throughput approaches for analyzing these digital allelotypes. The SNPs also have the advantage that their average spacing is about one every 300 bp in the human genome, denser than microsatellites. SNP Genotyping Technologies A large number of SNP genotyping technologies have emerged over the last few years (Table 3.1). One of the first rapid methods for SNP genotyping, TaqMan technology, uses real-time PCR (Gibson et al., 1996; Heid et al., 1996; Livak, 1999). The SNPs are distinguished by reporter oligonucleotides containing a fluorescent nucleotide at the 5 end and a quenching nucleotide at the 3 end. If the oligonucleotide is a perfect match to a PCR product flanking
the SNP, the Taq polymerase used for the PCR liberates the 5 fluor through the 5 to 3 exonuclease activity of the polymerase. The liberated fluor-substituted nucleotide is no longer inhibited by proximity to the quenching fluor and fluoresces brightly. This fluorescence increases with each cycle of the PCR and is measured in each cycle that is in real time. However, if the oligonucleotide does not bind because of the presence of an SNP, there is no increase in fluorescence. Because a PCR assay has to be set up for each SNP, the method is only of moderate throughput. Another moderate- to high-throughput technology for genotyping biallelic markers employs matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS) (Cantor, 2005; Griffin and Smith, 2000). Primer extension reactions are performed resulting in differing product lengths depending on the allele present for a particular SNP. The mass of the reaction products are then quantitated using the MALDI-TOF MS, permitting evaluation of which SNP is present. Throughput can be enhanced by multiplexing different assays in one sample. A dramatic increase in throughput came with the invention of photolithographically synthesized arrays of oligonucleotides for genotyping (Kreiner and Buck, 2005; Lipshutz et al., 1995). Affymetrix arrays consist of very high densities of 25-mers matching major or minor alleles of known SNPs in the human genome. Photolithographic activation of chemical groups is used to synthesize the oligonucleotides in situ. The photolithography permits a dramatic degree of spatial miniaturization with very dense packing of oligonucleotides and high-throughput analysis of SNPs. The presence or absence of a particular SNP in the target genome of interest is revealed by the presence or absence of a corresponding hybridization signal on the array. The arrays are spatially encoded, meaning that the position of an oligonucleotide on the array maps to the SNP it assays. The current feature size on these arrays permits up to 500,000 SNPs to be analyzed. Perlegen, a company that employs Affymetrix technology, has used custom-designed photolithographic arrays to query 1.5 106 SNPs (Hinds et al., 2005). Illumina arrays use sequence-encoding rather than spatialencoding for high-throughput genotyping. Beads with assay oligonucleotides are used to interrogate individual SNPs. A fluorescent extension reaction requiring correct hybridization of the assay oligonucleotide to genomic DNA detects which SNP is present (Shen et al., 2005).The small size of the beads means that hundreds of thousands of SNPs can be analyzed in parallel. Fluorescence readouts from each bead are captured using bundles of optical fibers. In addition to the assay oligonucleotide, the beads also contain decoding oligonucleotides that reveal which SNP the bead is interrogating. The decoding oligonucleotides are revealed using a series of fluorescently labeled probes. This bead-based method has been used to date to analyze up to a million SNPs in one assay. Sequencing Technologies The above genotyping technologies are high throughput but only detect already known SNPs. Most of these SNPs have been identified as a result of resequencing projects in the public domain (Wheeler et al., 2006) and represent reasonably common
Identifying Common and Rare Genomic Variations in the Population
■
35
OMIC SPACE
PC 2 Phenome PC 1
Metabolome
Proteome
Transcriptome
Genome
Figure 3.2 Omic space. Multiple partially orthogonal domains, where various sorts of biological networks, experimental findings, and model-based predictions are represented and integrated (Left). Tools for collection of genome-wide data relevant to omic space are being developed. The phenome (global phenotyping) represents much of classical biology and medicine, supplemented with exciting advances in fields such as molecular imaging (MRS, PET). Repositories for such data are in the early stages. One example is the Mouse Phenome database (http://www.jax.org/phenome). Mass spectrometry (MS) and NMR have been used for high-throughput analysis of the metabolome and for non-invasive diagnosis and screening using non-supervised algorithms such as PCA. An example of PCA used to differentiate disease and non-diseased samples is shown to the right, where PC1 and PC2 represent the first and second principal components, and red and blue points represent metabolites differentiating diseased and non-diseased samples, respectively. The proteome is studied using a number of technologies, including MS. The interactome (study of interactions between proteins) has been investigated using whole-genome yeast two-hybrid screens, literature-based analysis of pathways (e.g., gene ontology, http:// www.geneontology.org/), and co-affinity immunopurification. Protein products of genes mutated in inherited genetic disorders tend to interact with proteins involved in similar disorders (Right). Green, proteins involved in developmental disorders; blue, proteins involved in endocrine disorders. Whole-genome expression arrays have been used for transcriptome analysis, while high-density SNP chips, high-throughput sequencing, and HapMap sequences have been used at the genome level. Some databases, such as the “GenomePhenome Superbrain” (http://www.omicspace.riken.jp/), are designed to integrate data from the various omic domains. The elements (nodes) of each stage appear to have a scale-free topology (right, shown linked to transcriptome). Certain elements (hubs) have many connections while most have relatively few (see also Chapters 1 and 6).
alleles. However, the extent to which complex diseases are due to common or uncommon polymorphisms is unclear (see Common Disease/Common Variant Hypothesis below) (Smith and Lusis, 2002).
In the long term, the evaluation of rare SNPs in common complex disorders will require new sequencing technologies to affordably resequence multiple human genomes. This quest is known as “the $1000 genome” (Shendure et al., 2004). While
36
CHAPTER 3
TABLE 3.1
■
Genomic Approaches to Complex Disease
Genotyping and sequencing technologies
Technology
Applications
Organization
TaqMan
Custom SNPs, 102–103 assays
Applied Biosystems (https://www2. appliedbiosystems. com/)
Mass spectrometry Custom SNPs, 102–103 assays
Sequenom (http: //www.sequenom. com)
Photolithographic arrays
Predesigned SNPs, Affymetrix (http:// 105–106 assays www.affymetrix.com) Perlegen (http:// www.perlegen.com/)
Illumina arrays
Predesigned SNPs, Illumina (http:// 105–106 assays www.illumina.com)
Pyrosequencing
SNP discovery
454 Life Sciences (http://www.454. com)
Solexa sequencing arrays
SNP discovery
Illumina (http:// www.illumina.com)
Massively parallel signature sequencing
SNP discovery
Illumina (http:// www.illumina.com)
Single molecule DNA sequencing
SNP discovery
Helicos Biosciences (http://helicosbio. com)
certainly inspiring, it will likely be many years before this goal is achieved. Nevertheless, recent advances in rapid sequencing may improve our discovery of rare and private alleles (Table 3.1) (see Chapter 7). All of the developments increase speed by avoiding laborious molecular cloning steps in Escherichia coli. In one method developed by 454 Life Sciences, individual genomic DNA molecules are cloned on beads (Margulies et al., 2005). The beads are isolated from one another using oil/water emulsions to create spherical lipid bilayer vesicles about the size of a microorganism. Polymerase-mediated amplification of the DNA molecules on the beads can then occur without cross-contamination (emulsion PCR). Sequence information from each bead is then obtained using pyrosequencing. In this method, release of a pyrophosphate during incorporation of a nucleotide in an extension reaction is detected by light emission from a luciferase reaction. The small size of the beads allows large numbers of sequences to be obtained in parallel in picoliter reactors. Acquisition of the pyrosequencing output uses fiber optics. The throughput is about 25 Mb in an afternoon, although the average sequence reads are less than 100 bp. Polony sequencing is another approach using emulsion PCR, in which the individual beads are immobilized in a thin film of polyacrylamide after DNA amplification (Shendure et al.,
2005). The term “polony” is an abbreviation of “polymerase colony”. The DNA molecules represent self-ligated molecules which permit the acquisition of mate-paired sequence tags, that is, sequences from the two ends of the ligated molecule. Fluorescent sequencing by ligation in combination with a light microscope allows sequence tags of about 13 bp to be obtained from each end of the self-ligated molecule, a total of 26 bp. This method appears to be particularly useful for detecting translocations and inversions. Another approach, used by Solexa, employs PCR amplification of individual DNA molecules (Bennett et al., 2005). The individual PCR products remain localized to a small region of a glass slide because of the presence of capture oligonucleotides. The amplification results in “colonies” of different individual DNA molecule at a high spatial density on the glass slide. Fluorescent dideoxy sequencing reactions are then employed to interrogate each of the colonies in parallel, giving high-throughput sequences. Another bead-based sequencing method is massively parallel signature sequencing, MPSS (Brenner et al., 2000). A population of DNA molecules is ligated en masse to a highly complex mixture of identifying oligonucleotides. Amplification by PCR follows. Purification of multiple copies of a single class of DNA occurs when individual identifying oligonucleotides are captured by their complement on a bead. Sequence from each bead is then obtained by sequential hybridization reactions. The method gives sequence ladders of about 20 bp. A fourth method does not employ any type of DNA amplification, but rather uses fluorescent dideoxy sequencing of single DNA molecules (Braslavsky et al., 2003). The data capture uses a light microscope and the potential degree of parallelism is extremely high. Sequence reads of up to 5 bp have been obtained. At present, there is tremendous activity in developing cheaper, faster, and more accurate sequencing methods. Whether the next generation of sequencing tools uses one of the strategies described above or some completely unexpected technology, it seems clear that the contribution of rare and private polymorphisms to common genetic disease will become better understood in the coming years.
RELATING DNA VARIATION TO PHENOTYPES Candidate Gene Approach In the early days of complex trait genetics, the mid-1980s, the candidate gene approach was widely used. This approach evaluates whether common variations of biochemically defined candidate genes are associated with a particular disease by evaluating the frequency of various alleles in case versus control populations (Table 3.2). One of the first important examples was the gene for apolipoprotein E (APOE) which was found to exhibit three common alleles (E2, E3, and E4). The E2 form was originally found to be strongly associated with a relatively rare
Relating DNA Variation to Phenotypes
TABLE 3.2
■
37
Examples of sequence variants implicated in complex disease
Disease phenotype
Gene/sequence varianta
Identification strategy
Technology
References
Hyperlipidemia (type III)
ApoE2
Association/candidate gene
Isoelectric focusing
Wardell et al., 1982
Alzheimer’s disease
ApoE4
Association/candidate gene
PCR/restriction isotyping
Strittmatter et al., 1993
Diabetes mellitus (type II)
Calpain 10
Linkage/association
Sequencing, PCR-RFLP
Horikawa et al., 2000
Crohn’s disease
NOD2
Linkage/LD, Linkage, TDT
Microsatellites/ sequencing, allelespecific PCR
Hugot et al., 2001; Ogura et al., 2001
Familial combined hyperlipidemia
USF1
Linkage/isolated population/association
Sequencing
Pajukanta et al., 2004
Myocardial infarct risk
Lymphotoxin alpha
Association/LD
Sequencing/PCR isotyping
Ozaki et al., 2002
Macular degeneration
Complement factor H
Association
Affymetrix arrays
Klein et al., 2005
Low HDL cholesterol
ABCA1, APOA1, LCAT
Association
Sequencing
Cohen et al., 2004
Low LDL cholesterol
NPC1L1
Association
Sequencing/TaqMan
Cohen et al., 2006
Obesity
SLC6A14
Linkage/LD/isolated population/association
Microsatellites/ sequencing/MS
Suviolahti et al., 2003
Diabetes mellitus (type II)
TCF7L2
Linkage/isolated population/association
Microsatellites
Grant et al., 2006
Myocardial infarct risk
LTA4H
Isolated population/association
Sequencing
Helgadottir et al., 2006
Schizophrenia
NRG1
Linkage/isolated population/TDT
Fluorescence polarization
Stefansson et al., 2002
Crohn’s disease
5q31 cytokine gene cluster
Linkage/LD
Microsatellites/ sequencing
Rioux et al., 2001
Breast cancer
FGFR2, TNRC9/ LOC643714, MAP3K1, rs13281615 (8q), LSP1
Association
Affymetrix arrays/TaqMan/MS
Easton et al., 2007
Diabetes mellitus (type II)
TCF7L2, CDKAL1
Association
Illumina
Steinthorsdottir et al., 2007
Diabetes mellitus (type II)
9p21.3 (CDKN2A/ CDKN2B), IGF2BP2, CDKAL1, HHEX, SLC30A8
Association
Affymetrix
Diabetes Genetics Initiative et al., 2007; Zeggini et al., 2007
Diabetes mellitus (type II)
9p21.3 (CDKN2A/ CDKN2B), IGF2BP2, CDKAL1, HHEX, SLC30A8, TCF7L2, FTO PPARG, KCNJ11
Association/isolated population
Illumina
Scott et al., 2007
Myocardial infarct
9p21.3 (CDKN2A/ CDKN2B), MTHFD1L, PSRC1, MIA3, SMAD3, LPL
Association
Affymetrix
Samani et al., 2007
(continued )
38
CHAPTER 3
TABLE 3.2
■
Genomic Approaches to Complex Disease
(Continued)
Disease phenotype
Gene/sequence varianta
Identification strategy
Technology
References
Asthma
ORMDL3
Association
Illumina
Moffatt et al., 2007
Bipolar disease
16p12, KCNC2, GABRB1, GRM7, SYN3
Association
Affymetrix
Wellcome Trust Case Control Consortium, 2007
Myocardial infarct
9p21.3 (CDKN2A/ CDKN2B), MTHFD1L, ADAMTS17
Association
Affymetrix
Wellcome Trust Case Control Consortium, 2007
Crohn’s disease
NOD2, 5q31 cytokine gene cluster, IL23R, ATG16L1, 10q21, 5p13.1, IRGM, NKX2-3, PTPN2
Association
Affymetrix
Wellcome Trust Case Control Consortium, 2007
Rheumatoid arthritis
HLA-DRB1, PTPN22, IL2RA, IL2RB, TNFAIP2, GZMB, PRKCQ, KAZALD1, 12q24
Association
Affymetrix
Wellcome Trust Case Control Consortium, 2007
Diabetes mellitus (type I)
MHC, CTLA-4, PTPN22, IL2RA, IFIH1, 12q13, 12q24, 16p13, PTPN2, IL2 receptor
Association
Affymetrix
Wellcome Trust Case Control Consortium, 2007
Diabetes mellitus (type II)
PPARG, KCNJ11, TCF7L2, FTO, CDKAL1, HHEX,
Association
Affymetrix
Wellcome Trust Case Control Consortium, 2007
a
Where the responsible gene is unclear, the relevant chromosome region is given.
hyperlipidemia (Type III) and with low cholesterol levels in the population (Wardell et al., 1982). Later, the E4 allele was found to be strikingly associated with Alzheimer’s disease (Strittmatter et al., 1993). The candidate gene approach continues to be employed extensively, as thousands of association studies are reported each year. However, the strategy has some important problems. First, of course, the gene has to first be linked to the disease through biochemical studies or mechanistic hypotheses. Second, association tests are prone to artifacts due to population stratification and ethnic admixture. One way to overcome this latter problem is to perform family-based association testing such as the transmission disequilibrium test (TDT) (Spielman and Ewens, 1998). This test evaluates the role of a polymorphism in a disorder by assessing its frequency in affected versus non-affected siblings. The idea is that, since the siblings are raised in the same family, most of the differences in susceptibility will be due to genetics rather than to the environment. Linkage In the early 1990s, linkage analysis employing microsatellite markers became a common approach to map loci for complex diseases. In this approach, families with multiple affected members are phenotyped and genotyped. Regions of the genome relevant to the disease can then be detected using allele association or parametric approaches. One advantage of this approach is that a simplified
set of genes may be responsible for the disorder within a particular family compared to the population as a whole. However, this advantage melts away when multiple families are required, as is nearly always the case to obtain adequate statistical power. Another substantial disadvantage of linkage analysis in humans is the cost and difficulty of identifying multiple multigenerational families with the disorder of interest. In addition, mapping power in this design is limited and the responsible genes are frequently poorly localized to regions covering many megabases. A common paradigm since the late 1990s has been to carry out linkage analysis followed by targeted association testing of candidate genes in promising regions or to test for linkage disequilibrium (LD) with genetic markers in the region. Linkage disequilibrium is a situation in which neighboring alleles are linked more often than expected by chance, either because of low recombination rates between the markers or because of expansion of a founder population in which the alleles were originally linked. The first successful example of the twostep strategy was the identification of the calpain 10 gene as a causal gene in type 2 diabetes in a large set of Mexican families, although this finding remains somewhat controversial (Horikawa et al., 2000). A very clear subsequent success was the identification of the NOD2 gene as an important contributor to Crohn’s disease (Hugot et al., 2001; Ogura et al., 2001). Another example is the identification of the USF1 gene as a cause of familial combined hyperlipidemia (Pajukanta et al., 2004).
Relating DNA Variation to Phenotypes
Genome-wide Association Although the combination of linkage and association has been successful in several instances, only a handful of genes have been identified to date. However, with the advent of cheaper methods for the detection of polymorphisms, genome-wide association studies have become feasible (Hirschhorn, 2005; Hirschhorn and Daly, 2005) (see Chapter 8). Thus, association, the strategy first used for genetic dissection of human traits, has come back to center stage. The first example was a study of heart disease by Ozaki et al. (2002), who typed thousands of SNPs in thousands of individuals in Japan (Ozaki et al., 2002). The several genes exhibiting strong evidence of association with heart disease were then typed in a second set of individuals, and one gene, encoding lymphotoxin-, was found to exhibit a consistent association. Another very striking success was the identification of complement factor H in age-related macular degeneration (Klein et al., 2005). In that study, a relatively small group of individuals (96 cases, 50 controls) were typed for 116,204 SNPs, and a common non-synonymous SNP was found to be strongly associated with the disease even after a conservative Bonferroni correction for multiple comparisons. Such whole-genome association studies are rapidly becoming the major source of information on complex disorders in humans as the available technologies become more widely used and are allowing interrogation of 105–106 SNPs. Very recent successful examples include the identification of loci for breast cancer (Easton et al., 2007), type II diabetes (Diabetes Genetics Initiative et al., 2007; Scott et al., 2007; Steinthorsdottir et al., 2007; Zeggini et al., 2007), coronary artery disease (Samani et al., 2007), asthma (Moffatt et al., 2007), and a large study which identified loci contributing to bipolar disorder, coronary artery disease, Crohn’s disease, rheumatoid arthritis, and type 1 and type 2 diabetes (Wellcome Trust Case Control Consortium, 2007). Most gratifying is the high degree of agreement between different studies for identified loci for the same phenotype. Also interesting was the fact that large majority of the loci exerted small effects on the final disease risk (see Chapter 8). Copy Number Variation Another recent realization emerging from the use of microarray genotyping is the prevalence of copy number variants (CNVs) which constitute about 10% of the human genome (e.g., Sebat et al., 2004) (see Chapter 9). The CNVs are type of polymorphism in which segments of DNA with an average length of 500 kb show variation in copy number either due to deletion or duplication. This source of variation is beginning to be explored with respect to common complex disorders, one recent example being autism (Sebat et al., 2007). Common Disease/Common Variant Hypothesis The success of modern genotyping technologies in analyzing common complex disorders depends partly upon the correctness of the common disease/common variant hypothesis. This hypothesis states that common genetic diseases are due
■
39
to common polymorphisms. On the face of it, this hypothesis seems plausible, and there is evidence that it is at least partially true for many common complex diseases (Hirschhorn, 2005). However, it now appears that a significant fraction of common disease will be due to many different rare mutations. In fact, this was observed in a subset of the NOD2 mutations underlying Crohn’s disease (Hugot et al., 2001). More recently, studies of common variations in plasma lipoprotein levels, involving sequencing of candidate genes in many individuals, have clearly demonstrated the importance of rare alleles (Cohen et al., 2004, 2006). Thus, efficient sequencing approaches are likely to be important in understanding at least some forms of complex traits. Although there have been tremendous recent advances in genotyping speed, the throughput is still insufficient to interrogate all possible polymorphisms over the entire genome. In addition, even modern genotyping technologies will not easily identify rearrangements such as inversions, which may be more common than previously thought (Tuzun et al., 2005). Estimates based on mutation rates suggest that nearly every nucleotide in the genome is altered in at least one individual on the planet (see Chapter 1). Isolated Populations As described in a previous section, one of the major stumbling blocks in human genetics is allelic heterogeneity. One potentially helpful approach in analyzing complex human genetic traits is to use isolated populations. These populations are thought to have an extremely small number of founder individuals, often estimated as being in the hundreds. This bottleneck means that a much smaller number of genes and alleles cause the disease phenotype, simplifying the genetic heterogeneity of many complex diseases. Useful isolated populations have been identified in the Hutterites of rural North America, as well as populations in Finland and Iceland. Some theoretical studies suggest that for common complex disorders the advantages of using an isolated population may be blunted compared to that for rare alleles (Hirschhorn and Daly, 2005). This is because common disease alleles will be proportionally represented even in a relatively small founding population. Nevertheless, the promise of isolated populations has been fulfilled in at least a few examples. These include the identification of genes involved in familial combined hyperlipidemia (Pajukanta et al., 2004) and obesity (Suviolahti et al., 2003) in Finland and genes involved in type 2 diabetes (Grant et al., 2006), myocardial infarction (Helgadottir et al., 2006), and schizophrenia in Iceland (Stefansson et al., 2002). HapMap Project Human genetic analysis is currently limited by both genotyping power and ascertainment of sufficiently large study populations for analysis. Strategies are therefore needed to maximize the information obtained from these precious datasets. It was observed that contiguous regions of the mammalian genome tend to remain linked together on the chromosome in which
40
CHAPTER 3
■
Genomic Approaches to Complex Disease
they reside (Rioux et al., 2001). These regions in LD are called haplotype blocks and their existence led to the realization that one SNP, a so-called “tag SNP”, could in principle effectively “stand in” for a larger number of adjacent SNPs. It was conjectured that using carefully chosen tag SNPs could extend the power of present genotyping technologies, thus improving coverage of the genome (see Chapter 2). These observations led to the international HapMap project, an endeavor to comprehensively identify haplotype blocks in the human genome. The main problem in decoding haplotypes is that genotype information represents the admixture of two different linked haplotypes from the parental chromosomes. The admixed “messages” have to be deconvoluted to provide haplotype information. Two distinct approaches have been taken. In the publicly funded approach, genotyping of 269 individuals from four different ethnic human groups was used to identify blocks of SNP alleles that tend appear together more often that would be expected by chance (Altshuler et al., 2005). The other approach taken by the company Perlegen used purification of individual chromosomes by somatic cell genetics, so that the haplotypes could be read out directly (Hinds et al., 2005; Patil et al., 2001). In both approaches, blocks were found to extend for varying lengths in the genome, ranging from 1 kb to 100 kb. Despite this variation, the data should allow more efficient design of genotyping assays. In addition, the uncovered haplotype data will likely be useful in understanding the history and structure of human populations throughout the world. Quantitative Trait Locus Mapping in Model Organisms Animal models, particularly strains of mice and rats, exhibit common variations in traits that are relevant to most common diseases. The mapping of loci contributing to such traits using quantitative trait locus (QTL) analysis is straightforward (Flint and Mott, 2001). At present, thousands of loci for hundreds of complex traits have been mapped, but, apart from positional candidates, only a handful of genes have been identified. The QTL studies in mice suffer from some of the same problems as linkage in humans: the loci identified are usually very large and the underlying genes usually explain only a small fraction of the variance of the trait. Flint et al. (2005) provide an excellent summary of the status of QTL mapping and of new strategies. As discussed below, genome-wide “omics” technologies are expected to greatly facilitate disease gene identification in studies with animal models.
INTEGRATION OF “OMIC” TECHNOLOGIES WITH GENETICS Systems Biology High-throughput assays for DNA sequence variation, transcript levels, protein levels, protein interactions, and metabolites, in combination with data integration methods, are beginning to
transform studies of complex disease and of biology in general. This area of research has become broadly known as systems biology, the scientific discipline that seeks to quantify all of the elements of a biological system and assess their interactions (Hood et al., 2004). An important concept is that the interaction of the parts of a system creates properties or functions that would not be expected from looking at the individual components on their own (Table 3.3) (see Chapter 6). Transcriptomic Analysis The “omic” technology that has had the largest impact on complex disease is the measurement of transcript levels using highdensity microarrays.The original arrays consisted of spotted cDNA capture probes, but the more recent technologies, including Affymetrix and Illumina, utilize synthetic oligonucleotides as capture probes (discussed in Pravenec et al., 2003) (see Chapter 13). In any segregating population, there will be hundreds or thousands of genetic variations that affect gene expression. It seems clear that these variations in expression will make important contributions to complex disease. With the development of microarray technology, it is now possible to quantify transcript levels globally and to integrate such data with information relating to disease. One approach has been to map a critical genomic region for a disease using affected and unaffected individuals and then pinpoint the responsible gene by identifying differentially expressed genes in the critical region. For example, in the rat animal model, a combination of congenic mapping and microarray analysis identified the fatty acid translocase Cd36 gene as a cause of insulin resistance and hypertriglyceridemia (Aitman et al., 1999). In the case of complex, heterogeneous disorders, no one gene would necessarily be expected to exhibit altered expression in cases versus controls, but certain pathways might be perturbed. For example, Mootha and colleagues developed an analytic strategy, termed “Gene Set Enrichment Analysis,” to detect modest but coordinate changes in the expression of functionally related genes (Mootha et al., 2003). They applied the method to muscle biopsies from diabetic versus control individuals. Although no single gene was significantly associated with diabetes, the mitochondrial and oxidative phosphorylation pathways were significantly enriched in genes that were modestly perturbed in diseased individuals. Genetical Genomics A particularly powerful way of integrating expression and clinical data is to apply genetics, an approach that has been called “genetical genomics” (Jansen and Nap, 2001). Thus, if transcript levels are measured in segregating populations, the loci controlling transcript levels can be mapped using the same approaches that have been used to map loci for complex clinical traits. The loci controlling transcript abundance have been termed “expression QTLs,” or eQTLs (Schadt et al., 2003). When an eQTL encompasses the physical location of the gene encoding that transcript, it is likely that the causative genetic variation resides
Integration of “Omic” Technologies with Genetics
TABLE 3.3
■
41
Systems biology and genomic medicine
Disease phenotype
Systems biology approach
Insights
Technology
References
Insulin resistance/ hypertriglyceridemia
Transcript analysis
Cd3 gene causative
Microarrays
Aitman et al., 1999
Diabetes mellitus (type II)
Gene set enrichment analysis
Mitochondrial/oxidative phosphorylation plays a role
Microarrays
Mootha et al., 2003
Obesity
Genetical genomics
Molecular subtyping can better define complex phenotypes
Microarrays
Schadt et al., 2003
Hypertension
Genetical genomics
Correlation of biological and expression phenotypes
Microarrays
Hubner et al., 2005
Obesity, diabetes
Genetical genomics
Correlation of biological and expression phenotypes
Microarrays
Lan et al., 2006
Embryonic viability
Interactome mapping
Hub genes more likely to be essential for survival
Yeast 2 hybrid, co-affinity purification, literature curation
Gandhi et al., 2006
Inherited disease
Interactome mapping
Gene products from similar disorders tend to interact
Yeast 2 hybrid, co-affinity purification, literature curation
Gandhi et al., 2006
Cancer
Interactome mapping
Oncogenes can be related to known signal transduction pathways
Yeast 2 hybrid, co-affinity purification, genetic crosses in C. elegans
Tewari et al., 2004
within the gene itself (i.e., the transcript is likely to be regulated in cis). Conversely, when an eQTL does not encompass the physical location of the gene, the transcript must be regulated in trans (i.e., under the control of a different gene). Thousands of such eQTL have been mapped in various crosses in mice, as well as other experimental organisms, and less detailed maps have been produced in studies of cells from human pedigrees (Cheung et al., 2005; Doss et al., 2005; Ghazalpour et al., 2005; Hubner et al., 2005; Schadt et al., 2003). One of the initial surprises in these studies has been the extent of variation at the level of transcript levels. In recent studies in mice, where experiments are greatly simplified compared to human studies, a large F2 intercross between two common laboratory strains of mice revealed 6676 significant eQTLs (p 5 105) in liver, over a thousand of which were cis-acting eQTLs (Wang et al., 2006). Similarly, large numbers of common variations are likely to exist in human populations. In the near future, one can envision a genome-wide database relating common DNA variation with common transcript level variations (in other words, a map, such as the HapMap of all common variations in transcript levels). Obviously, such a map would help in prioritizing candidate gene for loci contributing to complex diseases (i.e., genes with cis-eQTLs at the
locus of interest would constitute a restricted set of candidates). Also, it might be expected that transcript levels of the causative genes will be significantly correlated with clinical trait values, so determining Pearson or Spearman correlation coefficients between these may be helpful. Genetical genomic data could also be used in a converse manner to exclude potential candidate genes (Lan et al., 2006). There are two important caveats in considering the relationship of an eQTL and a clinical trait. First, gene expression patterns are cell- and tissue-specific, and certain inaccessible tissues may be required to understand a complex trait (although it has been observed that cis-eQTLs usually tend to affect gene expression in all tissues in which that gene is expressed). The second is that the regulation of biologic processes occurs via control of gene expression in many, but certainly not all, situations. Molecular Subtyping Another use of genomic and other “omic” technologies is to define distinct molecular subtypes for a clinical phenotype. This has been particularly useful in the case of cancers (Bucca et al., 2004), but it should be applicable to most complex disorders. For example, in one study examining liver expression array data
42
CHAPTER 3
■
Genomic Approaches to Complex Disease
in an intercross, distinct subgroups of mice were identified with respect to body fat (Schadt et al., 2003). The array data were consistent with the idea that the classically defined trait of fat pad mass actually consists of at least two molecularly distinct subgroups. Analyzing these two subgroups separately gave much greater sensitivity in detecting loci for obesity. While such an approach has yet to be employed in human complex traits, there is a strong interest in using high-dimensional data methods, such as microarrays and mass spectrometry, to identify biomarkers for disease (Bailey and Ulrich, 2004). It is conceivable that some of these investigations may detect different subclasses of disease that would usually be lumped into one group using classical methods. Separate genetic analysis of these subclasses may well give greater sensitivity to detect the genes responsible for complex human conditions. A particularly powerful application of genetical genomics is to identify genes that are causally related to clinical traits. For example, in a recent study of a cross between two strains of mice differing in adiposity, Schadt and colleagues used liver expression array data to identify genes likely to be causally involved in visceral fat levels (Schadt et al., 2005). It was reasoned that genes controlling adiposity would likely also control expression of genes determining fat levels. However, correlation of gene expression levels with adiposity could represent an effect rather than a cause of obesity. Using mathematical conditioning, it was possible to distinguish between these possibilities. Some of the top candidates were then examined using transgenic mice either overexpressing or underexpressing the causal genes, and it was observed that a large fraction of those genes did indeed influence adiposity. Proteomics The above approaches were made possible by the high-throughput analysis of the transcriptome. The analysis of the full proteome has not yet been achieved, but rapid progress is being made (Domon and Aebersold, 2006). It is clear that the number of distinct proteins in mammals will certainly greatly exceed the number of genes, and proteomes will exhibit large and dynamic complexity. While proteomic and systems approaches have had only a limited role in understanding complex disease to date, this situation is bound to change as the depth and reliability of the databases continues to improve. Interactome The connections between gene products can be mapped using genome-scale surveys of protein-binding partners. The universe of protein–protein interactions is known as the interactome. Two widely used technologies for charting this space are yeast twohybrid screening and co-affinity purification combined with detection by MS (Cusick et al., 2005). Yeast (Gavin et al., 2002; Ho et al., 2002; Ito et al., 2001; Uetz et al., 2000), worms (Li et al., 2004), flies (Formstecher et al., 2005; Giot et al., 2003; Stanyon et al., 2004), and mammalian cells (Rual et al., 2005) have been investigated using these methods. For example, a recent study used yeast two-hybrid screening to examine all pairwise interactions
among 8100 human genes (38% of all genes) and detected 2800 interactions (Rual et al., 2005). A major disadvantage of high-throughput protein interaction surveys is their limited reliability. False-positive and false-negative results are currently estimated at 40–50% for both yeast two-hybrid and MS methods (de Silva and Stumpf, 2005; Rual et al., 2005). Another strategy to map genetic networks employs curation of protein interactions from the literature (Gandhi et al., 2006; Reguly et al., 2006). These datasets are thought to have higher reliability since many interactions have been verified by independent methods from more than one group. The Human Protein Reference Database (HPRD) contains 8800 proteins (42% of all genes) and 25,000 interactions (Gandhi et al., 2006). Nevertheless, curation databases suffer from selection bias and do not randomly sample all protein interactions. Furthermore, comparison of yeast two-hybrid data and literature-derived interactions suggests that the literature-curated data can also suffer from variable quality (Rual et al., 2005). The networks derived from interaction mapping are often scale-free. Such networks show a power law distribution relating the probability of interactions, P(k), of a node on the network with k other nodes: P(k) k, where = 2 (Carter et al., 2004). A feature of scale-free networks is the presence of a few highly connected nodes or hubs, representing genes that are highly connected to many other genes. This network class is robust to random disconnection of individual nodes, but is vulnerable to disruption of highly connected hubs. Correspondingly, genes in hubs are found significantly more likely to be essential for survival (Gandhi et al., 2006). Both scale-free and classical random graphs display “small world” behavior, in which moving between any two nodes on the network takes far fewer steps (edges) than naively expected. Interactome Insights Some interesting insights related to disease have emerged from analysis of these networks. For example, the HPRD revealed that protein products of genes mutated in inherited genetic disorders tend to interact with proteins involved in similar disorders. This observation suggested that certain gene subnetworks are involved in particular diseases and can help guide the search for new genetic culprits. In another example, it was recently shown that combining yeast two-hybrid data, co-affinity purification data, and genetic experiments in worms gave a much sharper outline of the TGF- signaling pathway than using any one dataset alone (Tewari et al., 2004). A total of nine new TGF- signaling related genes were found, of which seven were conserved in humans. One of these had functional homology to the human SNO/SKI oncoprotein. Metabolomics No less promising than proteomics in its relation to early diagnosis and treatment of disease is metabolomics (Claudino et al., 2007) (see Chapter 15). It is estimated that there are approximately 3,000 metabolites, compared with 20,000 genes. The
References
main analytic tool for the metabolome is proton nuclear magnetic resonance spectroscopy (1H NMR). Another useful approach is MS, which has greater processing cost per sample but is thought to have more sensitivity. Metabolomic approaches have been used successfully for non-invasive diagnosis of a number of disorders, including breast cancer, hepatitis, ovarian cysts, and ovarian cancer (Claudino et al., 2007). Many of these studies used unsupervised analytic approaches such as principal components analysis (PCA) to distinguish diseased and nondiseased samples (Griffin, 2006) (Figure 3.2).
■
43
to fruition with the first complete genome sequences of individual members of many species. With the goal of the $1000 genome on the horizon, it is probable that relatively soon we will have complete genome sequences for many individuals within one species. Recent genomic advances in transcript profiling, proteomics, metabolomics, and molecular imaging suggest that comprehensive knowledge in orthogonal descriptive directions is also possible (Figure 3.2). This is systems biology. The challenge now is to integrate these diverse modes of knowledge to a better understanding of complex genetics in both health and disease and make the field of genomic medicine a practical reality.
CONCLUSIONS AND PROSPECTS In his famous textbook Molecular Biology of the Gene (Watson, 1965), James Watson pointed out that the genome is not of infinite size (Olson, 2003). He therefore argued (within physical limits) that in principle we could know everything about the molecular description of a cell or organism. This vision has come
ACKNOWLEDGEMENTS Supported by NIH R01 NS050148, MH71779, NS043562 (D.J. Smith) and NIH HL30568, HL30568, HL28481, DK071673 (A.J. Lusis).
REFERENCES Aitman, T.J., Glazier, A.M., Wallace, C.A., Cooper, L.D., Norsworthy, P.J., Wahid, F.N., Al-Majali, K.M., Trembling, P.M., Mann, C.J., Shoulders, C.C. et al. (1999). Identification of Cd36 (Fat) as an insulin-resistance gene causing defective fatty acid and glucose metabolism in hypertensive rats. Nat Genet 21, 76–83. Altshuler, D., Brooks, L.D., Chakravarti, A., Collins, F.S., Daly, M.J. and Donnelly, P. (2005). A haplotype map of the human genome. Nature 437, 1299–1320. Bailey, W.J. and Ulrich, R. (2004). Molecular profiling approaches for identifying novel biomarkers. Expert Opin Drug Saf 3, 137–151. Bennett, S.T., Barnes, C., Cox, A., Davies, L. and Brown, C. (2005). Toward the $1000 human genome. Pharmacogenomics 6, 373–382. Botstein, D., White, R.L., Skolnick, M. and Davis, R.W. (1980). Construction of a genetic linkage map in man using restriction fragment length polymorphisms. Am J Hum Genet 32, 314–331. Braslavsky, I., Hebert, B., Kartalov, E. and Quake, S.R. (2003). Sequence information can be obtained from single DNA molecules. Proc Natl Acad Sci USA 100, 3960–3964. Brenner, S., Johnson, M., Bridgham, J., Golda, G., Lloyd, D.H., Johnson, D., Luo, S., McCurdy, S., Foy, M., Ewan, M. et al. (2000). Gene expression analysis by massively parallel signature sequencing (MPSS) on microbead arrays. Nat Biotechnol 18, 630–634. Bucca, G., Carruba, G., Saetta, A., Muti, P., Castagnetta, L. and Smith, C.P. (2004). Gene expression profiling of human cancers. Ann N Y Acad Sci 1028, 28–37. Cantor, C.R. (2005). The use of genetic SNPs as new diagnostic markers in preventive medicine. Ann N Y Acad Sci 1055, 48–57. Carter, S.L., Brechbuhler, C.M., Griffin, M. and Bond, A.T. (2004). Gene co-expression network topology provides a framework for molecular characterization of cellular state. Bioinformatics 20, 2242–2250. Cheung, V.G., Spielman, R.S., Ewens, K.G., Weber, T.M., Morley, M. and Burdick, J.T. (2005). Mapping determinants of human gene expression by regional and genome-wide association. Nature 437, 1365–1369.
Claudino, W.M., Quattrone, A., Biganzoli, L., Pestrin, M., Bertini, I. and Di Leo, A. (2007). Metabolomics: Available results, current research projects in breast cancer, and future applications. J Clin Oncol 25, 2840–2846. Cohen, J.C., Kiss, R.S., Pertsemlidis, A., Marcel, Y.L., McPherson, R. and Hobbs, H.H. (2004). Multiple rare alleles contribute to low plasma levels of HDL cholesterol. Science 305, 869–872. Cohen, J.C., Pertsemlidis, A., Fahmi, S., Esmail, S., Vega, G.L., Grundy, S.M. and Hobbs, H.H. (2006). Multiple rare variants in NPC1L1 associated with reduced sterol absorption and plasma low-density lipoprotein levels. Proc Natl Acad Sci USA 103, 1810–1815. Cusick, M.E., Klitgord, N.,Vidal, M. and Hill, D.E. (2005). Interactome: Gateway into systems biology. Hum Mol Genet 14 (Spec No. 2), R171–R181. de Silva, E. and Stumpf, M.P. (2005). Complex networks and simple models in biology. J R Soc Interface 2, 419–430. Diabetes Genetics Initiative et al. (2007). Genome-wide association analysis identifies loci for type 2 diabetes and triglyceride levels. Science 316, 1331–1336. Domon, B. and Abersold, R. (2006). Mass spectrometry and protein analysis. Science 312, 212–217. Doss, S., Schadt, E.E., Drake, T.A. and Lusis, A.J. (2005). Cis-acting expression quantitative trait loci in mice. Genome Res 15, 681–691. Easton, D.F., Pooley, K.A., Dunning, A.M., Pharoah, P.D., Thompson, D., Ballinger, D.G., Struewing, J.P., Morrison, J., Field, H., Luben, R. et al. (2007). Genome-wide association study identifies novel breast cancer susceptibility loci. Nature 447, 1087–1093. Flint, J. and Mott, R. (2001). Finding the molecular basis of quantitative traits: Successes and pitfalls. Nat Rev Genet 2, 437–445. Flint, J., Valdar, W., Shifman, S. and Mott, R. (2005). Strategies for mapping and cloning quantitative trait genes in rodents. Nat Rev Genet 6, 271–286. Formstecher, E., Aresta, S., Collura, V., Hamburger, A., Meil, A., Trehin, A., Reverdy, C., Betin, V., Maire, S., Brun, C. et al. (2005).
44
CHAPTER 3
■
Genomic Approaches to Complex Disease
Protein interaction mapping: A Drosophila case study. Genome Res 15, 376–384. Gandhi, T.K., Zhong, J., Mathivanan, S., Karthick, L., Chandrika, K.N., Mohan, S.S., Sharma, S., Pinkert, S., Nagaraju, S., Periaswamy, B. et al. (2006). Analysis of the human protein interactome and comparison with yeast, worm and fly interaction datasets. Nat Genet 38, 285–293. Gavin, A.C., Bosche, M., Krause, R., Grandi, P., Marzioch, M., Bauer, A., Schultz, J., Rick, J.M., Michon, A.M., Cruciat, C.M. et al. (2002). Functional organization of the yeast proteome by systematic analysis of protein complexes. Nature 415, 141–147. Ghazalpour, A., Doss, S., Sheth, S.S., Ingram-Drake, L.A., Schadt, E.E., Lusis, A.J. and Drake, T.A. (2005). Genomic analysis of metabolic pathway gene expression in mice. Genome Biol 6, R59. Gibson, U.E., Heid, C.A. and Williams, P.M. (1996). A novel method for real time quantitative RT-PCR. Genome Res 6, 995–1001. Giot, L., Bader, J.S., Brouwer, C., Chaudhuri, A., Kuang, B., Li, Y., Hao, Y.L., Ooi, C.E., Godwin, B., Vitols, E. et al. (2003). A protein interaction map of Drosophila melanogaster. Science 302, 1727–1736. Grant, S.F., Thorleifsson, G., Reynisdottir, I., Benediktsson, R., Manolescu, A., Sainz, J., Helgason, A., Stefansson, H., Emilsson, V., Helgadottir, A. et al. (2006). Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes. Nat Genet 38, 320–323. Griffin, J.L. (2006). The Cinderella story of metabolic profiling: Does metabolomics get to go to the functional genomics ball? Philos Trans R Soc Lond B Biol Sci 361, 147–161. Griffin, T.J. and Smith, L.M. (2000). Single-nucleotide polymorphism analysis by MALDI-TOF mass spectrometry. Trends Biotechnol 18, 77–84. Heid, C.A., Stevens, J., Livak, K.J. and Williams, P.M. (1996). Real time quantitative PCR. Genome Res 6, 986–994. Helgadottir, A., Manolescu, A., Helgason, A., Thorleifsson, G., Thorsteinsdottir, U., Gudbjartsson, D.F., Gretarsdottir, S., Magnusson, K.P., Gudmundsson, G., Hicks, A. et al. (2006). A variant of the gene encoding leukotriene A4 hydrolase confers ethnicity-specific risk of myocardial infarction. Nat Genet 38, 68–74. Hinds, D.A., Stuve, L.L., Nilsen, G.B., Halperin, E., Eskin, E., Ballinger, D.G., Frazer, K.A. and Cox, D.R. (2005). Whole-genome patterns of common DNA variation in three human populations. Science 307, 1072–1079. Hirschhorn, J.N. (2005). Genetic approaches to studying common diseases and complex traits. Pediatr Res 57, 74R–77R. Hirschhorn, J.N. and Daly, M.J. (2005). Genome-wide association studies for common diseases and complex traits. Nat Rev Genet 6, 95–108. Ho, Y., Gruhler, A., Heilbut, A., Bader, G.D., Moore, L., Adams, S.L., Millar, A., Taylor, P., Bennett, K., Boutilier, K. et al. (2002). Systematic identification of protein complexes in Saccharomyces cerevisiae by mass spectrometry. Nature 415, 180–183. Hood, L., Heath, J.R., Phelps, M.E. and Lin, B. (2004). Systems biology and new technologies enable predictive and preventative medicine. Science 306, 640–643. Horikawa, Y., Oda, N., Cox, N.J., Li, X., Orho-Melander, M., Hara, M., Hinokio, Y., Lindner, T.H., Mashima, H., Schwarz, P.E. et al. (2000). Genetic variation in the gene encoding calpain-10 is associated with type 2 diabetes mellitus. Nat Genet 26, 163–175. Hubner, N., Wallace, C.A., Zimdahl, H., Petretto, E., Schulz, H., Maciver, F., Mueller, M., Hummel, O., Monti, J., Zidek, V. et al.
(2005). Integrated transcriptional profiling and linkage analysis for identification of genes underlying disease. Nat Genet 37, 243–253. Hugot, J.P., Chamaillard, M., Zouali, H., Lesage, S., Cezard, J.P., Belaiche, J., Almer, S., Tysk, C., O’Morain, C.A., Gassull, M. et al. (2001). Association of NOD2 leucine-rich repeat variants with susceptibility to Crohn’s disease. Nature 411, 599–603. Ito, T., Chiba, T., Ozawa, R., Yoshida, M., Hattori, M. and Sakaki, Y. (2001). A comprehensive two-hybrid analysis to explore the yeast protein interactome. Proc Natl Acad Sci USA 98, 4569–4574. Jansen, R.C. and Nap, J.P. (2001). Genetical genomics: The added value from segregation. Trends Genet 17, 388–391. Klein, R.J., Zeiss, C., Chew, E.Y., Tsai, J.Y., Sackler, R.S., Haynes, C., Henning, A.K., SanGiovanni, J.P., Mane, S.M., Mayne, S.T. et al. (2005). Complement factor H polymorphism in age-related macular degeneration. Science 308, 385–389. Kreiner, T. and Buck, K.T. (2005). Moving toward whole-genome analysis: a technology perspective. Am J Health Syst Pharm 62, 296–305. Lan, H., Chen, M., Flowers, J.B., Yandell, B.S., Stapleton, D.S., Mata, C.M., Mui, E.T., Flowers, M.T., Schueler, K.L., Manly, K.F. et al. (2006). Combined expression trait correlations and expression quantitative trait locus mapping. PLoS Genet 2, e6. Li, S., Armstrong, C.M., Bertin, N., Ge, H., Milstein, S., Boxem, M., Vidalain, P.O., Han, J.D., Chesneau, A., Hao, T. et al. (2004). A map of the interactome network of the metazoan C. elegans. Science 303, 540–543. Lipshutz, R.J., Morris, D., Chee, M., Hubbell, E., Kozal, M.J., Shah, N., Shen, N., Yang, R. and Fodor, S.P. (1995). Using oligonucleotide probe arrays to access genetic diversity. Biotechniques 19, 442–447. Livak, K.J. (1999). Allelic discrimination using fluorogenic probes and the 5 nuclease assay. Genet Anal 14, 143–149. Margulies, M., Egholm, M., Altman, W.E., Attiya, S., Bader, J.S., Bemben, L.A., Berka, J., Braverman, M.S., Chen, Y.J., Chen, Z. et al. (2005). Genome sequencing in microfabricated high-density picolitre reactors. Nature 437, 376–380. Moffatt, M.F., Kabesch, M., Liang, L., Dixon, A.L., Strachan, D., Heath, S., Depner, M., von Berg, A., Bufe, A., Rietschel, E. et al. (2007). Genetic variants regulating ORMDL3 expression contribute to the risk of childhood asthma. Nature 448, 470–473. Mootha,V.K., Lindgren, C.M., Eriksson, K.F., Subramanian, A., Sihag, S., Lehar, J., Puigserver, P., Carlsson, E., Ridderstrale, M., Laurila, E. et al. (2003). PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet 34, 267–273. Ogura, Y., Bonen, D.K., Inohara, N., Nicolae, D.L., Chen, F.F., Ramos, R., Britton, H., Moran,T., Karaliuskas, R., Duerr, R.H. et al. (2001). A frameshift mutation in NOD2 associated with susceptibility to Crohn’s disease. Nature 411, 603–606. Olson, M.V. (2003). The new quantitative biology. Cold Spring Harb Symp Quant Biol 68, 495–502. Ozaki, K., Ohnishi, Y., Iida, A., Sekine, A., Yamada, R., Tsunoda, T., Sato, H., Sato, H., Hori, M., Nakamura, Y. et al. (2002). Functional SNPs in the lymphotoxin-alpha gene that are associated with susceptibility to myocardial infarction. Nat Genet 32, 650–654. Pajukanta, P., Lilja, H.E., Sinsheimer, J.S., Cantor, R.M., Lusis, A.J., Gentile, M., Duan, X.J., Soro-Paavonen, A., Naukkarinen, J., Saarela, J. et al. (2004). Familial combined hyperlipidemia is associated with upstream transcription factor 1 (USF1). Nat Genet 36, 371–376.
References
Patil, N., Berno, A.J., Hinds, D.A., Barrett, W.A., Doshi, J.M., Hacker, C.R., Kautzer, C.R., Lee, D.H., Marjoribanks, C., McDonough, D.P. et al. (2001). Blocks of limited haplotype diversity revealed by high-resolution scanning of human chromosome 21. Science 294, 1719–1723. Pravenec, M., Wallace, C., Aitman, T.J. and Kurtz, T.W. (2003). Gene expression profiling in hypertension research: A critical perspective. Hypertension 41, 3–8. Reguly, T., Breitkreutz, A., Boucher, L., Breitkreutz, B.J., Hon, G.C., Myers, C.L., Parsons, A., Friesen, H., Oughtred, R., Tong, A. et al. (2006). Comprehensive curation and analysis of global interaction networks in Saccharomyces cerevisiae. J Biol 5, 11. Rioux, J.D., Daly, M.J., Silverberg, M.S., Lindblad, K., Steinhart, H., Cohen, Z., Delmonte, T., Kocher, K., Miller, K., Guschwan, S. et al. (2001). Genetic variation in the 5q31 cytokine gene cluster confers susceptibility to Crohn disease. Nat Genet 29, 223–228. Rual, J.F., Venkatesan, K., Hao, T., Hirozane-Kishikawa, T., Dricot, A., Li, N., Berriz, G.F., Gibbons, F.D., Dreze, M., AyiviGuedehoussou, N. et al. (2005). Towards a proteome-scale map of the human protein-protein interaction network. Nature 437, 1173–1178. Samani, N.J., Erdmann, J., Hall, A.S., Hengstenberg, C., Mangino, M., Mayer, B., Dixon, R.J., Meitinger, T., Braund, P., Wichmann, H.E. et al. (2007). Genomewide association analysis of coronary artery disease. N Engl J Med 357, 443–453. Schadt, E.E., Monks, S.A., Drake, T.A., Lusis, A.J., Che, N., Colinayo, V., Ruff, T.G., Milligan, S.B., Lamb, J.R., Cavet, G. et al. (2003). Genetics of gene expression surveyed in maize, mouse and man. Nature 422, 297–302. Schadt, E.E., Lamb, J., Yang, X., Zhu, J., Edwards, S., Guhathakurta, D., Sieberts, S.K., Monks, S., Reitman, M., Zhang, C. et al. (2005). An integrative genomics approach to infer causal associations between gene expression and disease. Nat Genet 37, 710–717. Scott, L.J., Mohlke, K.L., Bonnycastle, L.L., Willer, C.J., Li, Y., Duren, W.L., Erdos, M.R., Stringham, H.M., Chines, P.S., Jackson, A.U. et al. (2007). A genome-wide association study of type 2 diabetes in Finns detects multiple susceptibility variants. Science 316, 1341–1345. Sebat, J., Lakshmi, B., Troge, J., Alexander, J., Young, J., Lundin, P., Maner, S., Massa, H., Walker, M., Chi, M. et al. (2004). Large-scale copy number polymorphism in the human genome. Science 305, 525–528. Sebat, J., Lakshmi, B., Malhotra, D., Troge, J., Lese-Martin, C., Walsh, T., Yamrom, B., Yoon, S., Krasnitz, A., Kendall, J. et al. (2007). Strong association of de novo copy number mutations with autism. Science 316, 445–449. Shen, R., Fan, J.B., Campbell, D., Chang, W., Chen, J., Doucet, D., Yeakley, J., Bibikova, M., Wickham Garcia, E., McBride, C. et al. (2005). High-throughput SNP genotyping on universal bead arrays. Mutat Res 573, 70–82. Shendure, J., Mitra, R.D.,Varma, C. and Church, G.M. (2004). Advanced sequencing technologies: Methods and goals. Nat Rev Genet 5, 335–344. Shendure, J., Porreca, G.J., Reppas, N.B., Lin, X., McCutcheon, J.P., Rosenbaum, A.M., Wang, M.D., Zhang, K., Mitra, R.D. and Church, G.M. (2005). Accurate multiplex polony sequencing of an evolved bacterial genome. Science 309, 1728–1732. Smith, D.J. and Lusis, A.J. (2002). The allelic structure of common disease. Hum Mol Genet 11, 2455–2461.
■
45
Spielman, R.S. and Ewens, W.J. (1998). A sibship test for linkage in the presence of association: the sib transmission/disequilibrium test. Am J Hum Genet 62, 450–458. Stanyon, C.A., Liu, G., Mangiola, B.A., Patel, N., Giot, L., Kuang, B., Zhang, H., Zhong, J. and Finley, R.L, Jr. (2004). A Drosophila protein-interaction map centered on cell-cycle regulators. Genome Biol 5, R96. Stefansson, H., Sigurdsson, E., Steinthorsdottir, V., Bjornsdottir, S., Sigmundsson, T., Ghosh, S., Brynjolfsson, J., Gunnarsdottir, S., Ivarsson, O., Chou, T.T. et al. (2002). Neuregulin 1 and susceptibility to schizophrenia. Am J Hum Genet 71, 877–892. Steinthorsdottir, V., Thorleifsson, G., Reynisdottir, I., Benediktsson, R., Jonsdottir, T., Walters, G.B., Styrkarsdottir, U., Gretarsdottir, S., Emilsson, V., Ghosh, S. et al. (2007). A variant in CDKAL1 influences insulin response and risk of type 2 diabetes. Nat Genet 39, 770–775. Strittmatter, W.J., Saunders, A.M., Schmechel, D., Pericak-Vance, M., Enghild, J., Salvesen, G.S. and Roses, A.D. (1993). Apolipoprotein E: High-avidity binding to beta-amyloid and increased frequency of type 4 allele in late-onset familial Alzheimer disease. Proc Natl Acad Sci USA 90, 1977–1981. Suviolahti, E., Oksanen, L.J., Ohman, M., Cantor, R.M., Ridderstrale, M., Tuomi, T., Kaprio, J., Rissanen, A., Mustajoki, P., Jousilahti, P. et al. (2003). The SLC6A14 gene shows evidence of association with obesity. J Clin Invest 112, 1762–1772. Tewari, M., Hu, P.J., Ahn, J.S., Ayivi-Guedehoussou, N., Vidalain, P.O., Li, S., Milstein, S., Armstrong, C.M., Boxem, M., Butler, M.D. et al. (2004). Systematic interactome mapping and genetic perturbation analysis of a C. elegans TGF-beta signaling network. Mol Cell 13, 469–482. Tuzun, E., Sharp, A.J., Bailey, J.A., Kaul, R., Morrison, V.A., Pertz, L.M., Haugen, E., Hayden, H., Albertson, D., Pinkel, D. et al. (2005). Fine-scale structural variation of the human genome. Nat Genet 37, 727–732. Uetz, P., Giot, L., Cagney, G., Mansfield,T.A., Judson, R.S., Knight, J.R., Lockshon, D., Narayan, V., Srinivasan, M., Pochart, P. et al. (2000). A comprehensive analysis of protein-protein interactions in Saccharomyces cerevisiae. Nature 403, 623–627. Wang, S.,Yehya, N., Schadt, E.E., Wang, H., Drake, T.A. and Lusis, A.J. (2006). Genetic and genomic analysis of a fat mass trait with complex inheritance reveals marked sex specificity. PLOS Genet 2, e15. Wardell, M.R., Suckling, P.A. and Janus, E.D. (1982). Genetic variation in human apolipoprotein E. J Lipid Res 23, 1174–1182. Watson, J.D. (1965). Molecular Biology of the Gene, 1st edition. W. A. Benjamin, New York. Weber, J.L. and May, P.E. (1989). Abundant class of human DNA polymorphisms which can be typed using the polymerase chain reaction. Am J Hum Genet 44, 388–396. Wellcome Trust Case Control Consortium (2007). Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 447, 661–678. Wheeler, D.L., Barrett, T., Benson, D.A., Bryant, S.H., Canese, K., Chetvernin,V., Church, D.M., DiCuccio, M., Edgar, R., Federhen, S. et al. (2006). Database resources of the National Center for Biotechnology Information. Nucleic Acids Res 34, D173–D180. Zeggini, E., Weedon, M.N., Lindgren, C.M., Frayling, T.M., Elliott, K.S., Lango, H., Timpson, N.J., Perry, J.R., Rayner, N.W., Freathy, R.M. et al. (2007). Replication of genome-wide association signals in UK samples reveals risk loci for type 2 diabetes. Science 316, 1336–1341.
46
CHAPTER 3
■
Genomic Approaches to Complex Disease
RECOMMENDED RESOURCES 454 Life Sciences www.454.com Affymetrix www.affymetrix.com Applied Biosystems https://www2.appliedbiosystems.com/ Gene ontology – the gene ontology classification, http://www.geneontology.org Genome-phenome superbrain http://omicspace.riken.jp HapMap http://www.hapmap.org/ Helicos Biosciences http://helicosbio.com Illumina www.illumina.com
Jackson laboratory – a repository for genetic and genomic information on the mouse, http://www.jax.org National Center for Biotechnology Information – repository for SNP information, comprehensive literature database and much more, www.ncbi.nlm.nih.gov Perlegen http://www.perlegen.com/ Sequenom www.sequenom.com UCSC Genome Browser – available genome sequences, http:// genome.ucsc.edu/
CHAPTER
4 Human Health and Disease: Interaction Between the Genome and the Environment Kenneth Olden
INTRODUCTION The importance of inheritance versus the environment in the origin of various diseases has been debated for more than a century. The phrase “nature versus nurture” was coined by Francis Galton (a 19th-century British scientist) to distinguish between characteristics one is born with and those that are acquired as a result of one’s surroundings. Early on, humans looked to the external environment to explain morbidity and mortality as they were cognizant of their exposure to natural agents that were poisonous. Because of investments in environmental health research, we now know that many of the chemical, physical, and biological agents in the environment not only cause acute illness and death, but also increase risk for various chronic diseases such as cancer, asthma, hypertension, stroke, Alzheimer’s, Parkinson’s, infections, and various developmental disorders. However, the era of dominance of external factors as the primary contributor to human disease gradually faded with the development and growth of Mendelian genetics (Dobzhansky, 1941) and the more recent field of genomics (Trask, 1999). In fact, the geocentric view of disease has dominated our concept of the origin of disease for the past 50 years. However, with the completion of the sequencing of the human genome (Lander et al., 2001; Venter et al., 2001), it now seems apparent that complex human Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
phenotypes cannot be explained based solely on the action of genes or their protein products. Genes account for only a small part of the phenotype, and the environment has a spectacular impact. That is, the onset and clinical course of most chronic diseases are influenced by complex interactions between genes and the environment (Olden, 2007). The concept of gene–gene and gene–environment interactions is not new. Garrod (1902) was one of the first to suggest that the effects of genes could be modified by the environment. He suggested that individual differences in genetics could play a role in variation in response to drugs, and that the effect of genotype could be modified by the diet. Wright (1932) also emphasized the existence of a functional relationship between various biological endpoints and networks of genes and environmental factors in his studies of mutation, selection, and breeding, and Dobzhansky and Levene (1955) reported the existence of genotype–environment interaction with respect to viability of some chromosomes in Drosophila. Whereas concerns about gene–environment interactions have had a long history (reviewed by Haldane, 1946), the importance of understanding these interactions as a means to prevent complex diseases emerged only over the past 15 years (Olden, 1993; Olden and Guthrie, 2000; Olden et al., 2001; Schwartz and Collins, 2007). The geocentric view of disease was virtually unchallenged until the leadership of Copyright © 2009, Elsevier Inc. All rights reserved. 47
48
CHAPTER 4
■
Genome and the Environment
the environmental health community began to emphasize gene– environment interaction as a top research priority in 1992 (Olden, 1993; Washington FAX, January 21, 1992). A few years later, the initiation of the Environmental Genome Project (EGP) (discussed later in this chapter) represented the first large-scale effort to discover susceptibility alleles likely to be important in gene–gene or gene–environment interactions (Brown and Hartwell, 1998; Guengerich, 1998; Kaiser, 1997; Olden and Wilson, 2000). Prior to the initiation of the EGP, most of the momentum for research on gene–environment interactions was occurring in the field of pharmacogenetics to identify genetic variants that influence drug efficacy. As a consequence, many published studies have linked specific polymorphisms with specific drug responses (Frye, 2004; Hulla et al., 1999; Nebert et al., 1996). Studies have shown that interindividual pharmacokinetic variation in rates of drug elimination can vary dramatically (Vessel, 1991). Multinational twin studies were conducted on the kinetics of several drugs to compare relative contribution of genetics and environmental factors with respect to interindividual variations. Results were remarkably similar for the various populations, and pharmacokinetic variation virtually disappeared within monozygotic twins, but was preserved within most dizygotic twins; meaning that variation in drug metabolism is primarily under genetic control. Altering the environment of the monozygotic twin had no effect on kinetics. If reported pharmacogenetic differences of 10- to 200-fold can be extrapolated directly to the risk of human diseases, one could conclude that an individual can be 10- to 200-fold more sensitive to a given drug or environmental chemical because of differences in expression and activity of metabolic enzymes. Recent studies have shown a similar association between specific diseases and various genetic polymorphisms. Some examples of both types of genotype–phenotype association studies will be described later. However, even when all the highly penetrant alleles have been discovered and all the gene–gene and gene–environment interactions are understood, it may still be difficult to relate genotype to phenotype because of the additional layers of complexity associated with post-translational processing and protein–protein interactions. Kouhry and Wagner (1995) suggested that there are at least six types of gene– environment interactions. In the first situation, neither the environment nor the genotype has any effect alone, but can interact to increase risk. In the second situation, the genotype has no effect in the absence of exposure, but can exacerbate the effects of the latter. The third situation is the converse of the second; that is, the environment is without effect alone, but can enhance the effect of the genotype. In the fourth situation, both the environmental exposure and the genotype increase risk for disease, but the combination is interactive or synergistic. The fifth and sixth situations represent cases in which gene–environment interactions are protective. Phenylketonuria is an example of disorder that is caused by both the susceptibility genotype and exposure to phenylalanine in the diet. In contrast, lung cancer can be caused by cigarette smoking even if the susceptibility genotype is not present, and acute hemolytic anemia can be caused by the susceptibility
genotype alone even in the absence of fava bean digestion. Individuals with the susceptibility genotype for a1-antitrypsin deficiency have a high risk for development of emphysema, but the risk is greatly increased by cigarette smoking. The model of gene–environment interaction that has been most studied relates to metabolic susceptibility genes and chemical carcinogenesis. Polymorphisms in carcinogen-metabolizing enzymes (to be discussed later) are relatively common in the population and can play a major role in the development of the common cancers. The phrase gene–environment interaction means that the direction and magnitude of the effect that a genetic variant has on the phenotype can vary as the environment changes; that is, genetic risk is modifiable in an environment-specific manner or that the phenotypic effect of a mutation or environmental exposure depends on the genetic background. Whether a particular variant allele is expressed, the degree to which it is expressed, and when it is expressed can be influenced by the environment or by other genes. However, since the genotype evolves or changes very slowly over hundreds of years, one can assume that most of the increase in disease burden in industrialized nations is the result of a variable environment interacting with a relatively constant genetic substrate. Similar to gene–environment interactions, variants in multiple genes can contribute to the expression of phenotype. For example, multiple genetic loci influence the rate of progression of HIV infection to full blown AIDS (Dean, 2003). The nonsense allele, V641, in the CCR2 gene and heterozygosity for the cytokine receptor/HIV-1 co-receptor CCR5- 32 allele is reported to be associated with the slower progression of AIDS (Dean et al., 1996; Smith et al., 1997); in contrast, a haplotype in the promoter region of CCR5 is associated with the more rapid progression of HIV infection to AIDS (Martin et al., 1998). Also, progression of HIV-infected patients with more heterologous HLA Class I genes is considerably slower (Carrington et al., 1999).
IMPORTANCE OF THE ENVIRONMENT Over the past 50 years, the prevalence of chronic diseases has reached epidemic proportions in industrialized countries, and various governments have expended enormous resources to identify genetic causes. Nevertheless, the expectation that simple direct paths would be found from genes to disease has not been realized. Several explanations have been put forward to explain failure, but gene–gene and gene environment interactions are among the most likely. The merging view is that exposure to environmental toxicants has conspired with genetics to promote the development of these disorders. It is now well established that alterations in high penetrant genes explain only a small fraction of complex diseases, and such genes represent only a small fraction of variations relative to the more common polymorphisms that have a less disruptive effect on protein function. However, it is much less obvious what portion of the disease burden is attributable to interactions among low penetrant
Importance of the Environment
genes versus interactions of low penetrant genes with the environment. Nevertheless, a prominent role for the environment is supported by findings of geographic differences in incidence of disease, by variations in trends over time and by studies of disease patterns in immigrant populations (Doll and Peto, 1981). Also, population-based, twin-cohort studies, the “gold standard” for distinguishing between the contribution of genetics and the environment, suggest that the environment plays a prominent role in the development of human disease. In many cases, concordance among monozygotic twins is very low. For example, such recent studies show that no more than one-third of the cancer burden can be attributed to the action of genes alone (Lichtenstein et al., 2000; Verkasala et al., 1999), only 15% of Parkinson’s disease (Tanner et al.,1999), and about a third of autoimmune diseases (Powell et al., 1999). In a much earlier landmark study, Doll and Peto (1981) estimated that about one-third of all cancers are caused by genes alone based on demographic distribution of the disease in the United States. And, a recent study (Khot et al., 2003) reported that 90% of individuals with severe heart disease have at least one or more of the four classic risk factors captured in the current definition of the environment (Olden et al., 2001). Because of these and other studies, it is now generally accepted that a single risk factor is rarely responsible for the development of most chronic diseases (see Chapter 3). Rather, diseases commonly arise from complex interactions involving genes, environmental factors, behavior, and stochastic events associated with DNA replication and endogenous metabolism, as depicted in the schematic model shown in Figure 4.1. Age and stage of development are also important factors in disease development, as they influence the timing of gene expression and susceptibility to environmental toxicants. Many people with susceptibility genes never actually develop the disease, even for diseases caused by highly penetrant alleles (e.g., BRCA1 and BRCA2 breast cancer genes) (Futreal et al., 1994). There is always a subpopulation of carriers of susceptibility genes who do not progress to the ultimate disease state. Whether one can account for the discordance between genotype and phenotype by lack of exposure to unknown environmental risk factors or to yet undiscovered genomic variations or gene–environment interactions
Intrinsic Genetic Susceptibility
Environmental Exposures
Human Health/Disease
Behavior/Age/Stage of Development
Figure 4.1 Schematic model of interactive factors involved in human health and disease.
■
49
is not known. However, existing evidence for a prominent role for the environment in the development of human disease is so compelling that Rothman et al. (2001) concluded that “the epidemiologic evidence accumulated to date indicates that environmental exposures, broadly defined to include lifestyle factors, are responsible for most cancers.” Others have concluded that the new data now being reported “blow away” the myth that “bad genes” are responsible for the majority of human morbidity and mortality (Olden, 2004a, b). That is, individual life history may be just as important as genetic background in predicting and understanding the development of complex traits. Genetics Loads the Gun, but the Environment Pulls the Trigger Is the relationship between genes and the environment analogous to that of a loaded gun and its trigger, as previously proposed? Olden and Wilson (2000) argued that the inheritance of disease susceptibility alleles creates the potential for adverse health outcomes, but that the adverse health outcome may never be realized provided the carrier can avoid environmental exposures capable of interacting with the susceptibility gene(s) to trigger disease development. If this analogy fully captures the relationship between environmental susceptibility alleles and specific environmental exposures, this can be good news because amelioration of environmental hazards could lead to a dramatic reduction in the incidence of chronic diseases in human populations. From an evolutionary point of view, one could argue that genes and gene variants that were adaptively selected in the past because they endowed the human species with survival or reproductive advantage, now have the potential to increase disease risk because of their incompatibility with the environment. The change in the environment, resulting from human activity and technology innovation, has outpaced evolution of the human genome to the extent that there is now a nature–nurture mismatch. For years, environmental health advocates have had to deal with the question of why more bodies are not piling up on the streets if industrial chemicals, such as benzene, vinyl chloride, and dioxins, are so harmful. Why can some people withstand exposure to nerve gases such as sarin, and organophosphate pesticides such as malathion and parathion? Why do fewer than 25% of smokers develop lung or bladder cancer? How do some individuals who work in the petrochemical industry survive years of exposure to benzene and never develop leukemia? Human population studies to assess gene–environment interactions are beginning to provide answers to these questions. For example, about 25% of Asians and 10% of Caucasians have a protective form of the gene that codes for paraoxonase (PON1, to be discussed later), an enzyme that detoxifies organophosphates, such as sarin and the above-mentioned insecticides, ten times faster than the more common form of the enzyme (Costa and Furlong, 2002). Also, common polymorphisms in genes for the cytochrome P450 enzymes determine how one metabolizes carcinogens (i.e., arylamine) in cigarette smoke or benzene in the environment of petrochemical workers (Lower et al., 1979; Smith et al., 2002). Such studies are beginning to
50
CHAPTER 4
■
Genome and the Environment
explain why a given polymorphism can be highly beneficial in one environment and virtually lethal in another (Boguski, 1999). Furthermore, they suggest that risk reduction can be achieved by re-engineering of genes or by modification of the environment. Now that the Human Genome Project (Lander et al., 2001; Venter et al., 2001) and the haplotype mapping (HapMap) project (Altshuler et al., 2005) have been completed, the major challenge facing biomedical researchers is to elucidate the interactive relationship between genes and between genes and the environment. While descriptive studies by toxicologists and epidemiologists have shed some light on the possible genetic and environmental risk factors, the actual causes and mechanisms of disease development remain poorly understood. Investments in the aforementioned sequencing and mapping projects led to the development of powerful new tools to conduct large-scale, population-based studies necessary to untangle complex interactions between genes and environmental factors. However, the design and conduct of such studies will require the development of even better tools and fundamental restructuring of the biomedical research enterprise to put greater, although not exclusive, emphasis on interdisciplinary research teams. Promise of New Technologies for Unraveling Complex Interactions Between the Human Genome and the Environment The Nobel Prize laureate Sydney Brenner has written that “progress in science depends on new technologies, new discoveries and new ideas, probably in that order” (Brenner, 2002). In no discipline is this more evident than in environmental health research. Environmental health regulation has long been constrained by limitations in our ability to determine the risk presented by specific xenobiotic agents. Integration of genomics and toxicology to develop a new field of toxicogenomics may revolutionize risk assessment and environmental health regulation (Lovett, 2000). Over the past 50 years, advances in molecular biology have dramatically improved our understanding of human biology. Such innovations include the discovery of the DNA double helix (Watson and Crick, 1953), development of recombinant DNA technology (Watson et al., 1983), the polymerase chain reaction (Chamberlin, 1982), siRNA technology (Baulcombe, 2002), gene and protein microarray technology (Brown and Botstein, 1999; Schena et al., 1995), and DNA sequencing methodology (Lander et al., 2001). For example, an important insight gained from gene sequencing studies is that the human genome consists of a large number of polymorphic sites, some of which are associated with altered biological function or phenotype (see Chapter 1). The finding of a large number of variant forms of genes suggested that genetic polymorphisms may play a significant role in susceptibility to disease or toxic exposure. While descriptive studies by toxicologists, molecular geneticists, and epidemiologists, working independently, have shed some light on this topic, the actual role of gene variation in susceptibility to disease development remains poorly understood.
THE ENVIRONMENTAL GENOME PROJECT Although we are beginning to understand how mutations in single genes effect phenotype, we have very limited knowledge of how variations in different genes influence each other and the sensitivity of biological mechanisms to environmental perturbation. But, with the completion of the reference sequence of the human genome, the field of genomics shifted from gene discovery to the study of gene function and elucidation of the genetic basis for differences in susceptibility to common diseases. Although reference is made to the human genome, the concept of a single genome is misleading. Whereas variation in the human genome is not greater than 0.1% among individuals in the population, this represents very significant variation considering that the human genome is composed of approximately three billion nucleotide base pairs. Genetic linkage analysis has been a powerful approach in the discovery of high penetrance allelic variants (e.g., BRCA1/2). However, this approach has not yet been particularly useful in the identification of low penetrance susceptibility alleles associated with the development of complex diseases such as cancer, cardiovascular disease, asthma, diabetes, Alzheimer’s and Parkinson’s disease because the relative low risk associated with a specific allele may be influenced by unknown environmental factors. Of the many types of variations in the human genome, single nucleotide polymorphisms (SNPs) are the most common (Collins et al., 1999). Estimates are that there is on average one SNP per 1300 base pairs, four to five per gene-coding region, or approximately 11 million in total in the human genome (Kruglyak and Nickerson, 2001). The EGP was formally initiated by the US National Institute of Environmental Health Sciences (NIEHS) in 1997 to investigate the role of genetic polymorphisms in susceptibility to environmentally induced disease (Brown and Hartwell, 1998; Kaiser, 1997, 2003). In this regard, NIEHS was way out front in recognizing the importance of genetic variation for understanding human disease. Following their lead, other large-scale public and private initiatives have been developed to identify the millions of SNPs in the human genome. Most SNPs can be expected to be neutral with respect to alteration of environmental sensitivity. To examine the extent of diversity in the human genome, the EGP was organized into five research areas: (1) SNP discovery; (2) functional analysis; (3) disease association studies; (4) development and refinement of technologies to improve sequencing, functional analysis, and population-based studies; and (5) development of policies to address possible ethical, social, and legal issues associated with genetic susceptibility studies. The 544 candidate genes included in the initial survey (see http://www.genome. utah.edu/genesnps and http://www.niehs.nih.gov/research/ supported/programs/egp/) had already been characterized with respect to function and were known to have a significant probability of playing a role in susceptibility to specific environmental exposures. The resequencing analysis of individual DNA samples was based on samples from 90 individuals, representative of the
The Environmental Genome Project
diversity in the US population. This design would allow for the detection of variation that occurs at a frequency of one percent or greater. As of September 2007, approximately 619 environmental response genes have been resequenced with the discovery of 88,016 SNPs. All of the SNPs entered into the database have been validated by multiple sequence analysis and include genotype frequency information. This highly useful SNP database is publicly available (http://www.egp.gs.washington.edu/welcome.html). Decades of studies by environmental health researchers, demonstrating tremendous variation in individual response to environmental exposures, provided the intellectual foundation for the massive resequencing effort undertaking in the EGP. Presumably, natural selection, resulting from exposures to various environmental insults, has operated over the course of human history to promote the evolution of sophisticated metabolic pathways to suppress phenotypic variation or buffer against toxic injury. Buffering is a fundamental aspect of biology, and there are likely to be multiple mechanisms involved, including genetic redundancy, feedback regulation, and cooperative biochemical interactions (Wagner et al., 1997; Gibson and Wagner, 2000; Hartman et al., 2001). Buffering was first inferred from the observation that organisms can acquire a wide range of phenotypes when subjected to varying genetic or environmental stressors (Scheiner et al., 1991). Collectively, the various buffering pathways involved in protecting the organism against environmental stressors have been referred to as the “environmental response machinery” (Olden and Wilson, 2000). According to this model, all human genes, including those that encode protein components of the environmental response machinery, are subject to genetic variability (mutations or polymorphisms) that can alter the metabolic efficiency of protective pathways. Some alleles can strengthen protection and others can reduce it. Another result emerging from the SNP discovery effort is that the variants tend to associate with one another in a nonrandom manner known as linkage disequilibrium (Carlson et al., 2004); this pattern of association is called a haplotype. On average, the EGP candidate genes are organized into 25 haplotypes (Wilson and Olden, 2004). The current thinking is that the generation of haplotype and linkage disequilibrium data will be extremely useful in large-scale genotyping studies to investigate the association between specific genes and disease. Such studies are now underway in several laboratories (see Chapter 8). However, the goal of assigning specific genotypes to phenotypes with respect to environmental sensitivity has proven to be a daunting assignment. For functional analysis of the SNPs discovered in the EGP, both in vitro biochemical and animal model studies have been initiated. For example, Furlong and his colleagues (Costa and Furlong, 2002) investigated the functional consequences of polymorphisms in paraoxonase (PON1) genes. PON1 is a highdensity lipoprotein associated with the plasma membrane that metabolizes oxidized lipids, environmental xenobiotics (e.g., organophosphate insecticides mentioned above) and some pharmaceutical agents (e.g., statins). Studies by Furlong and colleagues suggest that PON1 may play a significant role in the development of cardiovascular disease independent of environmental exposures (Costa and Furlong, 2002). However, their
■
51
studies have shown that the activity of PON1 is significantly altered by genetic variation and that these alterations influence sensitivity to the organophosphate insecticides chlorpyrifos and diazinon. For example, one of the two SNPs in the coding region significantly increased the substrate-specific activity of the enzyme and provides much greater protection against chlorpyrifos oxon than the wild-type PON1. However, the Q192R allele did not correlate with carotid artery disease in a case– control study (Jarvik et al., 2003). Whereas individual SNPs in the promoter region had no effect on enzyme activity, combinations of three of the common haplotypes of the five SNPs in this region of the gene did alter the plasma activity of PON1. While these studies do suggest that polymorphisms in PON1 may play a role in susceptibility to organophosphate toxicity, the exact involvement and possible interactions of the various SNPs in the PON1 gene are yet to be determined. In another project in the EGP, Martyn Smith and colleagues (Smith et al., 2001) identified a gene that codes for NAD(P)H: quinone receptor oxidoreductase1 (NQO1) and demonstrated its potential involvement in the etiology of leukemia. This enzyme plays an important role in protection against oxidative damage from exposure to quinones. And they found that the common C609T polymorphism resulted in an amino acid substitution with a complete loss of enzyme activity in homozygotes. The C609T variant was associated with a 2.5-fold increase in risk for leukemia in case–control studies (Krajinovic et al., 2002; Smith et al., 2001, 2002). In contrast, several polymorphisms in genes involved in folate metabolism reduced the risk of leukemia by two- to threefold in heterozygotes and 3- to 10-fold in homozygotes (Skibola et al., 1999, 2002). Folate is a source of precursors for DNA synthesis, repair and methylation; and dietary folate protects against leukemia (Thompson et al., 2001). Also, there is evidence for a role of gene–environment interactions between polymorphic variants of thymidylate synthetase and methylenetetrahydrofolate reductase (Skibola et al., 2002). Finally, for functional analysis of the various polymorphisms discovered in the EGP, NIEHS developed mouse genome centers in 2000. The mouse is the most widely used mammalian model system for the study of the impact of the environment on human health. The mouse is particularly useful in studies of gene–environment interactions because most human genes have counterparts in the mouse genome. Furthermore, extensive toxicological and pathological databases and molecular genetic techniques for construction of gene-knockout and knock-in strains have been developed using the mouse as a model. The mouse model also offers the advantage of using strain-specific differences in response to environmental exposure to investigate how the genetic background influences susceptibility. First, the mouse genome center resource was used to create genetic variation map databases of laboratory mouse strains for use in studies of gene–environment interactions. This was achieved by using knockout and knock-in technologies to generate polymorphic variations in candidate genes similar to those discovered in the EGP. Exposing such animals to putative environmental risk factors is the most obvious and direct way
52
CHAPTER 4
■
Genome and the Environment
to investigate the relationship between genetic background and susceptibility to environmental exposures. Second, since various mouse strains are known to exhibit significant variation in susceptibility to carcinogens and other environmental toxicants, efforts were undertaken to discover the haplotype maps of the 15 most commonly used mouse strains to identify genetic sequences that underlie differences in response (Olden 2004a; Pearson 2004). Once susceptibility haplotypes have been identified, one can track down the genes involved by crossing a susceptible strain with a resistant strain and identifying the chunks of the genome that are associated with the susceptibility. As the biological and genetic databases expand, the need for such mouse crosses might no longer be necessary.
PROBLEMATIC NATURE OF GENE– ENVIRONMENT INTERACTION STUDIES While there is considerable interest and enthusiasm for research on gene–environment interaction, there is also apprehension and uncertainty about the feasibility of such studies and practical questions about how to conduct them. Current studies underway in different parts of the world are problematic because they do not employ a standard definition of the environment nor standard metrics to assess exposure. But, even more problematic is the fact that methods to accurately measure exposure and to statistically model multiple interactive components that make modest contributions to the phenotype are not yet available. To alter the current state of affairs with respect to technology and database needs will require investments in the following areas of research: (i) identification of disease risk factors; (ii) elucidation of genetic differences and similarities between human and mouse; (iii) development of better tools and statistical models to assess exposure and interactions of multiple components; (iv) development of high-throughput, low-cost and more informative strategies to assess toxicity of drugs and environmental xenobiotics, and (v) elucidation of mechanisms and metabolic pathways influenced by gene–environment interactions. Furthermore, it has become clear that addressing the role of gene–environment interactions in the etiology of complex diseases will require the development of a robust framework to account for “the environment” (see Chapter 83). In spite of current understanding of the multifactorial nature of the etiologic mechanisms of chronic diseases, biomedical researchers still tend to focus on limited or circumscribed components of disease. Epidemiologists tend to focus on exposures; geneticists target susceptibility genes; cell/molecular biologists tend to explore mechanisms; and social scientists study behavior. While the various disciplinary approaches have led to new insights into disease causation, they are unlikely to be able to provide a coherent explanation for disease causation. Given this reality,
the environmental health community moved swiftly to take advantage of new databases and tools, derived from the human genome project, to identify and characterize all the environmental susceptibility alleles in the human genome and all the functional molecules (e.g., RNA, protein, carbohydrates, lipids, and metabolites) encoded by human and mouse genomes (Kaiser, 1997; Lovett, 2000; Olden, 2004a, b; Waters et al., 2003). With approximately 25,000 genes, hundreds of thousands of protein species, and numerous metabolites, the task of identifying and characterizing them with respect to function is an ambitious undertaking, one requiring improvement, standardization, and validation of existing methods to achieve reproducibility, increased sensitivity and specificity, and high throughput. Such studies will also require large-scale, multi-institutional collaboration and interdisciplinary expertise to amass and analyze such large databases. Until the latter part of the 19th century, infectious diseases were the chief causes of death; and their development was characterized by relatively short latency periods between exposure and the onset of illness. The relative ease in identifying the causative agents led to the development of effective prevention measures in the United States and other industrialized nations. Unfortunately, our capacity to prevent infectious diseases tracked with the emergence of chronic diseases, which have latency periods of 10–20 years. The increased prevalence of chronic diseases, the long latency period, and the more insidious nature of their causes necessitated development of new approaches to study their etiology. Epidemiologists responded by developing association studies to analyze the statistical relationship between two or more variables (e.g., genotype, exposure, and disease), based on two fundamental assumptions: first, that human disease does not occur at random; and second, that human disease has causal factors that can be identified by studying different populations. This disciplinary approach has been remarkably successful in the discovery of major risk factors for many common diseases, but has been less successful in discovering risks resulting from gene–gene or gene–environment interactions. Gene–environment interactions have been identified by taking advantage of existing models that take into account various ways that genetic effects can be modified by environmental exposures (Hunter, 2005). Epidemiologists have typically used one of three study designs: retrospective case–control, prospective cohort, and family-based. Family-based association design estimates association between genes and phenotypes based on the assumed Mendelian transmission of alleles from parent to offspring (Kraft and Hunter, 2005). In this study design, it is often difficult to obtain genetic data on parents for late-onset diseases, and exposure data is subject to recall bias as it is collected retrospectively. A further limitation of this approach is that the penetrance of the gene has to be sufficiently strong to cause familial aggregation. However, family-based studies allow one to assess the influence of genes over several generations; if the penetrance of the gene changes over time, this might represent the effect of the environment and lifestyle. In the case–control design, data on exposure is obtained after diagnosis of the disease, so errors in
Polymorphism and Disease Susceptibility: Case–Control Studies
estimating risk from gene–environment interactions can result from exposure recall or from bias in selection of the controls. Even if appropriate controls are found, incentive for participation in the study is often low; so, obtaining high participation is challenging. Also, it is difficult to estimate the effect of the disease on biomarkers used in the study. However, the case–control design is popular among epidemiologists because cases and controls can be enrolled in a short period of time, which leads to reduced cost and delay in reporting out the results. Finally, in the populationbased cohort design, information on exposure is collected prospectively, and tissue for DNA and baseline biomarker analysis is collected before the disease develops. The advantage of this study design is that assessment of exposure is more reliable and it does not suffer from selection bias. Unfortunately, such studies are very costly and time consuming and lack sufficient number of subjects to evaluate risks associated with gene–gene or gene–environment interactions. However, one can increase the power of populationbased cohort studies by pooling data from several studies. Because of the cost and time involved, most of the previous studies of gene–environment interactions have used the case– control design, which relied on family history to infer genetic predisposition and questionnaires to assess environmental and lifestyle exposures. In future studies of gene–environment interactions, it is likely that data from case–control and population-based cohort studies will be combined, and that genotype and exposure will be analyzed with greater precision using more direct and quantitative technologies. For example, the human genome and the various polymorphism discovery projects have opened up a vast new terrain to search for gene–gene or gene–environment interactions. Once these discovery efforts have evolved to the point that genetically susceptible populations have been identified, targeting such populations would increase the magnitude of relative risk, the significance of the observed associations, and the likelihood of discovering important gene–gene or gene– environment interactions. But, because of cost, rigorously designed case–control studies will remain the only option for assessing gene–environment interactions for rare diseases. Exploration of this new vast genomic landscape presents major challenges with respect to design and analysis of meaningful studies, and translation of the findings into public health and clinical practice. In addition to the challenges associated with population studies, the study of gene–environment interactions requires accurate information on exposure to eliminate a major source of uncertainty and the most serious problem in environmental health risk assessment. Not only is it difficult to measure exposure accurately, it is often prohibitively expensive for use in studies of large populations needed to test for association between genes and disease to correlate genetic variation with risk for disease.The current practice of estimating exposure using indirect surrogates, such as toxic release and production inventories and environmental monitoring, is inadequate and limits our understanding of dose–response relationships. Current methods available for measuring exposure include the use of environmental sensors, geographical information systems, biological sensors, direct analysis of body fluids, and more recently, toxicogenomics. Unfortunately,
■
53
all of these exposure assessment tools, with the possible exception of toxicogenomics, have serious limitations. The biological response to environmental exposure is so complex and involves so many interactive factors that the utilization of a combination of exposure assessment tools, capable of integrating all the interactive components, will be required in prospective epidemiologic studies. In my opinion, the development of the field of toxicogenomics will provide powerful and relatively inexpensive tools to identify biomarkers and to relate exposure and biological events during disease progression. In addition to impediments imposed by sample size and exposure assessment tools, population-based studies are fraught with other challenges related to the need for better statistical models. To define complex exposure-disease relationships and interactions between genes and environmental factors in the development of human disease, a comprehensive view of exposure is needed. The achievement of this goal will require improved sophisticated modeling capabilities to explain how components of the disease process respond when perturbed by environmental factors. Even in the simplest case of four combinations of genotype and exposure, where one is comparing risk of carrier versus noncarrier in exposed and nonexposed situations, there are several models for describing interactions between genetic susceptibility and environmental exposures. Of course, modeling of interactions gets more complex as the categories of environmental exposures and the number of variant alleles involved increase (Hunter, 2005). Also, new statistical methods may be needed to extract meaning from large datasets and to integrate information from many sources. Finally, the most critical technological challenge is how to relate exposure-disease association studies to pathways and mechanisms. To address this issue, NIEHS developed the National Center for Toxicogenomics (Lovett, 2000) to promote the development of technologies to monitor gene, protein, and metabolite expression in cells and tissues under various environmental conditions. To understand how genes and environmental factors interact to perturb biological pathways to cause injury or disease, scientists will need tools with the capacity to monitor the global expression of thousands of genes, proteins, and metabolites simultaneously. The generation of such data in multiple species can be used to identify conserved and functionally significant genes and pathways involved in gene–environment interactions.
POLYMORPHISM AND DISEASE SUSCEPTIBILITY: CASE–CONTROL STUDIES An article by Lower et al. (1979) transformed population health research and provided the foundation for the massive sequencing efforts to identify genetic variations involved in modulating human response to drugs and other environmental xenobiotics. This landmark investigation also demonstrated the power of hypothesis-driven, population-based studies for assessing disease risk associated with specific interactions between genes and the
54
CHAPTER 4
■
Genome and the Environment
environment. Furthermore, this pioneering study validated the use of epidemiological tools to investigate the biological significance of human genetic variants in disease risk. Such studies indicate that functional polymorphisms in genes that code for DNA repair or metabolism of environmental xenobiotics can modify the effects of environmental exposures. Gene variants can influence the balance of metabolic intermediates produced by altering the structure, function, and level of enzymes or the catalytic efficiency of the biotransformation process. Several large-scale projects are underway to build databases and to develop tools and infrastructure to investigate gene–gene and gene–environment interactions, and some results are already being reported. An encyclopedic review of the various studies is beyond scope of the present chapter, but a few examples will be presented as illustrative. Bladder Cancer and N-acetyltransferase 2 (NAT2) Polymorphism The seminal study by Lower and colleagues (1979) was prompted by their desire to elucidate the relationship between N-acetyltransferase phenotype and susceptibility to the development of bladder cancer from exposure to arylamines, as a result of exposure to cigarette smoke or chemical dyes. Earlier studies had suggested that (i) a significant portion of bladder cancer can be attributed to cigarette smoking and occupation exposures; (ii) the distribution of N-acetyltransferase activity in the liver is highly variable among individuals, and (iii) individuals with low enzyme activity (the so-called “slow acetylator phenotype”) were most susceptible to the development of bladder cancer (Case and Hasker, 1954; Cole, 1974; Hueper, 1969; Hueper et al., 1938). These observations led Lower et al. (1979) to propose that the slow acetylator phenotype would be over-represented in a population of bladder cancer patients occupationally exposed to arylamine dyes. Their hypothesis was confirmed in a cohort of bladder cancer patients who worked in the dye industry in Denmark. Bladder cancer has been described as a classic illustration of the principle of dose–effect modification of an environmental exposure by genetic polymorphisms (Kelada et al., 2003). Exposure to aromatic amines can cause bladder cancer depending on one’s NAT2 genotype/phenotype. Smokers who are slow acetylators have a higher risk of developing bladder cancer compared to smokers who are rapid acetylators. Marcus et al. (2000) calculated an odds ratio of 1.3–1.7 for smokers who are rapid acetylators, confirming that there is an interaction between NAT2 and smoking. Using nonsmoking rapid acetylators as the reference with estimates of the prevalence of smoking, these investigators were able to demonstrate that the populationattributable risk of the gene–environment interaction was 35% for slow acetylators who had ever smoked and 13% for rapid acetylators who had ever smoked. In laboratory studies designed to investigate plausible mechanisms to explain the results, Hein et al. (1993) and Mattano et al. (1989) demonstrated that NAT2 enzyme can bioactivate procarcinogenic aromatic amines in the bladder. Following N-oxidation of aromatic amines by cytochrome P4501A2 in the liver, O-acetylation of the resulting hydroxylamine by NAT2 can produce unstable acetoxy esters or
proximate carcinogens. The acetoxy ester intermediates can be N-acetylated or N-oxidized to yield N-hydroxy-N-acetyl aromatic amines that can bind to DNA to create point mutations or adducts. Myeloperoxidase Polymorphism and Lung Cancer Myeloperoxidase is a phase I enzyme found in the lysosomes of polymorphonuclear leukocytes and monocytes. It converts lipophilic carcinogens into hydrophilic forms. Exposure to tobacco smoke stimulates the recruitment of neutrophils into lung tissue with local release of myeloperoxidase. This enzyme converts 7,8-diol benzo(a)pyrene to the highly reactive and carcinogenic form called benzo(a)pyrene 7,8-diol epoxide, which can form DNA adducts in lung tissue. Myeloperoxidase can also convert other procarcinogens (e.g., polycyclic aromatic hydrocarbons and aromatic amines) in cigarette smoke to free radicals or reactive oxygen species. A common G to A transition at position 463 of the promoter region of the myeloperoxidase gene, which leads to loss of a SP1 transcription factor binding site in an Alu hormone responsive element has been shown to reduce myeloperoxidase mRNA expression (Kiyohara et al., 2005). Because transcriptional activity is decreased in individuals with the variant A allele, less enzyme is available to convert benzo(a)pyrene intermediate to the highly carcinogenic benzo(a)pyrene 7,8-diol epoxide species. The frequency of the A allele varies between Caucasian (23.4%) and Asian (14.4%) populations. As the myeloperoxidase G-463A variant allele is related to low metabolic activation, individuals with at least one variant allele may be associated with decreased risk of lung cancer. London et al. (1997) reported that Caucasian individuals with the A/A genotype were associated with a significant reduction in risk for lung cancer. Although the biological effects of each genotype (i.e., G/G, G/A and A/A) have not been clarified, lung cancer among the three genotypes decreases from G to A. Similar results have been obtained for Japanese and Hawaiian populations (LeMarchand et al., 2000). The association between myeloperoxidase genotype and lung cancer risk was modified by duration of smoking. In heavy smokers versus light smokers, the combined G/A and A/A genotypes had an odds ratio for squamous cell carcinoma of 6.22 for heavy smokers and 1.72 for light smokers, and for never smokers 1.14. Other studies showed that myeloperoxidase G/A genotype interacted with the glutathione S-transferase T1 (GSTT1) locus to significantly decrease lung cancer among males but not females (Cajas-Salizar et al., 2003; Feyler et al., 2002). NAD(P)H Quinine Oxidoreductose 1 Polymorphism and Lung Cancer Polymorphism in the gene that codes for NAD(P) quinine oxidoreductase 1 (NQO1) is also a lung cancer susceptibility gene that interacts with carcinogens in cigarette smoke (Kiyohara et al., 2005). NQO1 has a dual function in that it can both activate and detoxify carcinogenic quinoid compounds, such as benzo(a)pyrene. The catalytic reduction of benzo(a)pyrene prevents the formation of cancer-causing quinone-DNA adducts. The combined variant genotypes, Pro-Ser and Ser-Ser were
Polymorphism and Disease Susceptibility: Case–Control Studies
significantly associated with decreased risk of lung cancer in Mexican Americans (Wiencke et al., 1997). Similar results were obtained in three Japanese studies (cited in review article by Kiyohara et al., 2005); however, these genotypes were only weakly associated with lung cancer in Chinese, Hawaiians, Caucasians, and African-Americans. In terms of gene–environment interactions, there was an excess small cell lung cancer risk associated with the presence of the Ser allele in heavy smokers, where the odds ratio for the Ser-Ser and Pro-Ser genotypes combined was 12.5; in light smokers the odds ratio was 0.90. In contrast, the frequency of the NQO1 genotypes did not differ significantly between smokers and nonsmokers. The NQO1 Pro187Ser polymorphism was also involved in gene–gene interactions. For example, the combined NQO1 Pro-Pro and GSTT1 null genotype showed a significant association with lung adenocarcinoma risk, with an odds ratio of 4.61 (Sunaga et al., 2002). The odds ratio was calculated using the NQO1 Ser-Ser and GSTT1 nonnull genotypes combined as the reference. Although the risk of developing lung cancer conferred by the various susceptibility alleles is small, lung cancer is such a common malignancy that this small increase in risk translates to a large number of lung cancer cases in the population and is clearly important from a public health point of view. These results also show that lung cancer cannot be explained by variation in a single allele, but rather by multiple polymorphisms within genes involved in the same or different pathways. Discovery of other functional variants may shed some light on the development of this important malignancy, as the burden of this disease surely develops from complex interactions between multiple genetic and environmental factors over time. That is, the effect of polymorphisms may be best represented by their haplotypes. Fibrinogen Polymorphism and Coronary Heart Disease Fibrinogen is an acute phase protein and its plasma level is responsive to trauma, serum cholesterol, hormones, and infections. Increase in genetically determined levels of this glycoprotein is associated with risk of cardiovascular disease. Several epidemiologic studies have shown that gene–environment interactions involving fibrinogen polymorphisms play a role in coronary heart disease. The glycoprotein consists of three different polypeptide chains (a, b, and c) linked by disulfide bonds (Vischetti et al., 2002). Studies show that b-chain synthesis limits the rate of production of mature fibrinogen (Yu et al., 1984). Many polymorphisms have been discovered in the b-chain, two of which show a strong correlation with plasma concentration of the mature protein. The BclI and the G/A-455 polymorphisms have been selected for analysis of gene–environment interaction. Such studies have shown that moderate physical activity decreases the level of fibrinogen; however, these effects are evident only in carriers of the B2 allele of the BclI b-chain polymorphism. Montgomery et al. (1996) have shown that there is an acute rise in fibrinogen levels in response to exercise, and that this response is strongly influenced by the G/A-455 polymorphism in the b-chain gene; in fact, a gradient of increasing
■
55
acute fibrinogen response was seen across the G/G, G/A, and A/A genotypes, respectively. Zito et al. (1999) reported that infection with Helicobacter pylori, the pathogen responsible for most peptic ulcers and gastritis, resulted in a fourfold increase in risk of myocardial infarction; the response to infection was higher in carriers of the BclI polymorphism, with carriers of the B2 allele showing higher fibrinogen levels compared to the B1 homogygous subjects. Carriers of the B2 allele have increased risk for myocardial infarction and have the capacity to amplify the effect of seropositivity for Helicobacter. So patients with both seropositivity for Helicobacter and carriers of the B2 genotype have an additional increased risk. In another study, some subjects with high dietary fat intake developed abnormal high-density lipoprotein cholesterol concentrations, and some did not, depending on their genotype or polymorphisms in the hepatic lipase gene promoter (Ordovas and Corella, 2002; Ordovas 2004). Also, tobacco smokers developed coronary heart disease, or did not, depending on their lipoprotein lipase genotype (Talmud et al., 2000) and their apolipoprotein E4 genotype (Humphries et al., 2001). These results show that genetic polymorphisms in the b-chain gene do not cause myocardial infarction, but influence risk for the disease by rendering individuals more susceptible to environmental risk factors. Mental Disorders and Monoamine Oxidase A Polymorphism A functional polymorphism in the promoter region of the MAOA gene encoding the neurotransmitter metabolizing enzyme monoamine oxidase moderates the effects of child maltreatment on the cycle of violence (Moffitt et al., 2005). Their results show that maltreated children, with the polymorphism in the promoter region, produced low levels of monoamine oxidase A and were more prone to develop conduct disorders, to demonstrate antisocial personality, and to commit crimes as an adult than maltreated children with high-activity levels of monoamine oxidase A. In a related study, Caspi et al. (2003) demonstrated that the serotonin transporter (5-HTT ) gene moderated the effects of stressful life events on depression. Individuals with one or two copies of the 5-HTT polymorphic allele exhibited more depressive symptoms, diagnosable depression, and suicidal tendencies following stressful life events than individuals who were homozygous for the wild-type allele. Asthma and Gene Polymorphisms Asthma is a complex multifactorial disorder with variable phenotypes attributed to interaction of multiple genes with multiple environmental factors (see Chapter 89). The interplay between genes and the environment is best illustrated by the hygiene hypothesis to explain the increase in asthma prevalence. The fundamental tenet of this hypothesis is that the emergence and severity of the disease depends on the interaction between genetic background and environmental exposure. According to the hygiene hypothesis, children in industrialized nations have decreased incidence of bacterial infections, hence reduced exposure to endotoxins. In fact, this is supported by several studies
56
CHAPTER 4
■
Genome and the Environment
which show that exposure to increased endotoxin levels in the domestic environment is associated with decreased occurrence of asthma (for recent review see Kleeberger and Peden, 2005). The risk of developing asthma is influenced by the Western lifestyle, including diet, air pollution, reduced risk of infections, indoor air, and stress (reviewed by Arruda et al., 2005). The length and type of exposure, as well as the time when it occurs, can either promote or suppress the inflammatory process and the subsequent course of the disease. CD14 is an innate immunity gene that has been studied extensively with respect to its genetics, function, and distribution in various populations because of its important role in the development of asthma and other allergic reactions. The CD14 gene product is a component of the endotoxin-recognition receptor complex and is critical for the development of the optimal responses to low concentrations of the environmental stimulant via the Toll-like receptors (TLR4 and TLR2). By some unknown mechanism, both the membrane-bound and soluble forms of CD14 enhance the affinity of the interaction between environmental allergens and the TLRs. In 1999, Baldini et al., as part of the Tucson Children Respiratory Study, demonstrated an association between a SNP in the CD14 promoter (CD14*C-159T), increased serum levels of soluble CD14, and decreased total serum IgE. The discovery of an association between this polymorphism in the CD14 promoter and a reduction in serum IgE levels provided important support for the view that innate immunity genes are key determinants in the development of asthma. Several studies have shown that pro-oxidant environmental exposures are associated with increased incidence and exacerbation of asthma and other lung diseases. For example, a large study conducted in Southern California showed that in utero exposure to tobacco smoke was associated with several indicators of lung morbidity, including increased incidence of asthma diagnosis, increased need for emergency room use for airways symptoms, and increased occurrence of bronchitis (reviewed by Kleeberger and Peden, 2005). Numerous candidate genes have been shown to be associated with asthma and asthma-associated phenotypes, such as elevated IgE levels, bronchial hyperactivity, and SNPs within specific cytokines, IgE regulating enzymes, and phase I and phase II xenobiotic metabolism enzymes. However, only a few genes conferring significant risk have been discovered, and the functional significance of the various gene variants is largely unknown. A glutathione S-transferase M1 (GSTM1) null allele (M1*0), which results in no protein expression and no glutathione antioxidant protection, was shown to be strongly associated with decreased lung growth and function in a cohort of school children in Southern California due to increased oxidant stress (Gilliland et al., 2002a, b, c). These investigators also examined the relationship between the GSTM1 null genotype, the prevalence of asthma, and the in utero exposure to tobacco smoke (Gilliland et al., 2002a, b, c). Children with the GSTM1 null genotype and who were born to mothers who smoked during pregnancy showed increased prevalence of early onset asthma, persistent asthma, wheezing with exercise, and increased need for medical intervention for asthma. Children with the functional GSTM1 genotype did not have asthmatic symptoms; nor did the children
with GSTM1 null genotype if they were born to nonsmoking mothers. Studies have also shown that children with the GSTP1 Val 105/Val 105 genotype had fewer school absences related to asthma, which suggests that this antioxidant protective gene may influence asthma severity (Gilliland et al., 2002a, b, c). Both GSTM1 and GSTP1 appear to modulate the effect of exposure to particulates in diesel exhaust as measured by IgE response following a nasal challenge; both IgE and histamine levels were increased. In a study of the effects of diesel exhaust on response to ragweed challenge, subjects were stratified on the basis of GSTM1 and GSTP1 genotypes and were allowed to inhale diesel exhaust particles followed by ragweed allergen. Subjects with either the GSTM1 null or GSTP1 Ile 105 genotype exhibited a stronger response than did subjects with other GSTM1 and GSTP1 genotypes (Bastain et al., 2003; Gilliland et al., 2004). Studies of the risk for asthma from lifetime exposure to ozone show that GST proteins are important determinants of asthma severity (David et al., 2003). In studies of the combined genotypes of NQO1 Pro187Ser and GSTM1 null or functional sequence, the presence of the Pro-Pro allele of NQO1 significantly reduced asthma risk in subjects homozygous for the GST1 deletion. The studies on lung function and gene variation cited above are representative examples of published and ongoing studies; for a more comprehensive survey, the reader is referred to the excellent review by Kleeberger and Peden (2005) and to other chapters in this book.
EPIGENETICS AND THE ENVIRONMENT A discussion of gene–environment interactions could not be concluded without at least a brief mention of epigenetics, as it represents yet another mechanism by which the environment can influence phenotype or disease development (see Chapter 5). Cells of multicellular organisms are genetically the same, yet structurally and functionally diverse owing to the differential expression of genes. Many of these differences in gene expression arise during development, and most are stable throughout the life of the organism. However, it is important to note that epigenetic changes can occur in the adult. Such stable alterations are referred to as “epigenetic” in that they do not involve mutations in the DNA itself. Most of the research on epigenetic mechanisms has focused on DNA methylation during early development and gene silencing. But studies are now focusing on epigenetic regulation of gene expression response to environmental exposures as a possible pathway for disease development. For example, one could imagine that inactivation of proteins that buffer the cell against environmental stress could lead to disease by allowing the expression of previously suppressed genetic variants. Much attention has been devoted to diet as it represents an abundant source of methyl group donors. Also, diet is known to be a particularly significant determinant in the timing of disease expression. For example, dietary supplements such as folate or vitamins that influence the activity of methyl-transfer enzymes can modulate the rate of disease development and can exert
References
marked effects on the incidence of colon cancer (for review, see Van den Veyver, 2002). Reduced amounts of folate have been associated with genomic instability, neural tube defects, and genomic hypomethylation. In addition, a diet deficient in methylcontaining components has been shown to induce liver cancer. The possibility that classical genetic mechanisms may not be adequate to explain the development of complex diseases looms large on the horizon. Normal physiological responses to certain environmental stimuli may be mediated by epigenetic mechanisms, so progress is likely to require a more intense research effort to fully understand how humans and other organisms respond to environmental cues. For example, epigenetic changes might explain the phenotypic discordance often observed among monozygotic twins, differences in age of onset of diseases, fluctuation in diseases and sporadic cases of disease. Also, since epigenetic-mediated changes in gene expression can occur at a much faster rate than DNA mutation, it represents a more adaptable mechanism for organisms to respond to rapidly changing environmental conditions.
CONCLUSION Although most of the visible environmental problems of the 1950s and 1960s have been ameliorated, massive quantities of toxic agents are still polluting our environment. Whether current levels of exposure to these agents are contributing to the high or increasing incidence of cancer, Parkinson’s and Alzheimer’s disease, asthma, autism, learning disabilities, diabetes, or other complex disorders is a matter of considerable concern. Finding answers to these questions has been a slow and difficult process. The traditional methodologies available to
■
57
medical researchers have not been adequate to elucidate the intricate gene–environment interactions involved in the development of complex diseases. But, thanks to the rare confluence of technology breakthroughs and the rethinking and redirection of the environmental health sciences over the past decade, the link between the environment and human health and disease can now be investigated with more rigor and specificity. For example, the sequencing of the human genome and the development and application of high throughput technologies to monitor the expression of genes and proteins in response to specific environmental exposures has created an unparalleled opportunity to study gene–environment interactions. By using a combination of new “-omic” technologies (genomics, proteomics, and metabolomics), in combination with population studies and better exposure data, one can achieve an integrative view of gene–environment interaction at the level of the whole organism. So, perhaps the most important development in healthcare in the 21st century will be the incorporation of knowledge of gene–environment interactions into public health and the practice of medicine. Many of the expected breakthroughs from use of these technologies will almost definitely lead to new opportunities to prevent, diagnose, and treat human diseases.
ACKNOWLEDGEMENTS I am grateful to my colleague, Dr Samuel Wilson, for his intellectual input and leadership in developing many of the concepts and research initiatives on gene–environment interactions discussed in this chapter. I am also grateful to Ms Betty Mills for editorial and technical assistance.
REFERENCES Altshuler, D., Brooks, L.D. et al. (2005). A haplotype map of the human genome. Nature 437, 1299–1320. Arruda, L.K., Sole, D., Carlos, E. et al. (2005). Risk factors for asthma and atopy. Curr Opin Allergy Clin Immunol 5, 153–159. Baldini, M., Lohman, I.C., Halonen, M. et al. (1999). A polymorphism in the 5 flanking region of the CD14 gene is associated with circulating soluble CD14 levels and with total serum IgE. Am J Respir Cell Mol Biol 20, 976–983. Bastain, T.M., Gilliland, F.D., Li,Y.F. et al. (2003). Intra-individual reproducibility of nasal allergic responses to diesel exhaust particles indicates a susceptible phenotype. Clin Immunol 109, 130–136. Baulcombe, D. (2002). RNA silencing. Curr Biol 12, R82–R84. Boguski, M.S. (1999). Biosequence exegesis. Science 286, 453–455. Brenner, S. (2002). Detective Rummage Investigates. Scientist 16(16), 15. Brown, P.O. and Hartwell, L. (1998). Genomics and human diseases – variations on variation. Nat Genet 18, 91–93. Brown, P.O. and Botstein, D. (1999). Exploring the New World of the genome with DNA microarrays. Nat Genet 21, 33–37. Carlson, C.S., Eberle, M.A. et al. (2004). Selecting a maximally informative set of single-nucleotide polymorphisms for association analyses using linkage disequilibrium. Am J Human Genet 74, 106–120.
Carrington, M., Nelson, G.W., Martin, M.P. et al. (1999). Heterozygote advantage and B*35-Cw*04 disadvantage. Science 283, 1748–1752. Case, R.A.M. and Hasker, M.E. (1954). Tumors of the urinary in workmen engaged in the manufacture and use of certain dye stuff intermediates in the British chemical industry. Brit J Ind Med 11, 75. Cajas-Salizar, N., Sierra-Terres, C.H., Salama, S.A. et al. (2003). Combined effect of MPO, GSTM1 and GSTT1 polymorphisms on chromosome aberrations and lung cancer risk. Int J Hyg Environ Health 206, 473–483. Caspi, A., Sugden, K., Moffitt, T.E. et al. (2003). Influence of life stress on depression: Moderation by a polymorphism in the 5-HTT gene. Science 301, 386–389. Chamberlin, M. (1982). Bacterial DNA-dependent RNA polymerases. In The Enzymes, 3rd edition (P. Boyer, ed.) Vol. 15B. Academic Press, NY, pp. 61–108. Cole, P. (1974). Lower urinary tract, In Cancer Epidemiology and Prevention: Current Concepts (D. Schottenfield, ed.), Charles C. Thomas, Springfield, IL, pp. 233–262. Collins, F.S., Lonjou, C. and Morton, N.E. (1999). Genetic epidemiology of single-nucleotide polymorphisms. Proc Natl Acad Sci USA 96(26), 15173–15177.
58
CHAPTER 4
■
Genome and the Environment
Costa, L.G. and Furlong, C.E. (2002). Paraoxanase (PON1) in Health and Disease: Basic and Clinical Aspects. Kluwer Academic Publishers, Norwell, MA. David, G.L., Romieu, I., Sienra-Monge, J.J. et al. (2003). Nicotinamide adenine dinucleotide (phosphate) reduced: Quinine oxidoreductase and glutathione S-transferase M1 polymorphisms and childhood asthma. Am J Respir Crit Care Med 168(10), 1199–1204. Dean, M. (2003). Approaches to identify genes for complex human diseases: Lessons from Mendelian disorders. Hum Mutat 22, 261–274. Dean, M., Carrington, M., Winkler, C. et al. (1996). Genetic restriction of HIV-1 infection and progression of AIDS by a deletion allele of the CKR5 structural gene. Science 273(5283), 1856–1862. Dobzhansky, T. (1941). Genetics and the Origin of Species, 2nd edition. Columbia University Press, NY. Dobzhansky, T. and Levene, H. (1955). Genetics of natural populations. Genetics 40, 779–808. Doll, R. and Peto, R. (1981). The causes of cancer: Quantitative estimates of available risks of cancer in the United States today. J Natl Cancer Inst 66, 1191–1308. Feyler, A., Voho, A., Bouchardy, C. et al. (2002). Myeloperoxidase463G- a polymorphism and lung cancer risk. Cancer Epidemiol Biomarkers Prev 11, 1550–1554. Frye, R.F. (2004). Probing the world of cytochrome P450 enzymes. Mol Interv 4, 157–162. Futreal, P.A., Liu, Qingyun, et al. (1994). BRCA1 mutations in primary breast and ovarian carcinomas. Science 266, 120–122. Garrod, A. (1902). The incidence of alkatonuria: A study in chemical individuality. Lancet 2, 1616–1620. Gibson, G. and Wagner, G. (2000). Canalization in evolutionary genetics: a stabilizing theory? Bio Essays 22, 376–380. Gilliland, F.D., Li, Y.F., Dubeau, L. et al. (2002a). Effects of glutathione S-transferase M1, maternal smoking during pregnancy, and environmental tobacco smoke on asthma and wheezing in children. Am J Respir Crit Care Med 166, 457–463. Gilliland, F.D., Gouderman, W.J., Vora, H. et al. (2002b). Effects of glutathione S-transferase M1, T1 and P1 on childhood lung function growth. Am J Respir Crit Care Med 166, 710–716. Gilliland, F.D., Rappaport, E.B., Berhane, K. et al. (2002c). Effects of glutathione S-transferase P1, M1, and T1 on acute respiratory illness in school children. Am J Respir Crit Care Med 166, 346–351. Gilliland, F.D., Li, Y.F., Saxon, A. et al. (2004). Effect of glutathione S-transferase M1 and P1 genotypes on xenobiotic enhancement of allergic responses: Randomized, placebo-controlled crossover study. Lancet 363, 119–125. Guengerich, F.P. (1998). The environmental genome project: Functional analysis of polymorphisms. Environ Health Perspect 106, 365–368. Haldane, J.B.S. (1946). The interaction of nature and nurture. Ann Eugenics 13, 197–205. Hartman, J.L., Garvik, B. et al. (2001). Principles for the buffering of genetic variation. Science 291, 1001–1004. Hein, D.W., Doll, M.A., Rustan, T.D. et al. (1993). Metabolic activation and deactivation of arylamine carcinogens by recombinant human NAT1 and polymorphic NAT2 acetyltransferases. Carcinogenesis 14, 1633–1638. Hueper, W.C. (1969). Occupational and Environmental Cancer of the Urinary System.Yale University Press, New Haven, CT. p. 465. Hueper, W.C., Wiley, F.H. et al. (1938). Experimental production of bladder tumors in dogs by administration of beta-naphthylamine. J Ind Hyg Toxicol 20, 46.
Hulla, J.E., Miller, M.S. and Taylor, J.A. (1999). Symposium overview: The role of genetic polymorphisms and repair deficiencies in environmental disease. Toxicol Sci 47, 135–143. Humphries, S.E., Talmud, P.J., Hawe, E. et al. (2001). Apolipoprotein E4 and coronary heart disease in middle aged men who smoke: A prospective study. Lancet 358, 115–119. Hunter, D. (2005). Gene-environment interactions in human diseases. Nat Rev Genet 6, 287–298. Jarvik, G.P., Jampsa, R. et al. (2003). Novel paraoxonase: Nonsense and missense mutations predicted by functional genomic assay of status. Pharmacogenetics 13, 291–295. Kaiser, J. (1997). Environmental Institute lays plans for gene hunt. Science 298, 569–570. Kaiser, J. (2003). Tying genetic to the risk of environmental disease. Science 300, 563. Kelada, S.N., Eaton, D.L., Wang, S.S. et al. (2003). The role of genetic polymorphisms in environmental health. Environ Health Perspect 111, 1055–1064. Khot, U.N., Khot, M.B., Bajzer, C.T. et al. (2003). Prevalence of conventional risk factors in patients with coronary heart disease. JAMA 290, 899–904. Kiyohara, C., Yoshimasu, K., Takayama, K. et al. (2005). NQO1, MPO, and the risk of lung cancer: A huge review. Genet Med 7, 463–474. Kleeberger, S.R. and Peden, D. (2005). Gene-environment interactions in asthma and other respiratory diseases. Annu Rev Med 56, 383–400. Kouhry, M.J. and Wagner, D.K. (1995). Epidemiological evasuation of the use of genetics to improve the predictive value of disease risk factors. Am J Hum Genet 56, 835–844. Kraft, P. and Hunter, D. (2005). Integrating epidemiology and genetic association: The challenge of gene-environment interaction. Philos Trans R Soc Lond B Biol Sci 360, 1609–1616. Krajinovic, M., Sinnett, H. et al. (2002). Role of NQO1, MPO, and CYP2El genetic polymorphisms in the susceptibility of childhood acute lymphoblastic leukemia. Int J Cancer 97, 230–236. Kruglyak, L. and Nickerson, D.A. (2001). Variation is the spice of life. Nat Genet 27, 234–236. Lander, E.S., Linton, L.M., Birren, B. et al. (2001). Initial sequencing and analysis of the human genome. Nature 409, 860–921. LeMarchand, L., Seifriend, A., Lum, A. et al. (2000). Association of the myeloperoxidase-463G- a polymorphism with lung cancer risk. Cancer Epidemiol Biomarkers Prev 9, 181–184. Lichtenstein, P., Niels, V.H., Verkasala, P.K. et al. (2000). Environmental and heritable factors in the causation of cancer: Analysis of cohorts of twins from Sweden. N Eng J Med 343, 78–85. London, S.J., Lehman,T.A. and Taylor, J.A. (1997). Myeloperoxidase genetic polymorphism and lung cancer risk. Cancer Res 57, 5001–5003. Lovett, R.A. (2000). Toxicologists brace for genomics revolution. Science 289, 536–573. Lower, G.M., Nilsson, T. et al. (1979). N-acetyltransferase phenotype and risk of urinary bladder cancer: Approaches to molecular epidemiology. Preliminary results in Sweden and Denmark. Environ Health Perspect 29, 71–79. Marcus, P.M., Hayes, R.B., Vineis, P. et al. (2000). Cigarette smoking, N-acetyltransferase 2 acetylation status and bladder cancer risk: a case series meta-analysis of a gene-environment interaction. Cancer Epidemiol Biomarkers Prev 9, 461–467. Martin, M.P., Dean, M., Smith, M.W. et al. (1998). Genetic acceleration of AIDS progression by a promoter variant of CCR5. Science 282, 1907–1911. Mattano, S.S., Land, S., King, C.M. et al. (1989). Purification and biochemical characterization of hepatic arylamine-N-acetyltransferase
References
from rapid and slow acetylator mice: Identify with arylhydroxamic acid N,O-acetyltransferase. Mol Pharmacol 35, 599–609. Moffitt, T.E., Caspi, A. and Rutter, M. (2005). Strategy for investigating interactions between measured genes and measured environments. Arch Gen Psychiatry 62, 473–481. Montgomery, H.E., Clarkson, P., Nwose, O.M. et al. (1996). The acute rise in serum fibrinogen concentration with exercise is influenced by the G453-A polymorphism in the beta-fibrinogen gene. Arterioscler Thromb Vasc Biol 16, 386–391. Nebert, D.W., McKinnon, R.A. and Puga, A. (1996). Human drug metabolizing enzyme polymorphisms: Effects on toxicity and cancer. DNA Cell Biol 15, 273–280. Olden, K. (1993). Opportunities in environmental health research. Environ Health Perspect 101, 6–7. Olden, K. (2004a). Use of omic approaches in unraveling mechanisms of gene-environment interactions. Curr Genomics 5, 1–6. Olden, K. (2004b). Relevance of the National Institutes of Health Roadmap Initiative to the field of environmental health. Environ Health Perspect 112, 3. Olden, K. (2007). Commentary: From phenotype to genotype, to gene-environment interaction and risk for complex diseases. Int J Epidemiol 36, 11–18. Olden, K. and Guthrie, J. (2000). Perspective on environmental toxicant issues: Information gaps in basic environmental health sciences. Environ Epidemiol Toxicol 2, 172–178. Olden, K. and Wilson, S.H. (2000). Environmental health and genomics: Visions and implications. Nat Rev Genet 1, 149–153. Olden, K., Guthrie, J. and Newton, S.A. (2001). A bold new direction for environmental health research. Am J Public Health 91, 1964–1967. Ordovas, J.M. (2004). Pharmacogenetics of lipid diseases. Hum Genomics 2, 111–125. Ordovas, J.M. and Corella, D. (2002). Genetic variation and lipid metabolism: Modulation by dietary factors. Circulation 106, 2315–2321. Pearson, H. (2004). Mouse sequencing plan aims to boost models. Nature 432, 5. Powell, J.J., Vande Mater, J. and Gerschwin, M.E. (1999). Evidence for the role of environmental agents in the initiation and progression of autoimmune conditions. Environ Health Perspect 107, 667–672. Rothman, N.,Wacholder, S., Caporoso, M. et al. (2001). The use of common genetic polymorphisms to enhance the epidemiologic study of environmental carcinogens. Biochim Biophys Acta 1471, C1–C10. Scheiner, S.M., Caplan, R.L. et al. (1991). The genetics of phenotypic plasticity III: Genetic correlations and fluctuating asymmetries. J Evol Biol 4, 51–68. Schena, M., Shalon, D. et al. (1995). Protein quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470. Schwartz, D. and Collins, F. (2007). Environmental biology and human disease. Science 316, 695–696. Skibola, C.F., Smith, M.T. et al. (1999). Polymorphisms in the methylenetetrahydrofolate reductase gene are associated with susceptibility to acute leukemia in adults. Proc Natl Acad Sci USA 96, 12810–12815. Skibola, C.F., Smith, M.T. et al. (2002). Polymorphisms in the thymidylate synthetase and serine hydroxymethyltransferase genes and risk of adult acute lymphocytic leukemia. Blood 99, 3786–3791. Smith, M.W., Dean, M., Carrington, M. et al. (1997). Contrasting genetic influence of CCR2 and CCR5 variants on HIV-1 infection and disease progression. Science 277, 959–965. Smith, M.T., Wang, Y. et al. (2001). Low NAD(P): Quinone oxidoreductase activity is associated with increased risk of leukemia
■
59
with MLL translocations in infants and children. Blood 100, 4590–4593. Smith, M.T., Wang, Y., Skibola, C.F. et al. (2002). Low NAD(P)H: quinine oxidoreductase activity is associated with increased risk of leukemia with MLL translocations in infants and children. Blood 100, 4590–4593. Sunaga, N., Kohno, T., Yanagitami, N. et al. (2002). Contribution of the NQO1 and GSTT1 polymorphisms in lung adenocarcinoma susceptibility. Cancer Epidemiol Biomarkers Prev 11, 730–738. Talmud, P.J., Bujac, S., Hall, S. et al. (2000). Substitution of asparagine for aspartic acid at residue 9(D9N) of lipoprotein lipase markedly augments risk of coronary heart disease in male smokers. Atherosclerosis 149, 75–81. Tanner, C.M., Ohtman, R., Goldman, S.M. et al. (1999). Parkinson disease in twins: A etiologic study. JAMA 281, 341–346. Thompson, J.R., Gerald, P.F. et al. (2001). Maternal folate supplementation in pregnancy and protection against acute lymphoblastic leukemia in childhood: A case-control study. Lancet 358, 1935–1940. Trask, B.J. (1999). Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, pp. 303–313. Van den Veyver, I.B. (2002). Genetic effects of methylation diets. Annu Rev Nutr 22, 255–282. Venter, J.C., Adams, M.D., Myers, E.W. et al. (2001). The sequence of the human genome. Science 291, 1304–1351. Verkasala, P.K., Kopria, J., Kaskenuvua, M. et al. (1999). Genetic predisposition, environmental and cancer incidence: National Twin Study in Finland. Int J Cancer 83, 743–749. Vessel, E.S. (1991). Genetic and environmental actors causing variation in drug response. Mutat Res 247, 241–257. Vischetti, M., Zito, F., Donati, M.B. et al. (2002). Analysis of geneenvironment interaction in coronary heart disease: Fibrinogen polymorphisms as an example. Ital Heart J 3, 18–23. Wagner, G.P., Booth, G. et al. (1997). A population genetic theory of canalization. Evolution 51, 329–347. Waters, M., Boorman, G., Bushel, P. et al. (2003). Systems toxicology and the chemical effects of biological systems (CEBS) knowledge based. Environ Health Perspect Toxicogenomics 111, 15–28. Watson, J.D. and Crick, F.H.C. (1953). Genetic implications of the structure of deoxyribonucleic acid. Nature 171, 964–967. Watson, J.D., Tooze, J. and Kurtz, D.T. (1983). Recombinant DNA: a short course. W.H. Freeman, N.Y. Wiencke, J.K., Spitz, M.R., McMillan, A. et al. (1997). Lung cancer in Mexican-Americans and African Americans is associated with the wildtype genotype of the NAD(P) H-quinone oxidoreductase polymorphism. Cancer Epidemiol Biomarkers Prev 6, 87–92. Wilson, S.H. and Olden, K. (2004). The environmental genome project; phase I and beyond. Molecular Intervention 4, 147–156. Wright, S. (1932). The roles of nutrition, inbreeding, crossbreeding and selection in evolution. Proc. of the 6th International Congress of Genetics 11, 356–366. Yu, S., Sher, B. and Kudryk, B. (1984). Fibrinogen precursors. Order of assembly of fibrinogen chains. J Biol Chem 259, 10574–10581. Zito, F., Di Castelnuevo, A., D’Orazio, A. et al. (1999). Helicobacter pylori infection and the risk of myocardial infarction: Role of fibrinogen and its genetic control. Thromb Haemost 82, 14–18. http://egp.gs.washington.edu/welcome.html http://www.genome.utah.edu/genesnps http://www.niehs.nih.gov/research/supported/programs/egp/ http://ntp.nieh.nih.gov/
CHAPTER
5 Epigenomics and Its Implications for Medicine Moshe Szyf
INTRODUCTION The genome holds the entire information required to encode all the proteins expressed in an organism. However, expression of this information is programmed so that only a fraction of the potential repertoire of proteins is expressed in a given cell type. This programming is defined by the epigenome. The epigenome controls gene expression programs by defining the accessibility of transcription machineries, as well as other nuclear machineries, to genes (Razin, 1998). Programming of the epigenome occurs during embryonic development, whereby it establishes the diversity of gene expression patterns amongst organs, tissues, and cell types (Razin, 1998). These programs are stable and maintained through life. The epigenetic machinery is responsive to signaling pathways, which could be activated by different triggers. It has become clear recently that epigenetic programming could potentially change through life, and not just during embryogenesis, in response to environmental exposures such as nutrition, drugs, and social interactions (Meaney and Szyf, 2005). The epigenome thus mediates the effect of the environment on our genes. These effects could result in novel gene expression programs triggering new physiological, behavioral, or pathological states. Improper epigenetic programming might lead to consequences that could be similar to the effects of genetic mutations or deletions. For example, aberrant epigenetic programming could lead to activation of a gene that is normally silenced or
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 60
to inactivation of a gene that is normally active. An example is the aberrant silencing of tumor-suppressor genes in cancer, which could be mediated by either an inactivating mutation or by epigenetic silencing through DNA methylation (Baylin et al., 2001). Epigenetic differences amongst individuals can result in interindividual phenotypic differences with respect to behavior, physiology, and pathology and response to drugs similar to genetic polymorphisms. However, the extent to which epigenetic interindividual differences play a role in generating interindividual phenotypic differences is still unknown. In contrast to the genome, which is identical in different cell types and through life, the epigenome is dynamic and varies from cell type to cell type and from time point to time point in life. This diversity allows the epigenome to regulate the assortment of gene expression programs in a multicellular organism. The epigenome consists of chromatin and its modifications, as well as a covalent modification of cytosine rings found at the dinucleotide sequence CpG (CG) (Razin, 1998). The basic building block of chromatin is the nucleosome, which is formed of an octamer of histone proteins containing an H3-H4 tetramer flanked on either side with a H2A-H2B dimer (Finch et al., 1977). The N-terminal tails of these histones are extensively modified by methylation (Jenuwein, 2001), phosphorylation, acetylation (Wade et al., 1997), and ubiquitination (Shilatifard, 2006). Different histone variants also play a regulatory role (Henikoff et al., 2004). The pattern of histone modifications and
Copyright © 2009, Elsevier Inc. All rights reserved.
DNA Methylation Patterns
variants creates a “histone code” that delineates the parts of the genome expressed at a given point in time in a given cell type (Jenuwein and Allis, 2001). Specific transcription factors and transcription repressors recruit histone-modifying enzymes to specific genes and thus define the profile of histone modification around genes (Jenuwein and Allis, 2001). The best-investigated examples are histone acetyltransferases (HAT), which acetylate histone H3 and H4 tails at the lysine-9 (K9) residue, and histone deacetylases (HDAC), which deacetylate histone tails (Kuo and Allis, 1998). Histone acetylation is believed to be a predominant signal for an active chromatin configuration. Deacetylated histones signal inactive chromatin, chromatin associated with inactive genes. Many repressors and repressor complexes recruit HDACs to genes, thus causing their inactivation. Histone tail acetylation is believed to enhance the accessibility of a gene to the transcription machinery whereas deacetylated tails are tightly interacting with DNA and limit its accessibility to transcription factors (Kuo and Allis, 1998). Histone methylation at K9 of H3-histone signals inactivity and is determined by the recruitment of histone methyltransferases (HMTs) such as SUV3–9 to genes (Lachner et al., 2001). The heterochromatin-associated protein HP-1, which binds methylated histones and precipitates an inactive chromatin structure (Lachner et al., 2001), recognizes methylated histones. Recently described histone demethylases remove the methylation mark, causing either activation or repression of gene expression (Shi et al., 2004;Tsukada et al., 2006). Chromatin remodeling complexes, which are ATP dependent, alter the position of nucleosomes around the transcription initiation site and define its accessibility to the transcription machinery (Varga-Weisz and Becker, 2006). It is becoming clear now that there is an interrelationship between chromatin modification and chromatin remodeling. For example, the presence of BRG1 (the catalytic subunit of SWI/ SNF-related chromatin remodeling complexes) is required for histone acetylation and regulation of -globin expression during development (Bultman et al., 2005). The targeting of chromatin modifications and remodeling to discrete genetic loci is established by specific trans-acting factors, which might be responsive to cellular as well as extracellular signaling. Chromatin is dynamic and plastic and could respond by altered recruitment of the different histone modification enzymes in response to a variety of signals. In addition to chromatin, which is associated with DNA, the DNA molecule itself is chemically modified by methyl residues at the 5ⴕ position of the cytosine rings in the sequence CG in vertebrates (Razin, 1998) (see Chapter 11). What distinguishes DNA methylation in vertebrate genomes is the fact that not all CGs are methylated in any given cell type (Razin, 1998). Distinct CGs are methylated in different cell types, generating cell type– specific patterns of methylation. Thus, the DNA methylation pattern confers upon a genome itself its cell type identity (Razin, 1998). Since DNA methylation is part of the chemical structure of the DNA itself, it is more stable than other epigenetic marks and thus has extremely important diagnostic potential (Beck et al., 1999). It was originally believed that the DNA methylation
■
61
pattern is established during development and is then maintained faithfully through life by the maintenance DNA methyltransferase (DNMT) (Razin and Riggs, 1980). The DNA methylation reaction was believed to be irreversible; thus the only way methyl residues could be lost was through replication in the absence of DNMT by passive demethylation (Razin and Riggs, 1980). Recent data support the idea that similar to chromatin modification DNA methylation is also potentially reversible even in post-mitotic tissues (Weaver et al., 2004a, b). A hallmark of DNA methylation patterns is the correlation between chromatin and the DNA methylation pattern. Active chromatin is usually associated with unmethylated DNA while inactive chromatin is associated with methylated DNA (Razin, 1998). Current data suggest a bilateral relation between chromatin structure and DNA methylation. This interrelationship of chromatin modification state and DNA methylation has important implications for understanding of how the epigenome is generated and of its utility as a diagnostic and as a therapeutic target. The gene carries the basic unit of information, but this information has an impact only once it is properly programmed by the epigenome. Thus, the epigenome is a prime target for intervention in human diseases and other conditions that involve altered gene function. Genetic mutations and polymorphisms are irreversible. In difference from the genome, the epigenome is reversible. One could potentially alter epigenetic patterns by either targeting pharmacologically the different chromatin modifications and DNA methylation enzymes or affecting the environmental exposures and the signaling pathways mediating these effects. Epigenetic profiles have therefore important practical implications in prophylaxis, diagnosis, and therapeutics. A true realization of the potential of epigenetics will require a thorough understanding of the mechanisms leading to the formation of specific DNA methylation patterns and a comprehensive mapping of dynamic epigenomes with relation to different environmental exposures. This chapter will review the basic machineries involved in shaping the epigenome and the potential implications of epigenetics on understanding, diagnosing, treating, and preventing human pathologies.
DNA METHYLATION PATTERNS The DNA Methylation Pattern Is Cell Type Specific and Correlates with Chromatin Structure Modification of nucleotides in DNA by addition of a methyl group is widespread but not ubiquitous in biology. Both 5-methylcytosine and 6-methyladenine methylation are found in bacteria and are encoded by sequence-specific DNMTs. The methylated sequence distinguishes one strain from the other and serves for self versus non-self demarcation, protecting bacteria from invading DNA arriving from a different host. For this restriction modification system to function properly, it requires complete methylation of the host DNA in a unique 4–8 base sequence (Bestor, 1990; McClelland et al., 1994).
62
CHAPTER 5
■
Epigenomics and Genomic Medicine
DNA methyltransferase (DNMT)
HO HO
S-Adenosylmethionine (SAM)
—
CH N
N
N
CH2SCH2CH2CH(NH2)
—
O— —C
N
—
CH N
COOH
CH — CH3
COOH
—
—
O— —C
C
—
—
CH2SCH2CH2CH(NH2)
—
— N
—
N
CH
N
—
C
CH3
NH2
NH2
—
N
—
N
NH2
—
—
NH2
HO HO
S-Adenosylhomocystein (SAH)
Figure 5.1 The DNA methylation reaction. The DNMTs catalyze the transfer of a methyl group from the methyl donor S-adenosyl methionine onto the 5 in the cytosine ring in DNA.
DNA methylation is not found in all organisms. Certain yeast strains, including Saccharomyces cerevisiae, are devoid of DNA methylation. Since these organisms are eukaryotes with extensive epigenetic control in spite of lack of methylation, it suggests that the primary component of epigenetic control is chromatin, which is highly developed in yeast, and not DNA methylation, which is absent in yeast. Similarly, it was long believed that nematodes and insects are devoid of DNA methylation. However, recent data suggest that Drosophila flies do contain minute amounts of 5-methylcytosine, which is relatively abundant during early development (Lyko et al., 2000). In contrast to bacterial methylation systems, the primary methylated sequence in vertebrates is composed of only two bases, the dinucleotide sequence CG. In plants, CNG (where N is either A, G, or T) sequences are also methylated in addition to CGs (Razin and Riggs, 1980). Another feature that distinguishes bacterial from vertebrate methylation is the fact that only a fraction (80%) of the methylatable CG population is methylated in vertebrate DNA. Different CG sites are methylated in different tissues, creating a pattern of methylation that is gene and tissue specific (Razin and Riggs, 1980). In some cases, such as parentally imprinted genes (Swain et al., 1987) or genes residing on the inactive X chromosome (Mohandas et al., 1981), the two alleles of the same gene are differentially methylated in the same cell. DNA methylation is the only component of the covalent DNA structure that shows cell-, parent of origin- and allelespecific identity. An identical sequence could be either methylated or nonmethylated in different cell types or in different alleles in the same cell type. The DNA methylation pattern is not copied by the DNA replication machinery, but by independent enzymatic machinery the DNMTs (Razin and Cedar, 1977) (Figure 5.1). DNA methylation patterns in vertebrates are distinguished by their tight correlation with chromatin structure. Active regions of the chromatin, which enable gene expression, are associated with hypomethylated DNA whereas hypermethylated DNA is packaged in inactive chromatin (Razin and Cedar, 1977). It is generally accepted that DNA methylation plays an important role in regulating gene expression. Establishing and Maintaining DNA Methylation Patterns Vertebrates establish their DNA methylation patterns during development and these patterns are then maintained in each
cell type during life. Nevertheless, differences in environmental exposure result in differences in DNA methylation patterns that emerge later in life (Weaver et al., 2004a, b), even in monozygotic twins (Fraga et al., 2005). DNA methylation patterns might be affected by changes in the methyl content of diets, maternal behavior (Weaver et al., 2005), and exposures to certain drugs. Understanding the epigenetic drift during life and its environmental causes is now attracting considerable attention and could potentially develop into an important field in medicine. DNA Methyltransferases The methylation of DNA occurs immediately after replication by a transfer of a methyl moiety from the donor S-adenosyl-lmethionine (SAM) (AdoMet) in a reaction catalyzed by DNMTs (Figure 5.1). DNA methylation is only one of the methylation reactions occurring in the cell that uses AdoMet as a cofactor. The N-terminus of most cytosine 5-methyltransferases defines their unique features and regulatory functions which are responsible for specific targeting of DNMTs to their site of action. In mammalian DNMT1, the N-terminus contains a number of functional domains that integrate DNMT activity with chromatin-modifying enzymes such as histone deacetylase 1 (HDAC1) (Fuks et al., 2000) and HDAC2 (Robertson et al., 2000) and coordinate DNMT activity with the replication fork and DNA replication (Milutinovic et al., 2003). This obviously has therapeutic implications, since it might be possible to target these functions independently of DNA methylation and steer clear of the adverse effects associated with loss of DNA methylation (for a review see Szyf [2001]). De Novo and Maintenance Methyltransferase Activities It is critical to accurately replicate the methylation pattern during cell division. Razin and Riggs proposed more than two decades ago that the maintenance DNMT prefers a hemimethylated substrate (Razin and Riggs, 1980). Since hemimethylated sites are generated during DNA replication when a nascent unmethylated C is synthesized across a methylated C in the template parental strand, the DNMT accurately copies the methylation pattern of the template strand (Figure 5.2). Some DNMTs distinguish hemimethylated from nonmethylated sites. In support of the Razin and Riggs model, three distinct phylogenic DNMTs have been identified in mammals.
Chromatin Modification
De novo DNMT (DNMT3a, 3b, and DNMT1?)
Demethylase
DNA replication
Maintenance DNMT (DNMT1)
Figure 5.2 DNA DNMT and demethylases sculpt the methylation pattern. De novo DNMTs introduce new methyl groups into CG sites, which are unmethylated on both strands of DNA. Demethylases remove methyl groups from DNA. Maintenance DNMT copies the DNA methylation pattern from the parental strand into the nascent strand of DNA during DNA replication. Open circles indicate an unmethylated CG site in DNA; red circles indicate a methylated CG site.
DNMT1 shows preference for hemimethylated DNA in vitro, which is consistent with its role as a maintenance DNMT, whereas DNMT3a and DNMT3b methylate unmethylated and methylated DNA at an equal rate, which is consistent with a de novo DNMT role (Okano et al., 1998). Two additional DNMT homologs have been found: DNMT2, whose substrate and methylation activity is unclear (Vilain et al., 1998), and DNMT3L, which belongs to the DNMT3 family of DNMTs by virtue of its sequence and is essential for the establishment of maternal genomic imprints but lacks key methyltransferase motifs, and is possibly a regulator of methylation rather than an enzyme that methylates DNA (Bourc’his et al., 2001). Knockout mouse data indicate that DNMT1 is responsible for a majority of DNA methylation marks in the mouse genome (Li et al., 1992), whereas DNMT3a and DNMT3b are responsible for some but not all de novo methylation during development (Okano et al., 1999). It has recently been demonstrated that an additional factor is required for targeting DNMT1 to newly replicating hemimethylated DNA, the protein UHRF1 (ubiquitin-like, containing PHD and RING finger domains 1), also known as NP95 in mouse and ICBP90 in human (Bostick et al., 2007). Several lines of evidence indicate that DNMTs are targeted to specific sequences by sequence-specific factors, which recognize specific sequences of DNA. For example, the HMT EZH2 or the oncoprotein PML-RAR target DNMTs to specific sequences in DNA (Di Croce et al., 2002;Vire et al., 2006). Thus, in addition to template-directed DNA methylation in replicating DNA, specific factors are responsible for specific methylation in specific genes. Is DNA Methylation a Reversible Reaction? The most controversial issue in the DNA methylation field is the question of whether the DNA methylation reaction is reversible. A passive process through replication in the absence of DNMT
■
63
would result in loss of methylation. This could serve as a mechanism for both site specific and global demethylation (see Figure 5.2 for a scheme of the different DNA methylation reactions). There is a long list of data from both cell culture and early mouse development supporting the hypothesis that active methylation occurs in embryonal and somatic cells. There are now convincing examples of active, replication-independent DNA demethylation during development as well as in somatic tissues. Active demethylation was reported for the myosin gene in differentiating myoblast cells (Lucarelli et al., 2001), the INTERLEUKIN-2 gene upon T-cell activation (Bruniquel and Schwartz, 2003), the INTERFERON- gene upon antigen exposure of memory CD8 T cells (Kersh et al., 2006), and the glucocorticoid receptor gene promoter in adult rat brains upon treatment with the HDAC inhibitor trichostatin A (TSA) (Weaver et al., 2004a, b). Enzymatic Activities Responsible for Demethylation The characteristics of the enzymes responsible for active demethylation are controversial. One proposal has been that a G/T mismatch repair glycosylase also functions as a 5-methylcytosine DNA glycosylase, recognizing methylcytosines and cleaving the bond between the sugar and the base. The abasic site is then repaired and replaced with a nonmethylated cytosine resulting in demethylation (Jost, 1993). An additional protein with a similar activity was recently identified, the methylated DNA binding domain protein 4 (MBD4) (Zhu et al., 2000). Another report has proposed that MBD2 has demethylase activity. MBD2b (a shorter isoform of MBD2) was shown to directly remove the methyl group from methylated cytosine in methylated CGs (Bhattacharya et al., 1999). This enzyme was therefore proposed to reverse the DNA methylation reaction. However, other groups disputed this finding (Ng et al., 1999). Understanding the mechanisms responsible for demethylation is one of the most important questions in the field.
CHROMATIN MODIFICATION Modifications of Histones with Implications for Gene Expression Programming: The Histone Code The basic unit of the chromatin is the nucleosome, which is a dimer of two identical complexes, each consisting of four histone proteins: H2A, H2B, H3, and H4 (Pruss et al., 1995). The N-terminal tails of histones H3 and H4 are highly charged and are tightly associated with DNA. Chromatin had been viewed in the past as a static entity, which packaged DNA in a condensed form and maintained its integrity. However, a vast body of literature has established that histones undergo diverse covalent modifications, which include acetylation, methylation, phosphorylation, ubiquitination, and sumoylation (Strahl and Allis, 2000). Histone acetylation of H3 and H4 histone N-terminal tails is believed to be a predominant signal for active chromatin by enhancing the accessibility of the transcription machinery. Histone methylation at the K9 of histone H3 is repressive and results in exclusion of acetylation at the same residue, which is
64
CHAPTER 5
■
Epigenomics and Genomic Medicine
required for activation of gene expression. H3-histone K4 methylation is associated, on the other hand, with gene activity and transcription elongation (Santos-Rosa et al., 2002). These modifications are believed to form a “histone code” that regulates chromatin function by affecting the structural dynamics of the nucleosome (Jenuwein and Allis, 2001). Proteins That Interpret the Histone Code Histone modifications regulate chromatin function by either altering the accessibility of DNA to different trans-acting factors or recruiting specific proteins that recognize a single or conformational set of modifications (Strahl and Allis, 2000). Examples of proteins whose interaction with DNA is defined by different histone modifications include bromodomain-containing proteins, which interact with acetylated histones (Filetici et al., 2001), and chromodomain-containing proteins, which interact with methylated lysines (Lachner and Jenuwein, 2002). A classic example is the HP-1, which recognizes histone H3 methylated at K9 and is believed to be involved in gene silencing in heterochromatin (Lachner and Jenuwein, 2002). Different histone variants also play regulatory roles (Henikoff et al., 2004; Pusarla and Bhargava, 2005; Sarma and Reinberg, 2005; Zhang et al., 2005). Remodeling of chromatin is also critical for gene function and involves ATPasecontaining complexes such as SWI/SNF, which propels energydependent nucleosomal movement required for opening up of chromatin around transcription initiation regions (Gibbons, 2005; Muchardt and Yaniv, 1999;Wolffe and Hayes, 1999). The State of Modification of Histones Is a Dynamic Balance of Modifying and Demodifying Enzymatic Activities Histone acetylation is catalyzed by HATs, which transfer an acetyl group from the cofactor acetyl CoA onto the -position on lysine, and reversed by HDAC (Kuo and Allis, 1998). Two phylogenetic superfamilies of HAT are known to date: the GCN5 N-acetyltransferases related family GNAT and the MYST family (Verdone et al., 2005). PCAF is a mammalian transcriptional coactivator, which was discovered based on its homology to GCN5-HAT. PCAF acts as a transcriptional coactivator in many processes and is known to interact with the CREB binding protein (CBP) and its homolog p300. P300 and CBP are ubiquitously expressed transcription coactivators. The MYST superfamily includes the human acetyltransferases MOZ and Tip60 (Borrow et al., 1996). Tip60 is involved in histone acetylation mostly on histone H4 in response to DNA damage and transcriptional activation of genes involved in repair. MOZ is involved in acute myeloid leukemia as suggested by the fusion of MOZ and CBP found in these leukemias, which results in mistargeting of the fused protein causing aberrant activation of gene expression (Borrow et al., 1996). There are four distinct phylogenetic classes of HDACs known to date (see Holbert and Marmorstein [2005] for a review). Class 1 includes HDACs 1–3, 8; class 2 includes the human HDACs 4–7, 9, 10; class 3 includes the sirtuins; and class
4 includes recently identified HDAC 11. Class 1 and 2 HDACs were shown to be involved in malignancies in humans and are the target of the HDAC inhibitors TSA and SAHA. Alterations in class 3 HDACs or sirtuins were shown to be associated with age-associated diseases such as type II diabetes, obesity, and neurodegenerative disorders. The mechanism of action of class 1 and 2 HDACs seems to involve a mode of catalysis in which the bound zinc ion mediates the nucleophilic attack of a water molecule on the acetylated lysine substrate whereas the class 3 HDACs require the presence of the oxidized form of the cofactor nicotinamide adenine dinucleotide (NAD+) (Holbert and Marmorstein, 2005). Methylation marks on histones could either activate or silence gene expression as discussed above. Arginine and lysine HMTs catalyze the transfer of a methyl group from the methyl donor SAM to either arginines or lysines. In some cases, a single lysine can be mono-, di-, or tri-methylated, with different functional consequences for each of the three forms. Mammalian homologs of Suv3–9 that methylate lysine-9 on the tail of H3-histones were the first HKMT (histone lysine methyltransferases) to be identified (Rea et al., 2000). SET-containing HKMT methylates Lys-4, -9, -27, or -36 of histone H3 and lys-20 of histone H4. There are two distinct families of lysine methyltransferases SET which contain a SET domain of approximately 130 amino acids and Dot1p (Peters and Schubeler, 2005). All arginine methyltransferases are members of the PRMT family (for a review, see Cheng et al. [2005]). PRMT1 and PRMT4/CARM1 methylate histones H3, H4, and H2B. Although it was long believed that histone methylation is irreversible, histone demethylases have been identified that can remove either repressive or activating histone methylation marks (Metzger et al., 2005; Shi et al., 2004). Two classes of histone demethylases were recently discovered: FAD-dependent amine monooxygenases, which demethylate histones via an oxidation reaction releasing formaldehyde as the leaving group; and JmjC domain demethylases, which also release formadelhyde but require -ketoglutarate and Fe(II). Lysine-specific demethylase-1 (LSD1) is a FAD-dependent enzyme and demethylates monoand di-methyl H3 K4 (for a review see Holbert and Marmorstein [2005]). Since K4 methylation marks histones associated with active genes, it was anticipated that demethylation of this mark would have a repressive effect on gene expression. In agreement with this prediction, LSD1 was found in a complex with the CoREST complex, which represses neuronal genes in nonneuronal cells (Lee et al., 2005). A representative of the second class of histone demethylases, JHDM1 (JmjC domain-containing histone demethylase-1), which specifically demethylates histone H3 at lysine 36 (H3-K36), was recently discovered (Tsukadaet al., 2006). H3-K36 is probably involved in promoting transcription elongation and thus maintenance of a transcriptional active state (Peters and Schubeler, 2005). Since other methylation sites are known to exist in histones, it is anticipated that more demethylases with different specificities will be discovered in the coming years. In summary, the histone modification state is defined by a dynamic equilibrium of modifying and demodifying enzymes.
DNA Methylation and Chromatin States Cooperatively Determine the State of Activity of Genes
■
65
Transcription factor
X CH3
Ac
Transcription factor
Ac
CH3
Ac
Ac Sin3A
HDCA
MECP2
CH3
H3
CH3
CH3
SUV39 HP1
Figure 5.3 DNA methylation silences genes by two mechanisms. The first mechanism is direct (upper right). A methyl group in the recognition element for a transcription factor inhibits the binding of the transcription factor to the promoter resulting in inhibition of transcription. The second mechanism (bottom right) involves recruitment of MBDs which in turn recruit co-repressors (Sin3A), HDAC, HMTs (Suv39), and HP-1, a protein that binds K9 methylated H3-histones. This results in formation of an inactive chromatin structure and silencing of transcription.
The direction of the equilibrium is responsive to cellular signaling pathways which activate transcriptional activators or repressors. Transcription factors recruit HATs and K4-HMTs, while transcriptional repressors recruit HDACs, K9-HMTs, and K4-histone demethylases to suppress gene expression. The relative balance of activators and repressors thus defines the state of modification of a gene. This dependence on targeting of histone-modifying enzymes by sequence-specific factors explains the gene selectivity of the histone modification state (Figures 5.3, 5.4 and 5.5).
A second mechanism involves the recruitment of methyl-domain binding proteins such as MeCP2 to methylated genes (Nan et al., 1997). All these proteins share a methyl-binding domain, which recognizes methylated CG dinucleotides in regulatory regions of genes. MBD3 does not bind methylated DNA, but is probably recruited to methylated genes through its interactions with MBD2 (Zhang et al., 1999). MeCP2 recruits a repressive multiprotein complex, which contains the co-repressor Sin3A and histonemodifying enzymes such as HDAC1 and the HMT Suv3–9 (Fuks et al., 2003). Thus, methylation of DNA can lead to chromatin modification, which creates conditions for gene silencing.
DNA METHYLATION AND CHROMATIN STATES COOPERATIVELY DETERMINE THE STATE OF ACTIVITY OF GENES
DNMTs Form Protein–Protein Interactions with Chromatin-Modifying Enzymes There is a bilateral relation between chromatin and DNA methylation. As much as DNA methylation targets chromatinmodifying enzymes to genes, chromatin-modifying enzymes recruit DNA methylating enzymes (Figures 5.4 and 5.5). The mechanisms by which chromatin directs DNA methylation is via protein–protein interaction, through recruitment of DNMTs to specific genes by chromatin-modifying enzymes. A growing list of histone-modifying enzymes has been shown to interact with DNMT1, such as HDAC1 and HDAC2, the HMTs SUV3–9 and EZH2, a member of the multiprotein Polycomb complex PRC2, which methylates H3-histone at the
DNA Methylation Leads to Chromatin Inactivation DNA methylation can mark gene for silencing by two mechanisms (Figure 5.3). First, DNA methylation in a recognition element of a transcription factor such as MYC (Prendergast and Ziff, 1991) interferes with its binding. By blocking the binding of these transcription factors, recruitment of acetyltransferases is inhibited, probably resulting in a hypoacetylated chromatin structure, which is inaccessible to the transcription machinery.
66
CHAPTER 5
■
Epigenomics and Genomic Medicine
K27 residue (Fuks et al., 2000, 2003; Rountree et al., 2000;Vire et al., 2006). DNMT3a was recently also shown to interact with EZH2, which targets the DNA methylation-histone modification multiprotein complexes to specific sequences in DNA (Vire et al., 2006). The targeting of DNMTs is required not merely for initiating de novo methylation but also for the safeguarding of this pattern. Trans-acting repressors also target both histonemodifying enzymes and DNMTs to specific cis-acting signals. For example, the promyelocytic leukemia fusion protein PMLRAR engages HDACs and DNMTs to its target binding sequences and produces de novo DNA methylation of adjacent genes (Di Croce et al., 2002). There are also documented interactions between proteins, which read the DNA methylation and histone methylation marks and either histone or DNA modifying enzymes. The MBD MeCP2 interacts with the HMT SUV3–9 (Fuks et al., 2003), and in plants it was shown that Chromomethylase3 (CMT3), a plant CNG-specific DNMT, interacts with an Arabidopsis homolog of HP1, a protein that binds histone H3 methylated at Lysine-9 (Jackson et al., 2002). In summary, DNA methylation and histone modification are dynamically interrelated through protein–protein interactions between DNMT and MBDs, on the one hand, and histone-modifying enzymes and modified-histone binding proteins, on the other (see Figures 5.4 and 5.5 for a scheme). Chromatin Activation Leads to DNA Demethylation The relationship between DNA methylation and chromatin is bilateral in both directions of the DNA methylation equilibrium. Pharmacological histone acetylation using HDAC inhibitors leads to replication-independent demethylation in human cell lines (Cervoni and Szyf, 2001; Szyf et al., 1985) as well as in Neurospora crassa (Selker, 1998). TSA induces demethylation in non-replicating cells in the hippocampus of adult rats, illustrating the reversibility of the DNA methylation pattern even in somatic tissues (Weaver et al., 2004a, b). The relationship between histone acetylation and DNA demethylation could explain gene-targeted
demethylation (Figure 5.4). Transcription factors target HATs to specific genes, causing gene-specific acetylation, and thus facilitate their demethylation. For example, the intronic kappa chain enhancer and the transcription factor NF-kappaB are required for B-cell–specific demethylation of the kappa gene (Lichtenstein et al., 1994). The maize Suppressor-mutator (Spm) transposon demethylation is mediated by the transposon-encoded transcriptional activator TnpA protein (Bruniquel and Schwartz, 2003). The mechanism by which histone acetylation induces demethylation is still unknown. Increase in histone acetylation through triggering of cellular signaling pathways by environmental signals is a possible route for environmental signals to induce changes in the DNA methylation pattern. The partial overlap between HDAC inhibitors and DNA methylation inhibitors has to be considered when one uses these agents either therapeutically or for research purposes (see Figure 5.5 for a summary of the relationship between DNA and chromatin modifications and gene expression).
DNMTs Are Involved in Cellular Functions Independent of Their DNMT Activity Recent data suggest that DNMT proteins encode functional domains that enable them to interact with other cellular machineries, and thus DNMTs might exert their biological effects by methylation-independent mechanisms. DNMT1 was shown to have a number of distinct functional domains: a fork-targeting domain that targets it to the replication fork (Leonhardt et al., 1992), a stretch of amino acids that bind the replication fork protein PCNA (proliferating cell nuclear antigen) (Chuang et al., 1997), and domains that mediate the binding of DNMT1 to HDAC1 and 2 as well as the tumor-suppressor gene retinoblastoma (Rb) protein (Robertson et al., 2000). DNMT1 inhibitors were shown to induce expression of tumor-suppressor genes in a methylation-independent manner (Milutinovic et al., 2004). This possibility must also be considered when interpreting data from DNA methylation inhibition experiments.
DNMT Transcription repressor
HAT
dem
eth
CH3 CH3
CH3
Ac
TSA
Ac
Ac Ac Ac
HDAC
ylas
e
Transcription factor
HAT binding
CH3
CH3
Ac
Ac
X
demethylase
Figure 5.4 Gene activation by histone acetylation and DNA demethylation. An inactive gene is methylated and is associated with HDAC, DNMT, and hypoacetylated histones. Upon either pharmacological inhibition of HDACs with TSA or recruitment of HATs to the gene, the histones are acetylated. The “open” chromatin facilitates interaction of demethylases with the gene and DNA demethylation.
Epigenetics and Human Disease
MBD2AS SAM
demethylase
Ac
Ac
K9
K9 M
TR+ HDAC
■
67
INHAT M
DNMT
HAT
TSA
K9-
K9M
M
X
X
TA+ Histone dMTase?
Histone MTase K9-M
demethylase
K9-M M
X
DNMT
MG98 5-azaC
M
X
Figure 5.5 Epigenetic cycle. A gene could be found in different epigenetic states that are a balance of modification and demodification reactions. Methylation of DNA and chromatin modifications are interrelated. Ac, histone acetylation; horizontal arrow indicates transcription; empty circle, CG site in DNA; M in circle indicates methylated CG site; K9-M, methyl-K9-H3-histone; TR, transcriptional repressor; histone MTase, histone methyltransferase; HAT, histone acetyltransferase; INHAT, inhibitors of histone acetyltransferase; histone dMTase, histone demethylase; TA, transcription activator; MBD2 AS, antisense oligonucleotide inhibitor of MBD2 a putative demethylase; SAM, S-adenosylmethionine; MG98, antisense oligonucleotide inhibitor of DNMT; 5-azaC, a nucleoside inhibitor of DNA methylation.
EPIGENETICS AND HUMAN DISEASE Aberrant DNA Methylation and Chromatin Modification States in Cancer Aberrant epigenetic markings could have the same effect as genetic mutations, and thus they are candidates to be involved in any human disease caused by aberrant gene expression (for recent reviews see Feinberg [2007] and Gosden and Feinberg [2007]). Since epigenetic marks are potentially responsive to extracellular signals, they are candidates to mediate gene–environment interactions. Epigenetic changes are especially attractive candidates to mediate age-related disease, such as cancer, schizophrenia, type 2 diabetes, and asthma. Although the involvement of epigenetic changes in most disease is at this point in time still speculative, there are extensive data implicating DNA methylation and chromatin modification changes in cancer. The alterations in DNA methylation in cancer include a global loss in methylation concurrently with regional hypermethylation of specific genes (Feinberg and Vogelstein, 1983; Feinberg et al., 1988; Lu et al., 1983). Cancer progression involves concurrent silencing of genes such
as tumor-suppressor genes that block cancer growth, as well as activation of genes that promote metastasis such as certain proteases; the changes in DNA methylation appear to correspond with these changes in gene expression programming. Loss of methylation was observed in repetitive elements such as LINE1 (Chalitchagorn et al., 2004), in genes that promote the cancer state such as oncogenes (Feinberg and Vogelstein, 1983), as well as in pro-metastatic genes such as S100P (Sugimoto et al., 2004), S100A4 (Rosty et al., 2002), and uPA (Guo et al., 2002). Loss of methylation is associated with increased invasiveness in vitro (Figure 5.6). The regional hypermethylation observed in cancer is associated with changes in chromatin state and, as a consequence, with gene silencing. There is an increasing list of genes affected by hypermethylation that includes genes encoding tumor suppressors such as p16, RB, and BRCA1; adhesion molecules such as E CADHERIN; repair enzymes such as MGMT; and inhibitors of metastasis such as the different TIMPs (for review on tumorsuppressor gene methylation see Baylin et al. [2001]). The assortment of hypermethylated genes varies in different cancer types and at different cancer stages, creating a methylation profile that is unique for each cancer cell type. These methylation
68
CHAPTER 5
■
Epigenomics and Genomic Medicine
DNMT
DNA methylation
Demethylation
M M Tumor-suppressor genes
Metastasis genes
X Figure 5.6 DNA methylation changes in cancer. In cancer, tumor-suppressor genes are methylated and silenced, and metastatic genes are unmethylated and activated.
profiles are gaining increased importance in cancer diagnosis (Novik et al., 2002). In addition to changes in DNA methylation, the abundance of components of the DNA methylation machinery is altered in cancer. DNMT1, for example, was shown to be upregulated in many cancers by different oncogenic signaling pathways (Szyf, 2001). It is unclear whether the upregulation of DNMT1 is responsible for the hypermethylation of tumor-suppressor genes or whether aberrant targeting of DNNMT1 to specific genes by specific trans-acting factors such as the fusion protein PML-RAR causes gene-specific hypermethylation (Di Croce et al., 2002). DNMT1 might be promoting tumorigenesis by DNA methylationindependent mechanisms through protein–protein interactions with critical cellular regulatory proteins such as Rb, HDAC1, and PCNA as discussed above (Chuang et al., 1997; Milutinovic et al., 2004; Robertson et al., 2000). DNA Methylation and Late-Onset Human Disease Although cancer is the most well-established case for epigenetically driven human disease, there is some evidence suggesting involvement of epigenetic processes in other diseases. It stands to reason that epigenetic drift would lead to altered gene expression programming, which would lead to a pathological state. Good candidates are autoimmune diseases and inflammatory states such as asthma. It is by now well established that alteration in DNA methylation plays a role in the autoimmune disease Systemic Lupus Erythematosus (SLE) (Richardson, 2002). Drugs known to induce lupus such as procainamide and hydralazine induce also demethylation (Cornacchia et al., 1988; Deng et al., 2003; Scheinbart et al., 1991) and the demethylating agent 5-azaC induces lupus (Quddus et al., 1993). The genome of T cells from SLE patients is demethylated in comparison with T cells from control subjects (Balada et al., 2007). It is proposed that demethylation activates genes whose expression confers upon T cells self-reactivity. Interestingly, MBD2, which was implicated
in demethylation, is upregulated in T cells from Lupus patients (Balada et al., 2007). The involvement of DNA methylation in asthma was proposed (Kaminsky et al., 2006), and promoters of genes that encode cytokines involved in asthma were shown to be regulated by DNA methylation (Santangelo et al., 2002), but there is no evidence yet for aberrant DNA methylation in asthma. Similarly, it has been speculated that DNA methylation disregulation is involved in cardiovascular diseases (Corwin, 2004). There are some anecdotal data indicating involvement of DNA methylation in cardiovascular pathologies. Global hypermethylation was noted in leukocytes from chronic kidney disease, and hypermethylation was shown to associate with inflammation and increased mortality (Stenvinkel et al., 2007). There is some evidence for epigenetic dysregulation of estrogen receptor beta in atherosclerosis and vascular aging (Kim et al., 2007), and global hypomethylation was noted in atherosclerotic lesions (Hiltunen et al., 2002). These data justify further studies into the possible epigenetic basis of cardiovascular disease. If indeed epigenetic aberrations are found to be involved in cardiovascular diseases, this might result in new diagnostic, prophylactic, and therapeutic approaches. DNA methylation has also been proposed to be involved in insulin resistance and type 2 diabetes (Devaskar and Thamotharan, 2007). It is known that nutritional restriction during gestation is associated with development of insulin resistance and type 2 diabetes later in life (Devaskar and Thamotharan, 2007; Simmons, 2007). Uteroplacental insufficiency also causes global changes in methylation and histone acetylation in brain and liver in rodents (Ke et al., 2006; MacLennan et al., 2004). Although these studies did not establish a causal relationship between hypomethylation and type 2 diabetes, they are suggestive. Epigenetic mechanisms could provide a molecular link between early exposures and late onset of type 2 diabetes. The advent of techniques to perform whole-epigenome mapping will allow us to directly test the attractive hypothesis that epigenetic aberrations driven by different environmental exposures cause epigenetic aberrations that can lead to lateonset diseases such as type 2 diabetes or cardiovascular disease. There is emerging evidence that aberrations in DNA methylation are involved in mental health. Mutations in the MBD MECP2 lead to Rett syndrome (Amir et al., 1999; Wan et al., 1999), one of the most common causes of mental retardation in females. There are some data indicating aberrant methylation in late-onset mental pathologies, although it is unclear whether these changes in DNA methylation originated during embryogenesis or later in life as a response to an environmental exposure. The gene encoding REELIN, a protein involved in neuronal development and synaptogenesis, which is implicated in long-term memory, was found to be methylated in brains of schizophrenia patients. The methylation of REELIN was correlated with its reduced expression and increased DNMT1 expression in GABAergic neurons in the prefrontal cortex (Chen et al., 2002; Costa et al., 2002, 2003; Grayson et al., 2005; Veldic et al., 2007). Our preliminary data demonstrate hypermethylation of the GLUCOCORTICOID RECEPTOR exon 1f promoter and
Conclusions
its reduced expression in hippocampi of subjects who committed suicide, relative to age-matched control subjects (McGowan et al., unpublished data). The analysis of epigenetic aberration in mental disease is an active area of research, which is bound to produce exciting data in the coming years. Therapeutic Implications of Aberrant DNA Methylation Patterns in Cancer Epigenetic states, in contrast to genetic mutations, are potentially reversible by pharmacological agents (Figure 5.5). Thus, this field carries much promise for future novel therapeutics and prophylactics. A number of clinical trials are underway with catalytic inhibitors of DNMT1, antisense inhibitors of DNMT1, and HDAC inhibitors in different cancers, with the goal of activating silenced tumor-suppressor genes (Goffin and Eisenhauer, 2002; Szyf, 2003). It is anticipated that, as our understanding of other histone modifications is improved, inhibitors of HKMT and DNA demethylases will be used in cancer therapy as well. An important objective is to develop isotypic specific inhibitors, which would target the isotypes relevant to cancer and limit adverse effects. A critical issue here is to activate tumor-suppressor genes without activating pro-metastatic genes (Figure 5.6) and to identify the specific chromatin modification and DNA modification enzymes critical for tumor-suppressor genes. DNMT1 might promote cancer through DNA methylationindependent mechanisms in addition to a methylationdependent process (Szyf, 2001). Demethylation of DNA could trigger the induction of pro-metastatic genes as discussed above; it might be therefore ill advised to use catalytic inhibitors of DNMT1. A preferred approach is to target the methylationindependent activities of DNMT1 specifically. While the main focus in the field is on activating tumor-suppressor genes using DNA methylation and HDAC inhibitors, we suggest that it might be worthwhile to target the opposite process hypomethylation. Inhibitors of the DNA demethylation machinery might be utilized to silence pro-metastatic genes (Pakneshan et al., 2004). An important issue that needs to be determined in future studies is whether demethylation inhibitors would cause silencing of tumor-suppressor genes through hypermethylation, which could contraindicate their use. DNA methylation profiles are promising to turn into central diagnostic tools. We now know that not all tumors are methylated at the same sites. Each tumor type bears a “methylation profile,” a unique combination of methylated CGs, which has important predictive and diagnostic potential (Adorjan et al., 2002). The development of whole-genome high-density oligonucleotide arrays will allow in the future to map in detail the DNA methylation and chromatin changes associated with disease and behavioral abnormalities. These methylation profiles might offer an explanation for interindividual phenotypic diversity. Identifying target genes whose methylation and chromatin is changed in correlation with altered phenotypes will facilitate unraveling of the biological mechanisms behind these phenotypic differences.
■
69
The second area that carries great promise is the utilization of DNA methylation and epigenetic profiling in prophylactics. We need to map the response of DNA methylation to different environmental exposures such as social environment, food, and drugs. Epigenetics is a candidate to serve as the main mechanism used by the environment to stably reprogram genes. Understanding the relationship between the environment and disease-state–specific methylation profiles will allow charting prophylactic strategies to avoid deleterious exposure and to promote positive environmental interventions. One of the interesting pending questions in this respect is the issue of methyl-supplemented diets. Folate-fortified diets have been in place for some time. Folate-enriched diets might be helpful in protection from the global hypomethylation which is associated with cancer (Poirier, 2002). However, it is also possible that they might cause hypermethylation and silence critical genes such as tumor-suppressor genes. It is important to map the relationship between methyl supplementation and changes in DNA methylation profiles. Understanding how the environment utilizes epigenetics to sculpt the genome is perhaps one of the greatest challenges in biology, which would have a broad impact in medicine.
CONCLUSIONS Epigenetic mechanisms control gene expression programs and serve as an interface between the dynamic environment and the genome. Many of the proteins, which lay down the DNA methylation pattern as well as histone remodeling complexes and modifying enzymes, have been identified. Moreover, recent discoveries of chromatin demodification enzymes suggest that epigenetic processes are reversible and that the state of chromatin, as well as DNA methylation, is an equilibrium between modification and demodification processes. There is a bilateral relation between chromatin structure and demethylation, which is in part determined by protein–protein interactions between DNA methylation enzymes and chromatin modification enzymes. The establishment of new epigenetic patterns requires the targeting of DNA methylation and chromatin modification enzymes to specific genes, while maintenance of this pattern requires their continuous presence on these genes. Epigenetic proteins are targeted to specific sequences through interaction with specific trans-acting activators or repressors, which recognize specific cis-acting signals next to genes. Future experiments are required to understand the cellular signaling pathways leading to activation of these targeting factors and to determine whether they link physiological, environmental, or pathological signals with chromatin modification states and the DNA methylation pattern. One of the great challenges in the field is to define the mechanisms linking environment and the sculpting of the genes by epigenetic modifications. Epigenetic aberrations act similar to genetic mutations to precipitate pathological states. In cancer this is now well established. It stands to reason that similar processes are involved in other diseases such as mental disorders, cardiovascular disorders, and metabolic disorders. The main challenge now is to unravel
70
CHAPTER 5
■
Epigenomics and Genomic Medicine
the epigenetic code and identify the epigenetic profiles associated with human disease. Recent advances in whole-genome mapping using high-density oligonucleotide arrays and the use of 5-methylcytosine–specific antibodies, as well as chromatin immunoprecipitation (ChIP) with histone-modification–specific antibodies, will allow for mapping in unbiased way the state of methylation and chromatin modification of entire genomes and for identifying methylation profiles that are associated with disease. Since epigenomes vary from cell type to cell type and at different time points in life, this mapping effort is bound to be an enormous task that requires development of high-throughput methodologies. The whole-epigenome mapping efforts would comprehensively test for the first time the hypothesis that interindividual epigenetic differences are the basis for the well-known interindividual differences in appearance, behavior, and physiology. One provocative hypothesis that needs to be addressed by whole-epigenome mapping is that some of the environmentally-driven epigenetic changes are passed through the germline, thus explaining welldocumented trans-generational environmental effects. Since the epigenome is dynamic, it is potentially reversible. A further understanding of the DNA methylation and histone
modification machineries will allow us to determine whether we could intercede either pharmacologically or by behavioral and nutritional interventions to reverse deleterious epigenetic programming. We ought to link the main cellular signaling pathways, specific DNA methylation and chromatin modification enzymes, and DNA methylation profiles in order to design intelligent interventions that will have main positive effects in the absence of adverse outcomes. Epigenetics has made important headways and has advanced dramatically in the last few years; however, many methodological and conceptual challenges remain. If the promise of this field would be realized, it would revolutionize our understanding of biology, medical diagnosis, and therapeutics.
ACKNOWLEDGEMENTS The work from the author’s laboratory discussed in this chapter was funded by the National Cancer Institute of Canada, the Canadian Institute of Health Research, and the Canadian Institute of Advanced Research.
REFERENCES Adorjan, P. et al. (2002). Tumour class prediction and discovery by microarray-based DNA methylation analysis. Nucleic Acids Res 30, e21. Amir, R.E.,Van den Veyver, I.B., Wan, M., Tran, C.Q., Francke, U. and Zoghbi, H.Y. (1999). Rett syndrome is caused by mutations in X-linked MECP2, encoding methyl-CpG-binding protein 2. Nat Genet 23, 185–188. Balada, E., Ordi-Ros, J., Serrano-Acedo, S., Martinez-Lostao, L. and Vilardell-Tarres, M. (2007). Transcript overexpression of the MBD2 and MBD4 genes in CD4+ T cells from systemic lupus erythematosus patients. J Leukoc Biol 81, 1609–1616. Baylin, S.B., Esteller, M., Rountree, M.R., Bachman, K.E., Schuebel, K. and Herman, J.G. (2001). Aberrant patterns of DNA methylation, chromatin formation and gene expression in cancer. Hum Mol Genet 10, 687–692. Beck, S., Olek, A. and Walter, J. (1999). From genomics to epigenomics: A loftier view of life. Nat Biotechnol 17, 1144. Bestor, T.H. (1990). DNA methylation: Evolution of a bacterial immune function into a regulator of gene expression and genome structure in higher eukaryotes. Philos Trans R Soc Lond B Biol Sci 326, 179–187. Bhattacharya, S.K., Ramchandani, S., Cervoni, N. and Szyf , M. (1999). A mammalian protein with specific demethylase activity for mCpG DNA. Nature 397, 579–583. Borrow, J. et al. (1996). The translocation t(8;16)(p11;p13) of acute myeloid leukaemia fuses a putative acetyltransferase to the CREBbinding protein. Nat Genet 14, 33–41. Bostick, M., Kim, J.K., Esteve, P.O., Clark, A., Pradhan, S. and Jacobsen, S.E. (2007). UHRF1 plays a role in maintaining DNA methylation in mammalian cells. Science 317, 1760–1764.
Bourc’his, D., Xu, G.L., Lin, C.S., Bollman, B. and Bestor, T.H. (2001). Dnmt3L and the establishment of maternal genomic imprints. Science 294, 2536–2539. Bruniquel, D. and Schwartz, R.H. (2003). Selective, stable demethylation of the interleukin-2 gene enhances transcription by an active process. Nat Immunol 4, 235–240. Bultman, S.J., Gebuhr, T.C. and Magnuson, T. (2005). A Brg1 mutation that uncouples ATPase activity from chromatin remodeling reveals an essential role for SWI/SNF-related complexes in beta-globin expression and erythroid development. Genes Dev 19, 2849–2861. Cervoni, N. and Szyf, M. (2001). Demethylase activity is directed by histone acetylation. J Biol Chem 276, 40778–40787. Chalitchagorn, K., Shuangshoti, S., Hourpai, N., Kongruttanachok, N., Tangkijvanich, P., Thong-ngam, D., Voravud, N., Sriuranpong, V. and Mutirangura, A. (2004). Distinctive pattern of LINE-1 methylation level in normal tissues and the association with carcinogenesis. Oncogene 23, 8841–8846. Chen, Y., Sharma, R.P., Costa, R.H., Costa, E. and Grayson, D.R. (2002). On the epigenetic regulation of the human reelin promoter. Nucleic Acids Res 30, 2930–2939. Cheng, X., Collins, R.E. and Zhang, X. (2005). Structural and sequence motifs of protein (histone) methylation enzymes. Annu Rev Biophys Biomol Struct 34, 267–294. Chuang, L.S., Ian, H.I., Koh, T.W., Ng, H.H., Xu, G. and Li, B.F. (1997). Human DNA-(cytosine-5) methyltransferase-PCNA complex as a target for p21WAF1. Science 277, 1996–2000. Cornacchia, E., Golbus, J., Maybaum, J., Strahler, J., Hanash, S. and Richardson, B. (1988). Hydralazine and procainamide inhibit T cell DNA methylation and induce autoreactivity. J Immunol 140, 2197–2200.
References
Corwin, E.J. (2004). The concept of epigenetics and its role in the development of cardiovascular disease: Commentary on “new and emerging theories of cardiovascular disease”. Biol Res Nurs 6, 11–16. Discussion 21–23. Costa, E., Chen,Y., Davis, J., Dong, E., Noh, J.S.,Tremolizzo, L.,Veldic, M., Grayson, D.R. and Guidotti, A. (2002). REELIN and schizophrenia: A disease at the interface of the genome and the epigenome. Mol Interv 2, 47–57. Costa, E., Grayson, D.R., Mitchell, C.P., Tremolizzo, L., Veldic, M. and Guidotti, A. (2003). GABAergic cortical neuron chromatin as a putative target to treat schizophrenia vulnerability. Crit Rev Neurobiol 15, 121–142. Deng, C., Lu, Q., Zhang, Z., Rao, T., Attwood, J., Yung, R. and Richardson, B. (2003). Hydralazine may induce autoimmunity by inhibiting extracellular signal-regulated kinase pathway signaling. Arthritis Rheum 48, 746–756. Devaskar, S.U. and Thamotharan, M. (2007). Metabolic programming in the pathogenesis of insulin resistance. Rev Endocr Metab Disord 8, 105–113. Di Croce, L. et al. (2002). Methyltransferase recruitment and DNA hypermethylation of target promoters by an oncogenic transcription factor. Science 295, 1079–1082. Feinberg, A.P. (2007). Phenotypic plasticity and the epigenetics of human disease. Nature 447, 433–440. Feinberg, A.P. and Vogelstein, B. (1983). Hypomethylation of ras oncogenes in primary human cancers. Biochem Biophys Res Commun 111, 47–54. Feinberg, A.P., Gehrke, C.W., Kuo, K.C. and Ehrlich, M. (1988). Reduced genomic 5-methylcytosine content in human colonic neoplasia. Cancer Res 48, 1159–1161. Filetici, P., P, O. and Ballario, P. (2001). The bromodomain: A chromatin browser?. Front Biosci 6, D866–D876. Finch, J.T., Lutter, L.C., Rhodes, D., Brown, R.S., Rushton, B., Levitt, M. and Klug, A. (1977). Structure of nucleosome core particles of chromatin. Nature 269, 29–36. Fraga, M.F. et al. (2005). Epigenetic differences arise during the life-time of monozygotic twins. Proc Natl Acad Sci USA 102, 10604–10609. Fuks, F., Burgers, W.A., Brehm, A., Hughes-Davies, L. and Kouzarides, T. (2000). DNA methyltransferase Dnmt1 associates with histone deacetylase activity. Nat Genet 24, 88–91. Fuks, F., Hurd, P.J., Wolf , D., Nan, X., Bird, A.P. and Kouzarides, T. (2003). The methyl-CpG-binding protein MeCP2 links DNA methylation to histone methylation. J Biol Chem 278, 4035–4040. Gibbons, R.J. (2005). Histone modifying and chromatin remodelling enzymes in cancer and dysplastic syndromes. Hum Mol Genet 14(Spec No 1), R85–R92. Goffin, J. and Eisenhauer, E. (2002). DNA methyltransferase inhibitorsstate of the art. Ann Oncol 13, 1699–1716. Gosden, R.G. and Feinberg, A.P. (2007). Genetics and epigenetics– nature’s pen-and-pencil set. N Engl J Med 356, 731–733. Grayson, D.R., Jia, X., Chen,Y., Sharma, R.P., Mitchell, C.P., Guidotti, A. and Costa, E. (2005). Reelin promoter hypermethylation in schizophrenia. Proc Natl Acad Sci USA 102, 9341–9346. Guo, Y., Pakneshan, P., Gladu, J., Slack, A., Szyf, M. and Rabbani, S.A. (2002). Regulation of DNA methylation in human breast cancer. Effect on the urokinase-type plasminogen activator gene production and tumor invasion. J Biol Chem 277, 41571–41579. Henikoff , S., McKittrick, E. and Ahmad, K. (2004). Epigenetics, histone H3 variants, and the inheritance of chromatin states. Cold Spring Harb Symp Quant Biol 69, 235–243.
■
71
Hiltunen, M.O.,Turunen, M.P., Hakkinen,T.P., Rutanen, J., Hedman, M., Makinen, K., Turunen, A.M., Aalto-Setala, K. and Yla-Herttuala, S. (2002). DNA hypomethylation and methyltransferase expression in atherosclerotic lesions. Vasc Med 7, 5–11. Holbert, M.A. and Marmorstein, R. (2005). Structure and activity of enzymes that remove histone modifications. Curr Opin Struct Biol 15, 673–680. Jenuwein, T. (2001). Re-SET-ting heterochromatin by histone methyltransferases. Trends Cell Biol 11, 266–273. Jenuwein, T. and Allis, C.D. (2001). Translating the histone code. Science 293, 1074–1080. Jost, J.P. (1993). Nuclear extracts of chicken embryos promote an active demethylation of DNA by excision repair of 5-methyldeoxycytidine. Proc Natl Acad Sci USA 90, 4684–4688. Kaminsky, Z., Wang, S.C. and Petronis, A. (2006). Complex disease, gender and epigenetics. Ann Med 38, 530–544. Ke, X. et al. (2006). Uteroplacental insufficiency affects epigenetic determinants of chromatin structure in brains of neonatal and juvenile IUGR rats. Physiol Genomics 25, 16–28. Kersh, E.N., Fitzpatrick, D.R., Murali-Krishna, K., Shires, J., Speck, S.H., Boss, J.M. and Ahmed, R. (2006). Rapid demethylation of the IFN-gamma gene occurs in memory but not naive CD8 T cells. J Immunol 176, 4083–4093. Kim, J., Kim, J.Y., Song, K.S., Lee, Y.H., Seo, J.S., Jelinek, J., Goldschmidt-Clermont, P.J. and Issa, J.P. (2007). Epigenetic changes in estrogen receptor beta gene in atherosclerotic cardiovascular tissues and in-vitro vascular senescence. Biochim Biophys Acta 1772, 72–80. Kuo, M.H. and Allis, C.D. (1998). Roles of histone acetyltransferases and deacetylases in gene regulation. Bioessays 20, 615–626. Lachner, M. and Jenuwein, T. (2002). The many faces of histone lysine methylation. Curr Opin Cell Biol 14, 286–298. Lachner, M., O’Carroll, D., Rea, S., Mechtler, K. and Jenuwein, T. (2001). Methylation of histone H3 lysine 9 creates a binding site for HP1 proteins. Nature 410, 116–120. Lee, M.G., Wynder, C., Cooch, N. and Shiekhattar, R. (2005). An essential role for CoREST in nucleosomal histone 3 lysine 4 demethylation. Nature 437, 432–435. Leonhardt, H., Page, A.W., Weier, H.U. and Bestor, T.H. (1992). A targeting sequence directs DNA methyltransferase to sites of DNA replication in mammalian nuclei. Cell 71, 865–873. Li, E., Bestor, T.H. and Jaenisch, R. (1992). Targeted mutation of the DNA methyltransferase gene results in embryonic lethality. Cell 69, 915–926. Lichtenstein, M., Keini, G., Cedar, H. and Bergman, Y. (1994). B cellspecific demethylation: A novel role for the intronic kappa chain enhancer sequence. Cell 76, 913–923. Lu, L.J., Randerath, E. and Randerath, K. (1983). DNA hypomethylation in Morris hepatomas. Cancer Lett 19, 231–239. Lucarelli, M., Fuso, A., Strom, R. and Scarpa, S. (2001). The dynamics of myogenin site-specific demethylation is strongly correlated with its expression and with muscle differentiation. J Biol Chem 276, 7500–7506. Lyko, F., Ramsahoye, B.H. and Jaenisch, R. (2000). DNA methylation in Drosophila melanogaster. Nature 408, 538–540. MacLennan, N.K., James, S.J., Melnyk, S., Piroozi, A., Jernigan, S., Hsu, J.L., Janke, S.M., Pham, T.D. and Lane, R.H. (2004). Uteroplacental insufficiency alters DNA methylation, one-carbon metabolism, and histone acetylation in IUGR rats. Physiol Genomics 18, 43–50.
72
CHAPTER 5
■
Epigenomics and Genomic Medicine
McClelland, M., Nelson, M. and Raschke, E. (1994). Effect of sitespecific modification on restriction endonucleases and DNA modification methyltransferases. Nucleic Acids Res 22, 3640–3659. Meaney, M.J. and Szyf , M. (2005). Maternal care as a model for experience-dependent chromatin plasticity?. Trends Neurosci 28, 456–463. Metzger, E., Wissmann, M.,Yin, N., Muller, J.M., Schneider, R., Peters, A.H., Gunther, T., Buettner, R. and Schule, R. (2005). LSD1 demethylates repressive histone marks to promote androgenreceptor-dependent transcription. Nature 437, 436–439. Milutinovic, S., Zhuang, Q., Niveleau, A. and Szyf, M. (2003). Epigenomic stress response. Knockdown of DNA methyltransferase 1 triggers an intra-S-phase arrest of DNA replication and induction of stress response genes. J Biol Chem 278, 14985–14995. Milutinovic, S., Brown, S.E., Zhuang, Q. and Szyf , M. (2004). DNA methyltransferase 1 knock down induces gene expression by a mechanism independent of DNA methylation and histone deacetylation. J Biol Chem 279, 27915–27927. Mohandas, T., Sparkes, R.S. and Shapiro, L.J. (1981). Reactivation of an inactive human X chromosome: Evidence for X inactivation by DNA methylation. Science 211, 393–396. Muchardt, C. and Yaniv, M. (1999). ATP-dependent chromatin remodelling: SWI/SNF and Co. are on the job. J Mol Biol 293, 187–198. Nan, X., Campoy, F.J. and Bird, A. (1997). MeCP2 is a transcriptional repressor with abundant binding sites in genomic chromatin. Cell 88, 471–481. Ng, H.H., Zhang, Y., Hendrich, B., Johnson, C.A., Turner, B.M., Erdjument-Bromage, H., Tempst, P., Reinberg, D. and Bird, A. (1999). MBD2 is a transcriptional repressor belonging to the MeCP1 histone deacetylase complex. Nat Genet 23, 58–61. Novik, K.L., Nimmrich, I., Genc, B., Maier, S., Piepenbrock, C., Olek, A. and Beck, S. (2002). Epigenomics: genome-wide study of methylation phenomena. Curr Issues Mol Biol 4, 111–128. Okano, M., Xie, S. and Li, E. (1998). Cloning and characterization of a family of novel mammalian DNA (cytosine-5) methyltransferases [letter]. Nat Genet 19, 219–220. Okano, M., Bell, D.W., Haber, D.A. and Li, E. (1999). DNA methyltransferases Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian development. Cell 99, 247–257. Pakneshan, P., Szyf , M., Farias-Eisner, R. and Rabbani, S.A. (2004). Reversal of the hypomethylation status of urokinase (uPA) promoter blocks breast cancer growth and metastasis. J Biol Chem 279, 31735–31744. Peters, A.H. and Schubeler, D. (2005). Methylation of histones: playing memory with DNA. Curr Opin Cell Biol 17, 230–238. Poirier, L.A. (2002). The effects of diet, genetics and chemicals on toxicity and aberrant DNA methylation: an introduction. J Nutr 132, 2336S–2339S. Prendergast, G.C. and Ziff , E.B. (1991). Methylation-sensitive sequencespecific DNA binding by the c-Myc basic region. Science 251, 186–189. Pruss, D., Hayes, J.J. and Wolffe, A.P. (1995). Nucleosomal anatomy– where are the histones?. Bioessays 17, 161–170. Pusarla, R.H. and Bhargava, P. (2005). Histones in functional diversification. Core histone variants. Febs J 272, 5149–5168. Quddus, J., Johnson, K.J., Gavalchin, J., Amento, E.P., Chrisp, C.E., Yung, R.L. and Richardson, B.C. (1993). Treating activated CD4+ T cells with either of two distinct DNA methyltransferase inhibitors, 5-azacytidine or procainamide, is sufficient to cause a lupuslike disease in syngeneic mice. J Clin Invest 92, 38–53.
Razin, A. (1998). CpG methylation, chromatin structure and gene silencing – A three-way connection. Embo J 17, 4905–4908. Razin, A. and Cedar, H. (1977). Distribution of 5-methylcytosine in chromatin. Proc Natl Acad Sci USA 74, 2725–2728. Razin, A. and Riggs, A.D. (1980). DNA methylation and gene function. Science 210, 604–610. Rea, S. et al. (2000). Regulation of chromatin structure by site-specific histone H3 methyltransferases. Nature 406, 593–599. Richardson, B.C. (2002). Role of DNA methylation in the regulation of cell function: autoimmunity, aging and cancer. J Nutr 132, 2401S–2405S. Robertson, K.D., Ait-Si-Ali, S., Yokochi, T., Wade, P.A., Jones, P.L. and Wolffe, A.P. (2000). DNMT1 forms a complex with Rb, E2F1 and HDAC1 and represses transcription from E2F-responsive promoters. Nat Genet 25, 338–342. Rosty, C., Ueki, T., Argani, P., Jansen, M., Yeo, C.J., Cameron, J.L., Hruban, R.H. and Goggins, M. (2002). Overexpression of S100A4 in pancreatic ductal adenocarcinomas is associated with poor differentiation and DNA hypomethylation. Am J Pathol 160, 45–50. Rountree, M.R., Bachman, K.E. and Baylin, S.B. (2000). DNMT1 binds HDAC2 and a new co-repressor, DMAP1, to form a complex at replication foci. Nat Genet 25, 269–277. Santangelo, S., Cousins, D.J., Winkelmann, N.E. and Staynov, D.Z. (2002). DNA methylation changes at human Th2 cytokine genes coincide with DNase I hypersensitive site formation during CD4(+) T cell differentiation. J Immunol 169, 1893–1903. Santos-Rosa, H., Schneider, R., Bannister, A.J., Sherriff, J., Bernstein, B.E., Emre, N.C., Schreiber, S.L., Mellor, J. and Kouzarides,T. (2002). Active genes are tri-methylated at K4 of histone H3. Nature 419, 407–411. Sarma, K. and Reinberg, D. (2005). Histone variants meet their match. Nat Rev Mol Cell Biol 6, 139–149. Scheinbart, L.S., Johnson, M.A., Gross, L.A., Edelstein, S.R. and Richardson, B.C. (1991). Procainamide inhibits DNA methyltransferase in a human T cell line. J Rheumatol 18, 530–534. Selker, E.U. (1998). Trichostatin A causes selective loss of DNA methylation in Neurospora. Proc Natl Acad Sci USA 95, 9430–9435. Shi, Y., Lan, F., Matson, C., Mulligan, P., Whetstine, J.R., Cole, P.A., Casero, R.A. and Shi, Y. (2004). Histone demethylation mediated by the nuclear amine oxidase homolog LSD1. Cell 119, 941–953. Shilatifard, A. (2006). Chromatin modifications by methylation and ubiquitination: Implications in the regulation of gene expression. Annu Rev Biochem 75, 243–269. Simmons, R.A. (2007). Developmental origins of beta-cell failure in type 2 diabetes: The role of epigenetic mechanisms. Pediatr Res 61, 64R–67R. Stenvinkel, P. et al. (2007). Impact of inflammation on epigenetic DNA methylation – A novel risk factor for cardiovascular disease?. J Intern Med 261, 488–499. Strahl, B.D. and Allis, C.D. (2000). The language of covalent histone modifications. Nature 403, 41–45. Sugimoto, S. et al. (2004). Expression and regulation of tumor suppressor gene maspin in human bladder cancer. Cancer Lett 203, 209–215. Swain, J.L., Stewart, T.A. and Leder, P. (1987). Parental legacy determines methylation and expression of an autosomal transgene: A molecular mechanism for parental imprinting. Cell 50, 719–727. Szyf , M. (2001). Towards a pharmacology of DNA methylation. Trends Pharmacol Sci 22, 350–354. Szyf, M. (2003). Targeting DNA methylation in cancer. Ageing Res Rev 2, 299–328.
References
Szyf , M., Eliasson, L., Mann, V., Klein, G. and Razin, A. (1985). Cellular and viral DNA hypomethylation associated with induction of Epstein-Barr virus lytic cycle. Proc Natl Acad Sci USA 82, 8090–8094. Tsukada, Y., Fang, J., Erdjument-Bromage, H., Warren, M.E., Borchers, C.H., Tempst, P. and Zhang, Y. (2006). Histone demethylation by a family of JmjC domain-containing proteins. Nature 439, 811–816. Varga-Weisz, P.D. and Becker, P.B. (2006). Regulation of higher-order chromatin structures by nucleosome-remodelling factors. Curr Opin Genet Dev 16, 151–156. Veldic, M., Kadriu, B., Maloku, E., Agis-Balboa, R.C., Guidotti, A., Davis, J.M. and Costa, E. (2007). Epigenetic mechanisms expressed in basal ganglia GABAergic neurons differentiate schizophrenia from bipolar disorder. Schizophr Res 93, 33–41. Verdone, L., Caserta, M. and Di Mauro, E. (2005). Role of histone acetylation in the control of gene expression. Biochem Cell Biol 83, 344–353. Vilain, A., Apiou, F., Dutrillaux, B. and Malfoy, B. (1998). Assignment of candidate DNA methyltransferase gene (DNMT2) to human chromosome band 10p15.1 by in situ hybridization. Cytogenet Cell Genet 82, 120. Vire, E. et al. (2006). The Polycomb group protein EZH2 directly controls DNA methylation. Nature 439, 871–874. Wade, P.A., Pruss, D. and Wolffe, A.P. (1997). Histone acetylation: Chromatin in action. Trends Biochem Sci 22, 128–132. Wan, M. et al. (1999). Rett syndrome and beyond: recurrent spontaneous and familial MECP2 mutations at CpG hotspots. Am J Hum Genet 65, 1520–1529.
■
73
Weaver, I.C., Cervoni, N., Champagne, F.A., D’Alessio, A.C., Sharma, S., Seckl, J.R., Dymov, S., Szyf, M. and Meaney, M.J. (2004a). Epigenetic programming by maternal behavior. Nat Neurosci 7, 847–854. Weaver, I.C., Diorio, J., Seckl, J.R., Szyf, M. and Meaney, M.J. (2004b). Early environmental regulation of hippocampal glucocorticoid receptor gene expression: Characterization of intracellular mediators and potential genomic target sites. Ann N Y Acad Sci 1024, 182–212. Weaver, I.C., Champagne, F.A., Brown, S.E., Dymov, S., Sharma, S., Meaney, M.J. and Szyf, M. (2005). Reversal of maternal programming of stress responses in adult offspring through methyl supplementation: Altering epigenetic marking later in life. J Neurosci 25, 11045–11054. Wolffe, A.P. and Hayes, J.J. (1999). Chromatin disruption and modification. Nucleic Acids Res 27, 711–720. Zhang, H., Roberts, D.N. and Cairns, B.R. (2005). Genome-wide dynamics of Htz1, a histone H2A variant that poises repressed/ basal promoters for activation through histone loss. Cell 123, 219–231. Zhang, Y., Ng, H.H., Erdjument-Bromage, H., Tempst, P., Bird, A. and Reinberg, D. (1999). Analysis of the NuRD subunits reveals a histone deacetylase core complex and a connection with DNA methylation. Genes Dev 13, 1924–1935. Zhu, B., Zheng, Y., Hess, D., Angliker, H., Schwarz, S., Siegmann, M., Thiry, S. and Jost, J.P. (2000). 5-methylcytosine-DNA glycosylase activity is present in a cloned G/T mismatch DNA glycosylase associated with the chicken embryo DNA demethylation complex. Proc Natl Acad Sci USA 97, 5135–5139.
CHAPTER
6 Systems Biology and the Emergence of Systems Medicine Nathan D. Price, Lucas B. Edelman, Inyoul Lee, Hyuntae Yoo, Daehee Hwang, George Carlson, David J. Galas, James R. Heath and Leroy Hood
INTRODUCTION A new approach to biology, which we call systems biology, has emerged over the past 10 years or so – an approach that looks at biology as an informational science, studies biological systems as a whole, and recognizes that biological information in living systems is captured, transmitted, modulated, and integrated by biological networks that pass this information to molecular machines for execution. This approach differs from early “systems approaches to biology” in that it attempts both a bottomup approach (from large molecular datasets) and a top-down approach (from computational modeling and simulations) where there is an attempt to trace complex observations of phenotype back to the digital core encoded in the genome. New measurement and visualization technologies, together with powerful computational and modeling tools, have transformed systems biology by making it possible for the first time to execute the five uniquely defining features of contemporary systems biology: 1. measurements of the various types of biological information that are global or comprehensive to the greatest extent possible (e.g., measure the digital code of all genes or, for example, the concentrations of all mRNAs, all proteins, all metabolites, etc.); Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 74
2. the different levels of information (DNA, RNA, protein, etc.) must be integrated to understand or capture how the environment has modified the basic digital information of the genome at each level of the biological information hierarchy (DNA, RNA, proteins, interactions, biological networks, cells, individuals and ecologies) and thereby to induce biological responses; 3. all biological systems (e.g., networks) must be studied dynamically as they capture, transmit, integrate and utilize the biological information necessary for the execution the two most fundamental responses of living organisms – development or physiological responses to the organism’s environment; 4. all measurements must be quantitatively determined to the greatest extent possible; and 5. the global and dynamic data from the variety of information hierarchies must be integrated and modeled. These models essentially create hypotheses about biological functions and mechanisms of disease that can then be tested experimentally by systems perturbations – the enumeration of new datasets, their integration, modifications of models, in an iterative manner, until the working models reflect the reality of the experimental data. Once the models accurately fit the biological systems measurements, they can then be used to predict the results of perturbing the system in new ways (e.g., for Copyright © 2009, Elsevier Inc. All rights reserved.
Systems Science in Biology and Medicine
designing treatment strategies or for identifying likely underlying causes for disease). The information revolution in biology will enable systems medicine to emerge in the next 10 years. Key to this transformation will be harnessing computationally the vast amount of biological information becoming available through rapid advancement in measurement, visualization and computational technologies. Two fundamental types of biological information, the digital information of the genome and interacting information from the environment, are integrated together to specify the five fundamental mechanisms of life – evolution, development, physiological responses, aging and the onset and progression of disease (Hood et al., 2004). Advances in high-throughput DNA sequencing will enable each person’s genome to be sequenced rapidly and at a reasonable cost, providing digital information for each person and making possible the prediction of increasingly accurate probabilistic health futures. Also, the advance of highthroughput measurement technologies will enable the assessment of dynamic environmental information emerging from the integration of genome and environmental information from each individual, as reflected, for example, by the changing levels of proteins in the blood – thus providing a real time (current) health assessment of the individual. These technologies will generate tremendously large, dynamical datasets about health states and hence about the states of relevant biological networks of the individual patients. This detailed information will arise through the use of perturbation experiments coupled with global experimental technologies, and progress in systems biology research is increasingly elucidating the functions and structures of these networks. A systems approach to medicine argues that disease arises from disease-perturbed biological networks and that the dynamically changing, altered patterns of gene expression that are controlled by these perturbed networks give rise to the disease manifestations. Here, we present a systems view of biology and disease together with a discussion of some recent advances in state-of-the-art in vitro and in vivo diagnostics technologies, and we suggest how, as these technologies mature, they will move us towards a future of predictive, personalized, preventive, and participatory medicine (P4 medicine) (Hood et al., 2004).
SYSTEMS SCIENCE IN BIOLOGY AND MEDICINE Systems Biology How do systems approaches to biology and medicine lead to a revolutionary view of medicine? We present this systems view in some detail, because an understanding of P4 medicine and its implications for society are predicated upon understanding these systems principles. Two primary domains of biological information lend themselves readily to such systems-level analysis: the static, digital information of the genome, and the dynamic information arising from environmental interactions with the subcellular, cellular, and tissue levels of organization (Hood et al.,
■
75
2004). Digital genome information encodes two types of biological networks – protein interactions and gene regulatory networks. Protein networks transmit and use biological information for development, physiology and metabolism. Gene regulatory networks – transcription factors, their regulatory binding sites and the small RNAs that regulate networks of other transcription factors and other RNAs interacting with one another – receive information from signal-transduction networks, integrate and modulate it, and convey the processed information to networks of genes or molecular machines that execute developmental and physiological functions. In biological systems, these two types of networks are closely integrated. The organization of these networks can be inferred from various types of measurements including, for example, global measurements of dynamically changing levels of mRNAs and proteins during developmental and physiological responses, as well as large-scale measurements of protein–protein and protein–DNA interactions. There are multiple hierarchical levels of organization and information (for example, DNA, RNA and protein networks, cellular and metabolic networks, and organization and responses of organ systems). To understand biological systems, information must be gathered from as many information levels as possible and integrated into models that generate hypotheses about how biology works. Let us now consider the logic of systems approaches to medicine. Systems Medicine The central premise of systems medicine is that diseaseperturbed networks alter patterns of expression in genes and proteins – and that these altered patterns encode the dynamic pathophysiology of the disease and necessarily results in altered molecular fingerprints that can be detected clinically. Through advances in measurement technologies and computational analysis tools for tissue and blood analyses, we will be able to read these signals to make a multitude of diagnoses to distinguish health from disease and to determine the nature of any pathology. Multi-parameter analyses of biological information will be key to tracking these altered patterns in disease-perturbed networks. Diagnostic methods of the past have been pauciparameter in nature – usually measuring just a single parameter relevant to a specific disease state (e.g., PSA levels to assess prostate health), and so our ability to track health and disease has been limited. Current and emerging technologies are creating a transition into a new era of predictive, preventative, and personalized medicine. A causal disease perturbation could be the result of specific, disease-causing DNA mutations, pathogenic organisms, or other pathological environmental factors such as toxins. Molecular fingerprints of pathological processes can take on many molecular forms, including the analyses of proteins (Wang et al., 2005a, b), DNA (Papadopoulou et al., 2006), RNA (Scherzer et al., 2007), and metabolites (Solanky et al., 2003), as well as informative, post-translational modifications to these molecules such as protein phosphorylation. Signals related to health and disease can be found in multiple sites. For instance, many bodily fluids such as the blood, urine, saliva, cerebral spinal fluid and so forth can be sampled to
76
CHAPTER 6
■
Systems Biology and the Emergence of Systems Medicine
identify evidence of perturbed molecular fingerprints reflecting the altered expression patterns of genes and proteins in diseaseperturbed biological networks. Of these, the blood is likely the most information rich organ (or fluid) in that it bathes all of the tissues in the body, and it is easily accessible for diagnostic procedures. In addition to the biomolecules secreted into the blood from cells and tissues throughout the body, the transcriptomes and proteomes of cells circulating in the blood (e.g., white blood cells) are also potentially an abundant source of biomedically important information (Buttner et al., 2007). Thus, the amount of information available in the blood about health and disease is enormous if we learn to read and interpret the molecular signals. As an example of an environmental disease perturbation, we have studied the dynamic onset of the infectious prion disease in several strains of mice at the level of mRNA in the brain (the affected organ) and showed that a series of interlocking protein networks that surround the prion protein are significantly perturbed across the 150-day span from disease initiation to death. Figure 6.1 shows differential networks that were derived from comparisons of mRNA-expression patterns in normal and diseased brains at each of three time points after infection. They involve 67 proteins in the prion replication and accumulation network. The initial network changes occur well before the clinical signs of the disease can be detected and predict later widespread histopathological events. These dynamically changing, disease-perturbed networks lead to two important conclusions. First, some significant network nodal points change before the related clinical or histological changes are evident. Therefore, labeled biomarkers that are specific for the changing nodes or the biological processes they regulate could be used for in vivo imaging diagnostics even before symptoms arise, as has already been shown in patients (Golub et al., 1999; Quackenbush, 2006; Ramaswamy et al., 2001;Wang et al., 2005a, b). Alternatively, if some of these altered nodes encode secreted proteins, they could provide readily accessible in vitro diagnostic blood markers for early disease detection. Second, many of the sub-networks of proteins that change during the onset of disease affect changes in phenotypic traits that are consistent with the pathology of the disease. About 900 perturbed mRNAs appear to encode the core prion-disease response, and, during the progression of the priondisease process, several hundred undergo significant gene expression changes well before the clinical signs of prion disease. Many of these potential early disease “sentinels” are predicted to be secreted into the blood and therefore represent potential protein biomarkers for early disease diagnosis through blood protein analysis. One interesting new diagnostic method evaluates relative expression reversals in protein concentrations or gene expression levels. This procedure eliminates the need for data normalization or the establishment of population-wide thresholds. This approach has been successfully used to identify robust and accurate classifiers for prostate cancer (Xu et al., 2005), sarcomas (Price et al., 2007), and a variety of other cancers (Tan et al., 2005), as well as to predict treatment outcomes in breast
cancer (Ma et al., 2004), demonstrating the efficacy of even relatively low-order systems analysis in medicine. Another key issue that needs to be addressed is how to best decouple two primary dimensions of disease information: namely, disease stratification (e.g., which type of prostate cancer is present) and disease progression (e.g., stage of development of a particular prostate cancer stratification in time). Molecular signature changes can indicate the presence of different diseases (stratification) as well as their stage (progression). Thus, a key challenge in future will be to develop analysis tools that will enable us to differentiate the location of the physiologic state in regards to both of these critical clinical dimensions.
MULTI-PARAMETER BLOOD-BOURNE BIOMARKERS In principle, it is clear how blood samples containing secreted molecules from all of the tissues in the body can be used as a window into health or disease states (Anderson and Anderson, 2002; Fujii et al., 2004; Hood et al., 2004; Lathrop et al., 2003; Lee et al., 2006). In practice, however, the task of identifying markers of disease states in the vast array of secreted proteins can seem daunting. There are millions of different proteins present in the blood and they are expressed at levels that probably differ by 12 orders of magnitude (1012). Thus, there are indeed very significant challenges associated with defining relevant biomarkers in the plasma proteome, as is evidenced by the paucity of blood protein markers found thus far, despite significant efforts (Wilson, 2006). This fact highlights the need to develop wellfounded systems approaches to diagnostics in order to hasten the identification of such markers and to harness the tremendous information potential of the blood. While the potential information available to diagnose heath and disease is enormous, it is also true that the challenges of separating signal from background noise (biological or measurement noise) are also very significant. Sources of noise include error in measurements, polymorphisms in the population, environmental variations, stochastic variations, as well as signal dilution through mixing and other processes of molecule transport. Thus, enhancing signal while reducing noise will be a primary theme of research in predictive medicine for the foreseeable future. An important strategy for dealing with noise arising from genetic polymorphisms and health histories in the population are through dynamic, subtractive analyses carried out in the individual patient. In other words, measurements taken from individual patients at different time points (longitudinal data) can be used to perform subtractive analyses where only the differences observed are considered. Thus, each patient becomes their own control, which eliminates many of the sources of noise. Eventually, it will be important for each patient to have biannual blood analyses taken so that changes can be observed relative to the background of what is normal in each individual rather than relative to the population at large. Such databases of measurements will be essential to enabling personalized and predictive
22 wks
18 wks Clinical Signs
No Clinical Signs
PRND
LAMAL LAMRI
STIP
PRND
FCGRL
CSNK2A1
LRP2 A2M
LRP1 LRPAP1
C3
FYN
GM2A
LYN
SHC1
SHC1
GFAP
CTSB
YWHAZ VCL
ANXA2 CTSH
CST3 S100ALO
MEN
PPL
CTSD STAT3
PDGFRA
ESR1 NAPK1
STAT1
CEBPB
MEN
PPL
CTSD
MAPK1
STAT1
LAG3
CEBPB
CREBBP HDAC1
TRP53
ITGAV
2 wks
JUN
GSN PPL
CTSD STAT3
PDGFRA
ESR1
TGAV
MAPK1
STATL
CEBPB
CREBBP HDAC1
YWHAZ
VIM
MEN
LAG3 HDAC1
TRP53 RELA
RELA NDTCH1
GFAP
VCL
CTS3 SL00AL0
LAG3
TRP53
RELA JUN
PDGFRA
ESR1
STAT3
PSEN1 ANXA2
CTSH
GSN
VIM
CTSB
NOTCH1
12 wks
JUN
NOTCH1
20 wks
Figure 6.1 A central sub-network involved in neuronal pathology in prion disease in mice, shown at three times after inoculation. Red circles indicate increased levels of gene expression relative to controls.
Multi-Parameter Blood-Bourne Biomarkers
CREBBP
TTGAV
VCL
CTS3 SL00AL0
EZH2
BCL2L11 CTSB
YWHAZ
ANXA2 CTSH
GSN
VIM
GFAP
PSEN
DNM2
GRB2 BCL2
EZH2
CTSB
VAV1
CASP3
DNM2
BCL2L11
BCL2L11 SEN
SHC1
VAV1 GRB2
BCL2
EZH2
SYK
APLPL APP
CASP3
DNM2
GRB2 BCL2
PTK2B
SYK
APP
VAV1
CASP3
LYN
HEXA PTK2B
SYK APLP
FYN
GM2A
LYN
HEXA PTK2B
APP
CSTB
LRP1 LRPAP1
C3
FYN
GM2A
HEXA
CSTB
CAV1 PLD2
CAV1 PLD2
CAV1 PLD2
APLP1
PCGRL
CSNK2A1
APOB
LRP2
LRAP1
C1QA C1QR1 PCGR3
HSPAJA
A2M LRP1
STIP1
PRNP
PCGRL
CSNK2A1
APOB
LRP2
APCS
PLAUR
C1QA C1QR1 PCGR3
HSPAPA
A2M C3
STIP1
PRNP
PRND LAMR1
PLG
PLAU
APCS
PLAUR
C1QA C1QR1 FCGR3
HSPA1A APOB
LAMA1
LAMR1 PLG
PLAU
APCS
PLAUR
PRND
LAMAL
PLG
PLAL
■
77
78
CHAPTER 6
■
Systems Biology and the Emergence of Systems Medicine
medicine. Using each patient as their own control is one of the essential features of the emerging personalized medicine. Organ-Specific Blood Protein Fingerprints We discuss the organ-specific blood protein fingerprint approach to diagnostics because it lies at the very heart of the predictive medicine that will emerge over the next 10 years. Our feeling is that this will be one of the central data gathering strategies for predictive and personalized medicine and, accordingly, we will discuss it in detail. The idea is that disease arises from dynamically changing disease-perturbed networks, that the diseased organs will secrete proteins into the blood, and that if these proteins are encoded by disease-perturbed networks their levels of expression in the blood will be altered in a manner that reflects the specific nature of the disease. Indeed, this idea is the basis for the very broad range of blood biomarker studies that are being carried out by many scientific centers (Anderson and Anderson, 2002; Fujii et al., 2004; Hood et al., 2004; Lathrop et al., 2003; Lee et al., 2006). The major difficulty with this simple view is that if you identify multiple blood proteins whose changes are specific for a particular disease state (as compared against normal controls) and then examine the same blood markers for that disease in bloods drawn from individuals, with say 10 other pathologies, these marker proteins can change in unpredictable ways since multiple organs control the expression of most blood proteins and these organs respond differently to various environmental signals. The important point is that if a marker protein synthesized in five organs changes in the blood, we cannot be certain which organ(s) is the origin of the change. While that marker may sample a biological network relevant to disease diagnosis, since its origin is not clear its use as a disease biomarker may raise more questions than it answers. The solution to this dilemma is clear – employ organspecific blood protein biomarkers whose changes must therefore reflect changes only in the organ itself. If enough of these organspecific blood proteins are sampled, they will represent a survey of many different biological networks in the organ of interest and will provide sufficient diagnostic information for any disease. Eventually, correlation of these organ-specific biomarkers with more general biomarkers may prove to be the best long-term strategy for developing diagnostic fingerprints. We have several lines of preliminary evidence that suggest this organ-specific blood protein approach will be effective. In prostate cancer, for example, there are disease-mediated altered patterns of mRNA and protein expression in the prostate. Some of these genes are expressed primarily in the prostate (organspecific products) and some of these organ-specific proteins are secreted into the blood, where they collectively constitute a protein molecular fingerprint comprised of say 100 or more proteins whose relative concentration levels probably report the status of the biological networks in the prostate gland. We have demonstrated that changes in the blood concentrations of several of these prostate-specific blood proteins reflect the various stages of prostate cancer and, as discussed above, various brainspecific blood proteins also reflect the progression of prion disease in mice. We propose that the distinct expression levels of the
individual proteins in each fingerprint represent a multiparameter (and therefore potentially information-rich) diagnostic indicator reflecting the dynamic behavior of, for example, the disease-perturbed networks from which they arise. The analysis of 50 or so organ-specific proteins should allow us to both stratify diseases in the organ as well as determine their stage of progression. We have identified tens to hundreds of organ-specific transcripts in each of the 40 or so organs that we have examined in mouse and human. We can envision a time over the next 5–10 years when 50 or so organ-specific blood proteins will be identified for each of the 50 or so major organs and tissues in humans – so that computational analyses of the relative concentrations of the protein components in these organ-specific fingerprints will enable blood to become the primary window into health and disease. When we analyze data from these blood indicators, we also may be in a position to identify the dynamically changing disease-perturbed network(s). The analysis of these dynamic networks will allow us to study in detail the pathophysiology of the disease response and hence be in a position to think of new approaches to therapy and prevention. To generate the ability to assess simultaneously all 50 organ blood protein fingerprints in patients, we ultimately need to develop the microfluidic or nanotechnology tools for making perhaps 2500 rapid protein measurements (e.g., 50 proteins from each of 50 human organs) from a droplet of blood. Nanotechnology is necessary because only this severe miniaturization can allow the necessary thousands of measurements from a single drop of blood. In order to reach this stage we will also have to create the appropriate computational tools to capture, store, analyze, integrate, model, and visualize the information arising from this approach.
EMERGING IN VIVO AND IN VITRO TECHNOLOGIES In vitro Measurement Technologies P4 medicine will require that in vitro measurements of thousands of blood proteins be executed rapidly, automatically and inexpensively on small blood samples. This will require miniaturization, parallelization, integration and automation of tissue and blood purification and measurement technologies. In short, it will require making in vitro measurements inexpensively – perhaps for a penny or less per protein measured. This need for inexpensive in vitro measurements is driving the development of integrated microfluidics and nanotechnologies for in vitro diagnostics. We further argue that the complex changes in these protein levels must be analyzed computationally to search for the patterns that correlate with particular diseases – and indeed to stratify each disease as well as determine its stage of progression. To meet the expanding requirements of systems medicine, in vitro measurements of cells, proteins, mRNAs, etc., whether for fundamental biological studies or for pathological analysis,
Emerging In vivo and In vitro Technologies
must not just be inexpensive, but must also be sensitive, quantitative, rapid, and executed on very small quantities of tissues, cells, serum, etc. We are working towards developing chip platforms that can take a few microliters (a finger prick) of blood, separate the plasma from the serum, and then measure on the order of 100 or more proteins and/or mRNAs quantitatively, with high sensitivity and specificity, and in a few minutes time. We can envision a time 5–10 years hence when small hand-held devices will be able to make these 2500 measurements from a fraction of a droplet of blood, send them via wireless to a server for analysis and consequently inform the patient and physician as to the status of the patient. While many fundamental scientific challenges remain to be solved before this goal is achieved, none of those challenges appears insurmountable at this point. One will also be able to use the blood cells as powerful diagnostic markers – either of infectious diseases or of genetic diseases. Microfluidic cell-sorting technologies for being able to sort blood cells into their 10 or so individual types for analysis are now available. Even more important is the emergence of single-cell analytic tools where DNA, mRNAs or small RNAs, proteins and even metabolites can be analyzed rapidly from individual cells. The cells can also be perturbed with appropriate environmental stimuli to identify defects. It is possible that an appropriate analysis of immune cells (both innate and adaptive) from the blood will reveal important information about past antigenic history of the individual as well as current state of immunological responsiveness. Similarly, analysis of rare blood cells such as circulating cancer cells may also be utilized to guide therapies. Another type of in vitro measurement will be a determination of the complete genome sequence of individuals with a nanotechnology approach to sequencing single strands of DNA on a massively parallel basis (billions of DNA strands simultaneously). A device like this will emerge over the next 5–10 years and will allow millions, if not billions, of individual human genome sequences to be determined rapidly, inexpensively and in a massively parallel manner. The 2500 organ-specific blood marker measurements and the complete individual human genome sequences will be the heart of the predictive and personalized medicine that will emerge over the next 10 years or so. While transcriptomic approaches have proven useful for identifying informative molecular signatures for a number of diseases (Quackenbush, 2006), high-throughput proteomic characterizations have lagged behind. The reason for this difference is clearly the more difficult challenge of measuring proteins compared to mRNA transcripts. Emerging proteomics technologies hold the promise of greatly improving our ability to make detailed disease assessments using protein-based molecular signatures. One key advantage of protein signatures relative to gene expression is that the proteins found in the blood and other accessible bodily fluids exhibit slower degradation rates than their mRNA counterparts – and of course proteins are the direct agents mediating the disease process itself. Thus, protein concentration signatures represent an important class of molecular signature for disease diagnosis.
■
79
The Imaging of Proteomics and Blood Protein Biomarkers The blood is an ideal organ for identifying biomarkers because it interacts with virtually every organ and major tissue type in the body, and each secretes proteins into the blood. Moreover, the blood is easily accessible for diagnostic studies – and presumably the analysis of changes in protein levels or protein modifications that could serve as surrogates or reflections of health and disease. As noted earlier, the blood is an enormously complex organ; it contains hundreds of thousands to millions of proteins whose concentrations range over 12 orders of magnitude. Since there is for proteins no equivalent to the polymerase chain reaction (PCR) amplification procedure, only the more abundant proteins can be seen by conventional protein analysis studies. It should also be pointed out that many different features of proteins ultimately must be characterized or quantified: identification, quantification, chemical modifications, alternative mRNA splicing products, cellular localizations, three-dimensional structures (and their dynamics), as well as their functions. We here are only concerned with the quantification and identification of proteins. There are two general approaches to the analysis of protein mixtures. In one case protein-capture agents such as antibodies are used – and with appropriate controls these approaches can quantify the levels of proteins. These techniques include Western blot analyses, ELISA assays, surface plasmon resonance and protein chips. A second approach is to use mass spectrometry. Since proteins have large masses that are difficult to analyze accurately by mass spectrometry, most proteins are analyzed after converting them into tryptic peptides (or other proteolytic fragments) whose masses are far smaller and hence more easily analyzed. Because blood is such a complex mixture, proteins of interest are often enriched by some type of fractional procedure (charge, hydrophobility, size and/or the presence of modifications such as glycosylation or phosphorylation), before (or after) their conversion into peptides and analysis by mass spectrometry. Isotopically labeled and synthesized peptides may be used to identify and quantify (relative or absolute) the same peptides from unknown samples (and hence one obtains a proxy for the quantification of their corresponding proteins) very effectively by mass spectrometry. We discuss below examples of some of these techniques. Antibody Microarrays Using Surface Plasmon Resonance Imaging (SPRI) Protein chip methods hold great potential for broad quantitative screens of proteins, and a variety of techniques have been developed based on antibody binding (Haab et al., 2001; Olle et al., 2005; Song et al., 2007).Various types of antibody arrays have been used for biomarker discovery and protein profiling of serum from patients with prostate, lung, pancreas, and bladder cancer (Gao et al., 2005;Miller et al., 2003;Orchekowski et al., 2005;Sanchez-Carbayo et al., 2006). One emerging approach with tremendous promise is SPRI (Hu et al., 2007; Huber and Mueller, 2006; Koga et al., 2006), which enables real-time, label-free measurement of protein expression by large numbers of different antibodies. SPR-based chips have a detection sensitivity of 10–100 times less than ELISA
80
CHAPTER 6
■
Systems Biology and the Emergence of Systems Medicine
(Hu et al., 2007), but have a spatial resolution down to approximately 4 m (Lyon et al., 1998). It is thus possible to print up to 800 unique antibodies on Lumera Nanocapture Gold™ microarray slides and monitor the abundance of the target proteins in real time (Hu et al., 2007), even in complex samples such as blood, because the same slide can be regenerated for reuse many times in 10 minutes or less (Z. Hu, C. Lausted, unpublished observations). We are in the process of automating the sample introduction procedures using microfluidic chips, which means that SPRI has the capacity necessary to screen rapidly through tens to hundreds of patient samples. Thus, this approach holds tremendous promise to be able to screen through large numbers of proteins, including secreted proteins and cell-surface markers and not only measure their presence, but also abundance and the dynamics of their binding. The limitation of this technique is its dependence on the affinity and specificity of the antibodies it employs for detection – cross-reactivities in complex protein mixtures (such as blood) can pose significant problems. DNA-Encoded Antibody Libraries (DEAL) One recently developed technique that offers great potential for detailed analyses is DEAL. The primary advantage of DEAL is that it uses a single, robust chemistry – that of DNA hybridization – to spatially localize and detect proteins, mRNAs, and cells, all in a multiplexed fashion. Antibodies are typically too fragile to survive the fabrication procedures associated with assembling robust microfluidics chips, but DNA oligomers are significantly more robust. DEAL thus enables the detection of panels of protein biomarkers within a microfluidics environment and from very small quantities of biological material (100 nanoliters or so) (Yang et al., 2006). This amount of plasma can be readily separated from whole blood on-chip, thus allowing for the measurement of serum biomarkers from a fingerprick of whole blood. In addition, within the environment of flowing microfluidics, the rate-limiting step in performing a surface immunoassay is the kinetics of the binding of the analyte to the surface-bound capture agent (Zimmermann et al., 2005). Thus, DEAL-based immunoassays can be executed very rapidly. The versatility of DEAL also enables multiplexed cell sorting and localization, followed by few- or single-cell measurements of protein, RNA, and other biomolecules (Bailey et al., 2007). DEAL can be engineered into a highly sensitive and very rapid measurement technique, with a reported detection limit of 10 f M for the protein IL-2, 150 times more sensitive than the corresponding commercial ELISA assay. This sensitivity can be applied to the isolation of rare cells based on combinations of cell-surface markers, enabling the isolation and addressing of individual cancer stem cells. DEAL can also be used to make single-cell measurements of secreted proteins from each of these isolated single cells. Thus, DEAL offers superb sensitivity and the ability to perform spatially multiplexed detection for characterization of rare cell types, such as circulating cancer cells or cancer stem cells. These advantages still face the inherent limitations of antibodies, so the development of new approaches to generating
protein-capture agents is a critical part of future development of comprehensive blood diagnostics. Mass Spectrometry-based Techniques Isotopic Tagging for Relative and Absolute Quantification (iTRAQ) of Proteins Analyzed by Mass Spectrometry Stable isotope labeling enables the quantitative analysis of protein concentrations through mass spectrometry (MS). One stateof-the-art technique for quantitative MS is iTRAQ (Ross et al., 2004), which uses stable isotope labeling of proteolytic peptides. This technique modifies primary amino acid groups of peptides by linking a mass balance group, and a reporter group by forming an amide bond. When MS/MS is used for analysis with iTRAQ-tagged peptides, the mass balancing carbonyl moiety is released as a neutral fragment generating 4–8 distinct sets of peptides whose relative abundances can be determined quantitatively. Because eight different iTRAQ reagents are currently available, comparative analysis of a set of two to eight samples is feasible within a single MS/MS run (Hu et al., 2007). The, iTRAQ technology represents the state-of-the-art in quantitative proteomics and represents a promising technology for using proteomics to differentiate key differences in protein networks. Glycopeptide Capture – Front-End Enrichment of Blood Proteins Containing Sugar Residues Mass spectrometry-based methods will allow for the identification of proteins spanning approximately three orders of magnitude in concentration from a given sample. Therefore, methods that can select specified fractions of the proteome are important for simplifying the sample sufficiently to identify the proteins of interest. One recently developed approach is the shotgun glycopeptides capture approach (Sun et al., 2007). This approach selects for N-linked or O-linked glycosylated peptides by selectively coupling these peptides to beads – allowing the uncoupled peptides to be washed away – then the glycosylated peptides are be released and analyzed by mass spectrometry. Both secreted and cell-surface proteins are enriched for glycopeptides compared with their nuclear and cytoplasmic counterparts. Thus, this approach can be used to, for example, identify candidates for unique cell-surface markers to make identifications of clinically relevant cellular subpopulations. In Vivo Molecular Diagnostics As it relates to personalized and predictive medicine, in vivo molecular diagnostics will also require the development of a diverse library of molecular imaging probes. These modular tools can be used to: 1. identify the specific location of disease-perturbed networks in patients; 2 link in vivo molecular measurements in diseased tissue in patients to in vitro measurements; 3. rapidly assess the efficacy of personalized therapeutics; and 4. validate that a drug is hitting its target and inducing the desired pharmacological outcome.
Conclusions and Perspectives
Although there are many in vivo imaging modalities, perhaps the best current method from the point of view of personalized and predictive medicine in patients is positron emission tomography (PET) molecular imaging (Czernin and Phelps, 2002). For PET, trace quantities of radiolabeled molecular probes are injected into the patient. As the probes circulate through the body and its various organ systems, they interact with target proteins to provide imaging assays for, as examples, the rate of metabolic processes, the concentration of receptors in signal transduction, enzyme activity, DNA-replication rates, hormone status, pharmacokinetics, and pharmacodynamics.
COMPUTATIONAL AND MATHEMATICAL CHALLENGES IN SYSTEMS MEDICINE Molecular Signature Classifiers Computational methods are needed to reduce the high degree of data dimensionality associated with global datasets to identify molecular signatures that can be used for disease diagnosis. Despite notable and significant challenges that remain (Dupuy and Simon, 2007; Simon, 2005), computational analyses to identify molecular signatures from global gene expression datasets that can be used for diagnosis and treatment selection (Quackenbush, 2006) is an area of research that has shown significant promise. These studies typically involve the collection of samples from two or more classes (e.g., cancer versus normal, or responsive versus non-responsive to treatment) and the use of a set of data on which to train the classifier and another set on which to test. In the absence of a true test set, re-sampling methods (such as cross-validation) are generally used to estimate likely performance of the classifier on future data. The ability to generate an accurate classifier is a function of factors such as: (1) the size of the training set relative to the number of features, (2) the computational method used, and (3) the inherent distinctness of the selected phenotypes. Typically, the number of samples is far less than the number of transcripts, and consequently over-fitting is a significant problem. This leads to the need for computational methods that avoid over-fitting when selecting a classifier. A variety of methods have been applied to disease diagnoses, including approaches based on support vector machines (Furey et al., 2000; Ramaswamy et al., 2001) and relative expression reversals (Geman et al., 2004; Price et al., 2007; Tan et al., 2005; Xu et al., 2005), among many others. Application of these methods has perhaps been applied most extensively to the study of cancer, leading to the discovery of molecular classifiers of varying degrees of accuracy to identify prognostic signatures for breast cancer (Adler and Chang, 2006; Buyse et al., 2006; Dai et al., 2005; Foekens et al., 2006; Glinsky et al., 2004a, b; Goncalves et al., 2006; Ivshina et al., 2006; Liu et al., 2007; Ma et al., 2004; Nuyten et al., 2006; Park et al., 2007; Pawitan et al., 2005; Thomassen et al., 2007; van ‘t Veer et al., 2002; van de Vijver et al.,
■
81
2002; Wang et al., 2005a, b; Weigelt et al., 2005), ovarian cancer (De Cecco et al., 2004; Smith, 2002; Spentzos et al., 2004), colon cancer (Barrier et al., 2006; Giacomini et al., 2005), prostate cancer (Dhanasekaran et al., 2001; Glinsky et al., 2004a, b; Halvorsen et al., 2005; Lin et al., 2005; Luo et al., 2002; Singh et al., 2002; Xu et al., 2005), and brain cancer (Fuller et al., 2005; Kim et al., 2002), among others. Biological Networks and the Interactome Among the most promising opportunities presented by systems biology is the integrated assessment of multiple biological parameters to magnify our understanding of complex developmental or pathological states. The interaction among all individual biomolecules in complex regulatory, signal transduction, and feedback networks present in eukaryotic organisms forms the theoretical framework for this higher-order analysis and is referred to as the interactome. These dynamic associations include protein–protein interactions, transcriptional regulation, and post-transcriptional gene silencing by short-interfering RNA and microRNA (Ruvkun, 2001). Multiple high-throughput methods have been developed to characterize the interaction between proteins, including the yeast-two-hybrid system (Ito et al., 2001) and surface plasmon resonance (Usui-Aoki et al., 2005). These methods are to determining interactome interactions what microarray studies are to determining relative mRNA concentrations.
CONCLUSIONS AND PERSPECTIVES Nanotechnology and microfluidic sensing chips with integrated microfluidic delivery devices will be developed over the next 5–10 years so that we reach a point at which thousands of proteins and/or RNAs can be analyzed quantitatively from only a fraction of a droplet of blood. This will allow physicians to assess the health/disease status of virtually every organ in the body at 6-month intervals and will become the first critical element of predictive medicine. We also predict that nanotechnology approaches to the sequencing of single DNA strands will over the next 5–10 years lead to instrumentation that can rapidly, cheaply and simply sequence individual human genomes. This will be a second foundation of predictive medicine – assessing the variability in individual human genomes to generate probabilistic future health histories for each individual human, and integrating it with diagnostics. Similarly, in vivo imaging agents will permit the functional visualization of informational molecules and drugs in humans, which will provide powerful and informative in vivo diagnostics approaches. These measurement technologies, together with both new computational approaches to extract information from these data and the systems view of medicine, will lead to three important revolutions in medicine. First, over the next 10–20 years, a medicine – P4 medicine – that is predictive, personalized, preventive and participatory will emerge. Obtaining individual human genome sequences will lead to individual predictive health histories, whereas biannual blood measurements of thousands of proteins from the molecular
82
CHAPTER 6
■
Systems Biology and the Emergence of Systems Medicine
fingerprints of human organs will give us a real-time assessment of individual health. Predictive medicine will result in a personalized medicine that focuses on the illnesses of an individual and eventually their wellness, preventing disease rather than treating it. The systems approach to disease will in time permit the identification of the key disease-perturbed subnetworks and the identification of important nodal points, which, if perturbed by drugs, could make the network behave in a more normal fashion or at least delete the more deleterious effects of the disease-perturbed network. This capability will provide a new approach for drug-target discovery, and a powerful and rapid new approach for developing drugs. Predictive, personalized and preventive medicine, if appropriately orchestrated with patient-oriented interpretations, will also enable patients to understand more deeply and actively participate in personal choices about illness and wellness. Participatory medicine will necessitate the development of powerful new approaches to handling enormous amounts of personal information in a secure manner, and to a new form of medical education of individual patients as well as their physicians. Thus, over the next 5–20 years, medicine will become predictive, personalized, preventive and participatory (P4 medicine). Another very exciting idea is that, in the long-term, systems analysis of blood offers the potential to open up a new avenue for studying human biology – when we learn to read and interpret the information inherent in secreted organ-specific protein patterns. The key patterns in these secreted proteins represent different network perturbations, and hence, different diseases. When secreted proteins enter the blood, they can provide a novel means to study biology in higher organisms and to identify drug targets through linking blood measurements to perturbations in underlying biological networks. Importantly, learning to study perturbations in underlying networks through secreted protein patterns (to whatever resolution is possible) will provide access to studying biological networks in vivo that are not amenable to direct experimentation. Developing the capacity to identify in vivo network perturbations through secreted protein measurements in the blood will open up a new avenue to drug
target identification and will provide a novel means to discover the perturbed sub-networks. Since taking blood is relatively non-invasive, it has less potential than biopsies or other invasive techniques of greatly distorting the system being studied. Thus, this approach has the potential to strongly complement existing approaches to study human biology. The effect of drugs, toxicity, human development and even aging may all be amenable to study through the blood if we can learn to read and interpret the signals in the proteins, which has the potential to be very significant in human biology. Second, as the sensitivity of measurements increases (both in vitro and in vivo), we will achieve a digitalization of biology and medicine – that is, the ability to extract relevant information content from single individuals, single cells and ultimately single molecules – with its attendant economies of scale. Just as Moore’s law led in time to the widespread digitalization of information technologies and communications, the exponentially increasing ability to extract quantized biological information from individual cells and molecules will transform biology and medicine in ways that we can only begin to imagine. Finally, all of these changes – a systems approach to medicine with its focus on disease prevention and more efficient drug discovery, the introduction of increasingly inexpensive nanotechnology based diagnostics and in vivo measurement technologies, the highly accurate and specific molecular characterization of the systems biology of disease, and the digitalization of medicine – will start to reduce the inexorably increasing costs of health care. This can, in principle, enable us to afford to provide for the more than 45 million people in the United States who currently lack health insurance, to reduce the crushing costs of healthcare to society, and to export our digital predictive and preventive medical approach to the developing world. Just as the mobile phone has become a fundamental communication mode in developing countries and has changed the lives of much of the world’s population, so digital medicine of the 21st century can bring to the world’s citizens a global and strongly improved state of human health and healthcare. In our view P4 medicine will become the foundational framework for global medicine.
REFERENCES Adler, A.S. and Chang, H.Y. (2006). From description to causality: Mechanisms of gene expression signatures in cancer. Cell Cycle 5, 1148–1151. Anderson, N.L. and Anderson, N.G. (2002). The human plasma proteome: History, character, and diagnostic prospects. Mol Cell Proteomics 1, 845–867. Bailey, R.C., Kwong, G.A., Radu, C.G., Witte, O.N. and Heath, J.R. (2007). DNA-encoded antibody libraries: A unified platform for multiplexed cell sorting and detection of genes and proteins. J Am Chem Soc 129, 1959–1967. Barrier, A., Boelle, P.Y., Roser, F., Gregg, J., Tse, C., Brault, D., Lacaine, F., Houry, S., Huguier, M., Franc, B. et al. (2006). Stage II colon cancer prognosis prediction by tumor gene expression profiling. J Clin Oncol 24, 4685–4691.
Buttner, P., Mosig, S., Lechtermann, A., Funke, H. and Mooren, F.C. (2007). Exercise affects the gene expression profiles of human white blood cells. J Appl Physiol 102, 26–36. Buyse, M., Loi, S., van’t Veer, L., Viale, G., Delorenzi, M., Glas, A.M., d’Assignies, M.S., Bergh, J., Lidereau, R., Ellis, P. et al. (2006). Validation and clinical utility of a 70-gene prognostic signature for women with node-negative breast cancer. J Natl Cancer Inst 98, 1183–1192. Czernin, J. and Phelps, M.E. (2002). Positron emission tomography scanning: Current and future applications. Annu Rev Med 53, 89–112. Dai, H., van’tVeer, L., Lamb, J., He,Y.D., Mao, M., Fine, B.M., Bernards, R., van de Vijver, M., Deutsch, P., Sachs, A. et al. (2005). A cell proliferation signature is a marker of extremely poor outcome in a subpopulation of breast cancer patients. Cancer Res 65, 4059–4066.
References
De Cecco, L., Marchionni, L., Gariboldi, M., Reid, J.F., Lagonigro, M.S., Caramuta, S., Ferrario, C., Bussani, E., Mezzanzanica, D., Turatti, F. et al. (2004). Gene expression profiling of advanced ovarian cancer: Characterization of a molecular signature involving fibroblast growth factor 2. Oncogene 23, 8171–8183. Dhanasekaran, S.M., Barrette, T.R., Ghosh, D., Shah, R., Varambally, S., Kurachi, K., Pienta, K.J., Rubin, M.A. and Chinnaiyan, A.M. (2001). Delineation of prognostic biomarkers in prostate cancer. Nature 412, 822–826. Duarte, N.C., Becker, S.A., Jamshidi, N., Thiele, I., Mo, M.L.,Vo, T.D., Srivas, R. and Palsson, B.O. (2007). Global reconstruction of the human metabolic network based on genomic and bibliomic data. Proc Natl Acad Sci USA 104, 1777–1782. Dupuy, A. and Simon, R.M. (2007). Critical review of published microarray studies for cancer outcome and guidelines on statistical analysis and reporting. J Natl Cancer Inst 99, 147–157. Foekens, J.A., Atkins, D., Zhang,Y., Sweep, F.C., Harbeck, N., Paradiso, A., Cufer, T., Sieuwerts, A.M., Talantov, D., Span, P.N. et al. (2006). Multicenter validation of a gene expression-based prognostic signature in lymph node-negative primary breast cancer. J Clin Oncol 24, 1665–1671. Fujii, K., Nakano, T., Kawamura, T., Usui, F., Bando, Y., Wang, R. and Nishimura, T. (2004). Multidimensional protein profiling technology and its application to human plasma proteome. J Proteome Res 3, 712–718. Fuller, G.N., Mircean, C., Tabus, I., Taylor, E., Sawaya, R., Bruner, J.M., Shmulevich, I. and Zhang, W. (2005). Molecular voting for glioma classification reflecting heterogeneity in the continuum of cancer progression. Oncol Rep 14, 651–656. Furey, T.S., Cristianini, N., Duffy, N., Bednarski, D.W., Schummer, M. and Haussler, D. (2000). Support vector machine classification and validation of cancer tissue samples using microarray expression data. Bioinformatics 16, 906–914. Gao,W.M., Kuick, R., Orchekowski, R.P., Misek, D.E.,Qiu, J., Greenberg, A.K., Rom, W.N., Brenner, D.E., Omenn, G.S., Haab, B.B. et al. (2005). Distinctive serum protein profiles involving abundant proteins in lung cancer patients based upon antibody microarray analysis. BMC Cancer 5, 110. Geman, D., d’Avignon, C., Naiman, D.Q. and Winslow, R.L. (2004). Classifying gene expression profiles from pairwise mRNA comparisons. Stat Appl Genet Mol Biol 3(Article19). Giacomini, C.P., Leung, S.Y., Chen, X., Yuen, S.T., Kim, Y.H., Bair, E. and Pollack, J.R. (2005). A gene expression signature of genetic instability in colon cancer. Cancer Res 65, 9200–9205. Glinsky, G.V., Glinskii, A.B., Stephenson, A.J., Hoffman, R.M. and Gerald, W.L. (2004). Gene expression profiling predicts clinical outcome of prostate cancer. J Clin Invest 113, 913–923. Glinsky, G.V., Higashiyama, T. and Glinskii, A.B. (2004). Classification of human breast cancer using gene expression profiling as a component of the survival predictor algorithm. Clin Cancer Res 10, 2272–2283. Golub, T.R., Slonim, D.K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J.P., Coller, H., Loh, M.L., Downing, J.R., Caligiuri, M.A. et al. (1999). Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring. Science 286, 531–537. Goncalves, A., Esterni, B., Bertucci, F., Sauvan, R., Chabannon, C., Cubizolles, M., Bardou, V.J., Houvenaegel, G., Jacquemier, J., Granjeaud, S. et al. (2006). Postoperative serum proteomic profiles
■
83
may predict metastatic relapse in high-risk primary breast cancer patients receiving adjuvant chemotherapy. Oncogene 25, 981–989. Haab, B.B., Dunham, M.J. and Brown, P.O. (2001). Protein microarrays for highly parallel detection and quantitation of specific proteins and antibodies in complex solutions. Genome Biol 2, RESEARCH0004. Halvorsen, O.J., Oyan, A.M., Bo, T.H., Olsen, S., Rostad, K., Haukaas, S.A., Bakke, A.M., Marzolf, B., Dimitrov, K., Stordrange, L. et al. (2005). Gene expression profiles in prostate cancer: Association with patient subgroups and tumour differentiation. Int J Oncol 26, 329–336. Hood, L., Heath, J.R., Phelps, M.E. and Lin, B. (2004). Systems biology and new technologies enable predictive and preventative medicine. Science 306, 640–643. Hu, Z., Hood, L. and Tian, Q. (2007). Quantitative proteomic approaches for biomarker discovery. Proteomics. In review. Huber, W. and Mueller, F. (2006). Biomolecular interaction analysis in drug discovery using surface plasmon resonance technology. Curr Pharm Des 12, 3999–4021. Ito, T., Chiba, T., Ozawa, R., Yoshida, M., Hattori, M. and Sakaki, Y. (2001). A comprehensive two-hybrid analysis to explore the yeast protein interactome. Proc Natl Acad Sci USA 98, 4569–4574. Ivshina, A.V., George, J., Senko, O., Mow, B., Putti, T.C., Smeds, J., Lindahl, T., Pawitan,Y., Hall, P., Nordgren, H. et al. (2006). Genetic reclassification of histologic grade delineates new clinical subtypes of breast cancer. Cancer Res 66, 10292–10301. Joshi-Tope, G., Gillespie, M., Vastrik, I., D’Eustachio, P., Schmidt, E., de Bono, B., Jassal, B., Gopinath, G.R., Wu, G.R., Matthews, L. et al. (2005). Reactome: A knowledgebase of biological pathways. Nucleic Acids Res 33, D428–D432. Kim, S., Dougherty, E.R., Shmulevich, I., Hess, K.R., Hamilton, S.R., Trent, J.M., Fuller, G.N. and Zhang, W. (2002). Identification of combination gene sets for glioma classification. Mol Cancer Ther 1, 1229–1236. Koga, H., Kyo, M., Usui-Aoki, K. and Inamori, K. (2006). A chip-based miniaturized format for protein-expression profiling: the exploitation of comprehensively produced antibodies. Electrophoresis 27, 3676–3683. Lathrop, J.T., Anderson, N.L., Anderson, N.G. and Hammond, D.J. (2003). Therapeutic potential of the plasma proteome. Curr Opin Mol Ther 5, 250–257. Lee, H.J., Lee, E.Y., Kwon, M.S. and Paik, Y.K. (2006). Biomarker discovery from the plasma proteome using multidimensional fractionation proteomics. Curr Opin Chem Biol 10, 42–49. Lin, B., White, J.T., Lu, W., Xie, T., Utleg, A.G., Yan, X., Yi, E.C., Shannon, P., Khrebtukova, I., Lange, P.H. et al. (2005). Evidence for the presence of disease-perturbed networks in prostate cancer cells by genomic and proteomic analyses: A systems approach to disease. Cancer Res 65, 3081–3091. Liu, R., Wang, X., Chen, G.Y., Dalerba, P., Gurney, A., Hoey, T., Sherlock, G., Lewicki, J., Shedden, K. and Clarke, M.F. (2007). The prognostic role of a gene signature from tumorigenic breastcancer cells. N Engl J Med 356, 217–226. Luo, J., Dunn,T., Ewing, C., Sauvageot, J., Chen,Y.,Trent, J. and Isaacs,W. (2002). Gene expression signature of benign prostatic hyperplasia revealed by cDNA microarray analysis. Prostate 51, 189–200. Lyon, L.A., Musick, M.D. and Natan, M.J. (1998). Colloidal Au-enhanced surface plasmon resonance immunosensing. Anal Chem 70, 5177–5183.
84
CHAPTER 6
■
Systems Biology and the Emergence of Systems Medicine
Ma, X.J., Wang, Z., Ryan, P.D., Isakoff, S.J., Barmettler, A., Fuller, A., Muir, B., Mohapatra, G., Salunga, R., Tuggle, J.T. et al. (2004). A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell 5, 607–616. Miller, J.C., Zhou, H., Kwekel, J., Cavallo, R., Burke, J., Butler, E.B., Teh, B.S. and Haab, B.B. (2003). Antibody microarray profiling of human prostate cancer sera: Antibody screening and identification of potential biomarkers. Proteomics 3, 56–63. Mulquiney, P.J. and Kuchel, P.W. (2003). Modelling Metabolism with Mathematica Detailed Examples Including Erythrocyte Metabolism. CRC Press, Boca Raton, Fla. Nuyten, D.S., Kreike, B., Hart, A.A., Chi, J.T., Sneddon, J.B.,Wessels, L.F., Peterse, H.J., Bartelink, H., Brown, P.O., Chang, H.Y. et al. (2006). Predicting a local recurrence after breast-conserving therapy by gene expression profiling. Breast Cancer Res 8, R62. Olle, E.W., Sreekumar, A., Warner, R.L., McClintock, S.D., Chinnaiyan, A.M., Bleavins, M.R., Anderson, T.D. and Johnson, K. J. (2005). Development of an internally controlled antibody microarray. Mol Cell Proteomics 4, 1664–1672. Orchekowski, R., Hamelinck, D., Li, L., Gliwa, E., vanBrocklin, M., Marrero, J.A.,Vande Woude, G.F., Feng, Z., Brand, R. and Haab, B.B. (2005). Antibody microarray profiling reveals individual and combined serum proteins associated with pancreatic cancer. Cancer Res 65, 11193–11202. Palsson, B. (2004). Two-dimensional annotation of genomes. Nat Biotechnol 22, 1218–1219. Papadopoulou, E., Davilas, E., Sotiriou, V., Georgakopoulos, E., Georgakopoulou, S., Koliopanos, A., Aggelakis, F., Dardoufas, K., Agnanti, N.J., Karydas, I. et al. (2006). Cell-free DNA and RNA in plasma as a new molecular marker for prostate and breast cancer. Ann N Y Acad Sci 1075, 235–243. Park, E.S., Lee, J.S., Woo, H.G., Zhan, F., Shih, J.H., Shaughnessy, J.D. and Frederic Mushinski, J. (2007). Heterologous tissue culture expression signature predicts human breast cancer prognosis. PLoS ONE 2, e145. Pawitan,Y., Bjohle, J., Amler, L., Borg, A.L., Egyhazi, S., Hall, P., Han, X., Holmberg, L., Huang, F., Klaar, S. et al. (2005). Gene expression profiling spares early breast cancer patients from adjuvant therapy: Derived and validated in two population-based cohorts. Breast Cancer Res 7, R953–R964. Price, N.D. and Shmulevich, I. (2007). Biochemical and statistical network models for systems biology. Curr Opin Biotechnol 18, 365–370. Price, N.D., Reed, J.L. and Palsson, B.O. (2004). Genome-scale models of microbial cells: Evaluating the consequences of constraints. Nat Rev Microbiol 2, 886–897. Price, N.D., Schellenberger, J. and Palsson, B.O. (2004). Uniform sampling of steady-state flux spaces: Means to design experiments and to interpret enzymopathies. Biophys J 87, 2172–2186. Price, N.D.,Trent, J., El-Naggar, A.K., Cogdell, D.,Taylor, E., Hunt, K.K., Pollock, R.E., Hood, L., Shmulevich, I. and Zhang, W. (2007). Highly accurate two-gene classifier for differentiating gastrointestinal stromal tumors and leiomyosarcomas. Proc Natl Acad Sci USA 104, 3414–3419. Quackenbush, J. (2006). Microarray analysis and tumor classification. N Engl J Med 354, 2463–2472. Ramaswamy, S., Tamayo, P., Rifkin, R., Mukherjee, S., Yeang, C.H., Angelo, M., Ladd, C., Reich, M., Latulippe, E. and Mesirov, J.P. (2001). Multiclass cancer diagnosis using tumor gene expression signatures. Proc Natl Acad Sci USA 98, 15149–15154.
Ross, P.L., Huang, Y.N., Marchese, J.N., Williamson, B., Parker, K., Hattan, S., Khainovski, N., Pillai, S., Dey, S., Daniels, S. et al. (2004). Multiplexed protein quantitation in Saccharomyces cerevisiae using amine-reactive isobaric tagging reagents. Mol Cell Proteomics 3, 1154–1169. Ruvkun, G. (2001). Molecular biology. Glimpses of a tiny RNA world. Science 294, 797–799. Sanchez-Carbayo, M., Socci, N.D., Lozano, J.J., Haab, B.B. and CordonCardo, C. (2006). Profiling bladder cancer using targeted antibody arrays. Am J Pathol 168, 93–103. Scherzer, C.R., Eklund, A.C., Morse, L.J., Liao, Z., Locascio, J.J., Fefer, D., Schwarzschild, M.A., Schlossmacher, M.G., Hauser, M.A., Vance, J.M. et al. (2007). Molecular markers of early Parkinson’s disease based on gene expression in blood. Proc Natl Acad Sci USA 104, 955–960. Simon, R. (2005). Roadmap for developing and validating therapeutically relevant genomic classifiers. J Clin Oncol 23, 7332–7341. Singh, D., Febbo, P.G., Ross, K., Jackson, D.G., Manola, J., Ladd, C., Tamayo, P., Renshaw, A.A., D’Amico, A.V., Richie, J.P. et al. (2002). Gene expression correlates of clinical prostate cancer behavior. Cancer Cell 1, 203–209. Smith, D.I. (2002). Transcriptional profiling develops molecular signatures for ovarian tumors. Cytometry 47, 60–62. Solanky, K.S., Bailey, N.J., Beckwith-Hall, B.M., Davis, A., Bingham, S., Holmes, E., Nicholson, J.K. and Cassidy, A. (2003). Application of biofluid 1 H nuclear magnetic resonance-based metabonomic techniques for the analysis of the biochemical effects of dietary isoflavones on human plasma profile. Anal Biochem 323, 197–204. Song, S., Li, B., Wang, L., Wu, H., Hu, J., Li, M. and Fan, C. (2007). A cancer protein microarray platform using antibody fragments and its clinical applications. Mol Biosyst 3, 151–158. Spentzos, D., Levine, D.A., Ramoni, M.F., Joseph, M., Gu, X., Boyd, J., Libermann, T.A. and Cannistra, S.A. (2004). Gene expression signature with independent prognostic significance in epithelial ovarian cancer. J Clin Oncol 22, 4700–4710. Stark, C., Breitkreutz, B.J., Reguly, T., Boucher, L., Breitkreutz, A. and Tyers, M. (2006). BioGRID: A general repository for interaction datasets. Nucleic Acids Res 34, D535–D539. Sun, B., Ranish, J.A., Utleg, A.G.,White, J.T.,Yan, X., Lin, B. and Hood, L. (2007). Shotgun glycopeptide capture approach coupled with mass spectrometry for comprehensive glycoproteomics. Mol Cell Proteomics 6, 141–149. Tan, A.C., Naiman, D.Q., Xu, L., Winslow, R.L. and Geman, D. (2005). Simple decision rules for classifying human cancers from gene expression profiles. Bioinformatics 21, 3896–3904. Thiele, I., Price, N.D.,Vo, T.D. and Palsson, B.O. (2005). Candidate metabolic network states in human mitochondria. Impact of diabetes, ischemia, and diet. J Biol Chem 280, 11683–11695. Thomassen, M.,Tan, Q., Eiriksdottir, F., Bak, M., Cold, S. and Kruse,T.A. (2007). Prediction of metastasis from low-malignant breast cancer by gene expression profiling. Int J Cancer 120, 1070–1075. Usui-Aoki, K., Shimada, K., Nagano, M., Kawai, M. and Koga, H. (2005). A novel approach to protein expression profiling using antibody microarrays combined with surface plasmon resonance technology. Proteomics 5, 2396–2401. van ‘t Veer, L.J., Dai, H., van de Vijver, M.J., He,Y.D., Hart, A.A., Mao, M., Peterse, H.L., van der Kooy, K., Marton, M.J., Witteveen, A.T. et al. (2002). Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530–536.
Recommended Resources
van de Vijver, M.J., He, Y.D., van’t Veer, L.J., Dai, H., Hart, A.A., Voskuil, D.W., Schreiber, G.J., Peterse, J.L., Roberts, C., Marton, M.J. et al. (2002). A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347, 1999–2009. Wang, X.,Yu, J., Sreekumar, A., Varambally, S., Shen, R., Giacherio, D., Mehra, R., Montie, J.E., Pienta, K.J., Sanda, M.G. et al. (2005). Autoantibody signatures in prostate cancer. N Engl J Med 353, 1224–1235. Wang, Y., Klijn, J.G., Zhang, Y., Sieuwerts, A.M., Look, M.P., Yang, F., Talantov, D., Timmermans, M., Meijer-van Gelder, M.E., Yu, J. et al. (2005). Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet 365, 671–679. Weigelt, B., Hu, Z., He, X., Livasy, C., Carey, L.A., Ewend, M.G., Glas, A.M., Perou, C.M. and Van’t Veer, L.J. (2005). Molecular
■
85
portraits and 70-gene prognosis signature are preserved throughout the metastatic process of breast cancer. Cancer Res 65, 9155–9158. Wilson, J.F. (2006). The rocky road to useful cancer biomarkers. Ann Intern Med 144, 945–948. Xu, L., Tan, A.C., Naiman, D.Q., Geman, D. and Winslow, R.L. (2005). Robust prostate cancer marker genes emerge from direct integration of inter-study microarray data. Bioinformatics 21, 3905–3911. Yang, S., Undar, A. and Zahn, J.D. (2006). A microfluidic device for continuous, real time blood plasma separation. Lab Chip 6, 871–880. Zimmermann, M., Delamarche, E., Wolf, M. and Hunziker, P. (2005). Modeling and optimization of high-sensitivity, low-volume microfluidic-based surface immunoassays. Biomed Microdevices 7, 99–110.
RECOMMENDED RESOURCES Website
Article
www.systemsbiology.org Contains overviews of systems biology discipline, particularly in “Systems Biology in Depth” section of the website.
Hood, L., Heath, J.R., Phelps, M.E. and Lin, B. (2004). Systems biology and new technologies enable predictive and preventative medicine. Science 306, 640–643.
This page intentionally left blank
Section
Technology Platforms for Genomic Medicine
2
7. DNA Sequencing for the Detection of Human Genome Variation and Polymorphism 8. Genome-Wide Association Studies and Genotyping Technologies 9. Copy Number Variation and Human Health 10. Inter-Species Comparative Sequence Analysis: A Tool for Genomic Medicine 11. DNA Methylation Analysis: Providing New Insight into Human Disease 12. Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine 13. DNA Microarrays in Biological Discovery and Patient Care 14. Proteomics: The Deciphering of the Functional Genome 15. Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms 16. Comprehensive Analysis of Gene Function: RNA interference and Chemical Genomics
CHAPTER
7 DNA Sequencing for the Detection of Human Genome Variation and Polymorphism Samuel Levy and Yu-Hui Rogers
INTRODUCTION The discovery and characterization of DNA polymorphisms in human populations is an important step toward understanding the contribution of genome sequence variants to predisposition, onset and progression of disease phenotypes. By way of example, molecular epidemiology studies have been able to establish the importance of the wide range of DNA variants that cause monogenic disorders such as cystic fibrosis, where over 1000 mutations in the CFTR gene have been identified and implicated in the disease phenotype (Rowntree and Harris, 2003). The impact of the mutational landscape of the CFTR gene on biological function is complex, and emerging evidence implicates additional modifier genes contributing to the range of disease phenotypes observed (Davies et al., 2005). The situation is likely to be compounded further when identifying and characterizing the contribution of multiple genetic loci to phenotypes found in polygenic diseases such as coronary heart disease, diabetes and hypertension (see Chapter 3). While our understanding of how polymorphisms can contribute either singly or collectively to either disease onset or progression is still actively being developed, so too are the technologies that enable their detection and ascertainment in human populations.
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 88
Human DNA polymorphism detection has been greatly empowered in the last decade by the production of a high quality, mostly finished, DNA sequence of the human genome. The production of this sequence, and the fact that several distinct DNA sources were employed, enabled the direct identification of 2.1 million single nucleotide polymorphisms (SNPs) (Lander et al., 2001; Venter et al., 2001) to supplement those already found in the public variant databases (dbSNP). The discovery of these SNPs in a small number of humans set the stage for an ambitious experiment, the HapMap project, to define common haplotypes through the use of 4 million SNP types in 269 distinct humans (Consortium, 2005). It is anticipated that these datasets will enable the identification of a set of informative SNPs with which disease association studies can be accomplished with improved success. Indeed, early successes highlighting the utility of the HapMap data in identifying alleles implicated in complex diseases such as asthma (Laitinen et al., 2004), age related macular degeneration (Klein et al., 2005) and type II diabetes (Sladek et al., 2007), amongst others, have clearly set the stage for more studies of this kind (see Chapter 8). Essentially, one potential study design would likely involve the use of SNPs as markers for a disease in an affected DNA population compared to a control DNA population. Subsequently, any marker statistically associated with
Copyright © 2009, Elsevier Inc. All rights reserved.
DNA Sequencing
the disease group would inform the genomic loci requiring further sequence analysis to detect the causative variant. There are currently over 10 million SNPs and other polymorphisms such as insertion/deletion variants (indels) in public databases that potentially provide a marker set for efficient disease-gene association studies. It is important to note that even this large variant set still might not represent the variants causative of diseases. This is likely due to the fact that significant polymorphism discovery efforts to date have been performed in a random fashion across the genomes of only a limited number of individuals. Therefore, we still need to discover genetic variation in regions of functional DNA sequence in the genomes of individuals with disease, in order to provide an accurate understanding of disease etiology. In this chapter, we outline the sequencing protocols and the attendant computational approaches employed for the discovery and initial characterization of DNA variants, with an emphasis on the most commonly used Sanger-based sequencing (Sanger et al., 1977) methodologies established in our laboratories. These polymorphisms range from the substitution of single bases (i.e., SNPs) to the insertion or deletion of nucleotides in size ranges from 1 basepair (bp) to greater than 20,000 bp. Recent reports on the unexpected number and scope of larger scale copy number variants or structural variants ranging from 1–3 million bases (Mb) in size (Redon et al., 2006; Sebat et al., 2004) will not be addressed here since current methods used for their analysis involve either genotyping or microarray-based comparative genome hybridization, as presented elsewhere (see Chapter 9). We also highlight, where possible, how the application of these methods and advances has impacted our understanding of human biology. Several recent reviews addressed the methodologies and development of advanced sequencing methods and recent commercial platforms that exploit their use (Fan et al., 2006; Shendure et al., 2004). These advances in genome sequencing for variant detection are of particular importance as we have begun to see genome sequences and complete variants sets of individual humans, including as of this writing James D. Watson by 454 Life Sciences (Wheeler et al., 2008) and J. Craig Venter (Sanger-based sequencing (Levy et al., 2007)). The discussion in this chapter will focus primarily on mature approaches for use in high-throughput experimental environments, as well as on the types of experimental results that can and have been generated. It will be evident that comprehensive experimental designs and economies of scale are possible when applying these techniques in a production environment.
■
89
and genomic variants therein (Osoegawa et al., 2001). Other sequencing projects have employed expression sequence tag (EST) or cDNA sequencing of mRNA libraries to characterize variants primarily at the transcript, and by inference, at the protein level (Strausberg et al., 2003). All these approaches benefit from the random sampling of DNA regions by selecting bacterial colonies containing DNA inserts, DNA sequence determination using capillary-based electrophoresis of incorporated dye terminator nucleotide bases and subsequent sequence alignment of reads to construct reliable assemblies of genomic or mRNA molecules. Other approaches to DNA sequence elucidation that will be covered in this section involve the targeted amplification of defined genomic loci typically for study in large populations, the use of microfabricated high-throughput bead or surfacebased sequencing devices and high-density oligonucleotide arrays. A comparison of these distinct methodologies is provided in Table 7.1. Whole Genome Shotgun Sequencing Polymorphism discovery on a genome-wide level can be accomplished by applying the Sanger sequencing technique (Sanger et al., 1977) and the whole genome shotgun (WGS) strategy (Fleischmann et al., 1995; Venter et al., 2001; Venter et al., 1996) on a small pool of DNA samples. Owing to its high accuracy, versatility and ability to generate complete DNA sequence information, the Sanger sequencing method has been considered the definitive approach for DNA variant discovery. The WGS methodology relies upon generating a random sampling of genomic regions at a sufficiently deep coverage level to produce enough sequence data that can then be assembled by sequence alignment methods. First, multiple copies of the genome are randomly shredded into pieces of approximately 2–40 kilobasepairs (Kb) in size and subsequently cloned into a vector or plasmid. These constructs are then replicated in bacteria and sequenced from both ends to produce pairs of linked sequences, termed “mate pairs”, representing 500–800 bp at the end of each insert. These sequence reads are then assembled computationally to generate a set of contiguous high-quality regions with potential information regarding any associated sequence gaps of a known size. Currently, the Sanger-based WGS sequencing process, excluding the library construction step, is typically fully supported by equipment automation, as outlined in Figure 7.1, and managed as a production “pipeline” with varying degrees of integration between different steps employing a laboratory information management system.
DNA SEQUENCING DNA sequencing methods applied to either single or multiple DNA samples provide the definitive approach to discovering polymorphisms. These polymorphisms can be detected at the genomic and/or the mRNA level depending on the source of the material employed. For example, genomic DNA subcloned into BACs or in small to medium insert size bacterial libraries were created to produce a finished human genome sequence
Library Construction Genomic DNA is randomly sheared via nebulization to produce fragments with a distribution of approximately 2–40 Kb. The DNA fragments are ligated to adapters after end-polishing and size selection. After several rounds of purification, the fragments are inserted into plasmid vectors. The resulting library is electroporated into E. coli cells.
90
CHAPTER 7
TABLE 7.1
■
DNA Sequencing for the Detection of Human Genome Variation and Polymorphism
Comparison of distinct approaches in DNA sequence acquisition and the detection of variants Sanger based WGS
Targeted PCR
Sequence by synthesis
Sequence by hybridization
Current platforms
ABI 3730 xl
ABI 3730 xl
454, Illumina/Solexa, ABI SOLiD (strictly hybridization-ligation chemistry)
Affymetrix, Perlegen, Nimblegen
Read length (bp)
500–800
500–800
25–200
60–100
Detection SNP/Indel
Y/Y
Y/Y (up to amplicon size)
Y/Y
Y/Y (primarily deletions in sample)
Unbiased polymorphism detection
Y
Limited to targeted regions
Y
Limited to targeted regions
Biased polymorphism detection
N
Y
Y
Y
Construction of longrange haplotypes
Y
Limited to targeted regions
Uncertain
Limited to targeted regions
Current cost per sample $2 million
N
Uncertain
Y
Y
WGS Library
Figure 7.1
Clone picking and growth
Template preparation
Sequencing reaction set up and clean-up
Electrophoresis and sequence detection
Whole Genome Shotgun sequencing process flow.
Colony Plating and Picking The E. coli cells containing the plasmid vectors with inserted DNA fragments are then spread onto agar plates and incubated to form E. coli colonies. The colonies are picked into liquid media and incubated for further growth to generate sufficient numbers of plasmid template copies. Subsequently, these template copies are isolated and purified from the host cells. Template Preparation The alkaline lysis plasmid isolation approach (Sambrook et al., 1989) is the most widely utilized method for plasmid template isolation due to its robustness. Bacterial cells are lysed, cell debris removed by centrifugation and plasmid DNA recovered from the cleared lysate by isopropanol precipitation. Sequencing Reactions For DNA sequence determination, the most widely used technique is the Sanger dideoxy sequencing method (Sanger et al., 1977). It mimics the natural process of DNA replication in vivo with the presence of target DNA molecules, universal M13 sequencing primers, polymerase, deoxynucleotides (dNTPs) as well as chain terminating dideoxynucleotides which interrupt chain elongation upon incorporation generating DNA
fragments of different lengths. Several improvements have been made to the original method to greatly enhance its throughput, robustness and detection sensitivity. These include the development of multicolor fluorescent detection (Smith et al., 1986), fluorescently tagged chain-terminators (Prober et al., 1987), a cycle sequencing protocol (McCombie et al., 1992) and fluorescence resonance energy transfer (FRET) dyes (Ju et al., 1995; Lee et al., 1997). Currently, most sequencing reactions are carried out using FRET dye terminators, modified Taq polymerase and a cycle sequencing protocol. Sequence Detection The technology for the electrophoresis-based separation of Sanger sequenced DNA fragments and the detection of individual bases have advanced substantially in the past two decades. As a result, the throughput, detection sensitivity and costs of sequencing, as well as the accuracy and quality of the data generated by these sequencers, have improved significantly. The invention of multicolor fluorescent sequencing in 1986 (Smith et al., 1986) replaced laborious and hazardous radioisotope labeling techniques (Sanger et al., 1982) with automated signal detection, high-speed computational recording and signal processing. In the mid-1990s, the successful integration of the capillary-based
DNA Sequencing
T
I
R
■
91
O Consensus
reads
Figure 7.2 Display and detection of a single nucleotide polymorphism in a human genome assembly of a single DNA donor. Sequence reads overlap and can be assembled to form the consensus sequence of the contig where 5 of the 12 reads show a T variation at a C consensus base (variant position is highlighted with an arrow). T type of sequence read, either assembled contig or individual read, I unique identifier, R co-ordinates of alignment, O strand orientation that aligns.
electrophoresis and multicolor fluorescent detection (Bashkin et al., 1996a; Bashkin et al., 1996b; Behr et al., 1999) further enabled the creation of the true high-throughput sequencing machines. Today, state-of-the-art capillary-based sequencers have the ability to read over 1 million bp of sequence per 24-hour period with long read lengths (an average of 800 bp per read) and an average quality value (QV) of greater than 30 (99.9% accuracy). Whole Genome Assembly The whole genome assembly process utilizes several aspects of experimental design to ensure that unambiguous construction of long contiguous sequence can be generated. These include the generation of sequence reads from each end of clone inserts using universal primers pairs to ensure that a majority of inserts will have their ends sequenced. This enables both the generation of contiguous sequence and the potential for ordering and orienting larger sequence segments essentially created by sequence alignment of the individual reads whose mate pair relationship is known. A variety of genome assembler software packages have been designed with this basic rationale at their core (Celera Assembly (Myers et al., 2000), Phusion (Mullikin and Ning, 2003)). The Atlas assembler uses as input both BAC-based clone sequences and reads generated via the WGS strategy (Havlak et al., 2004), an approach that was used to assemble the rat genome (Gibbs et al., 2004). There are two basic steps to assembly: (i) the creation of unique regions of contiguous (contigs) assemblies from the sequence overlaps between reads and (ii) the sequential end-toend organization of contigs by employing the mate pair information contained in the reads constituting each contig.The successful application of this strategy was a significant challenge for the sequencing and assembly of human DNA since greater than 45%
of the genome sequence is repetitive in nature (Lander et al., 2001; Venter et al., 2001). The repetitive nature of the human and other mammalian genomes thus confounds the ability to determine accurate read sequence overlaps and the placement of contigs into the correct order and orientation. Our solution is to employ clone insert libraries of a size larger than the corresponding repeat regions, typically either fosmid or BAC libraries (40 kb and 100 kb respectively), thus enabling the spanning of repeats by assembling adjacent unique sequence (Venter et al., 2001). The selection of a range of insert sizes libraries is typically built into the experimental design when sequencing a new genome. Variant Detection Variants can be reliably detected from assembled sequence reads from either a single or multiple DNA donors in the region of unique genome sequence assembly. The example in Figure 7.2 show the reads assembled from a single human identifying a known SNP found in a coding region of the APOE gene and, as expected, displays a close to equiprobable occurrence of both allelic forms (five reads with the T allele and seven reads with C allele). This approach has been employed with success to identify 974,400 heterozygous variants in a single canine individual (Kirkness et al., 2003). The inclusion of random shotgun sequence reads from distinct DNA donors is another approach to employing genome assembly and read placement to identify polymorphic site. This approach was employed using five DNA donors in the initial shotgun sequence assembly of the human genome (Venter et al., 2001). An important factor in using a WGS assembly approach to detecting polymorphic sites is that enough sequence coverage from a single donor should be available. In this manner it is possible to provide a good quality assembly with associated consensus sequence and associated allelic variants. This means that
92
CHAPTER 7
■
DNA Sequencing for the Detection of Human Genome Variation and Polymorphism
additional donors can be included either post-assembly or as part of the assembly process if the species in question does not possess extensive interindividual polymorphism. Recent work attempting to detect both SNP and indels in an individual human sample suggests that detection of heterozygous variants, with 99% probability using a Sanger chemistry-based WGS approach, necessitates a variant being reported by 20 reads or more (Levy et al., 2007). Another challenge to polymorphism detection during genome assembly is the occurrence of many small heterozygous indel polymorphisms in human DNA, typically between 1–20 bp length (Ball et al., 2005). These indel events result in the occurrence of long and short alleles, requiring a methodology to accurately separate the reads from each allele, thereby providing an accurate representation of the indel. Subsequently, a practical decision needs to be made regarding which form will be represented in the consensus sequence. Typically this can be the “major” allele even though in a single donor this will be present effectively at equal probability. Lastly, a strategy needs to be implemented to best catalog the occurrence and type of the other allele or alleles, in the case of sequencing multiple DNA donor samples. An example of the detection and display of an indel from a genome assembly of a single DNA donor can be seen in Figure 7.3. The figure displays the consensus sequence (the contig sequence) containing the short, major, allele since gaps are included in the alignment, whilst the long form (inclusion of ATTCT) is found in two of the eight assembled reads. We recently applied a modified sequence assembly algorithm to detect heterozygous indels variants from the diploid genome sequence of an individual human (Levy et al., 2007). Using this approach we detected over 345,000 heterozygous
indels events of lengths 1–571 bp with at least 200 indels in protein-coding regions of genes. PCR Amplicon Re-Sequencing The goal of directed sequencing or re-sequencing is to use the existing sequence data from a particular genomic locus or loci to determine sequence variations of these same loci in different DNA samples. This provides the nature and the frequency of DNA polymorphisms and can be used to assess whether these variants could be implicated in, for example, either disease onset or progression. PCR re-sequencing methodology is widely accepted as the gold standard (Kwok and Chen, 2003) for the discovery of polymorphisms (both SNPs and indels) in targeted regions since it provides the most complete information including the genotype, the location and sequence context.This approach has been applied to understand sequence variation in populations (Crawford et al., 2004) and for the identification of variants and haplotypes that could potentially explain disease phenotypes in cancer (Cox et al., 2005). It combines the powerful PCR technique (Mullis et al., 1986) with the informative Sanger sequencing method (Sanger et al., 1977). Rapid advancement of the technologies and dramatic reduction in costs for direct sequencing of PCR products, as well as the completion of the reference human genomes in 2001 (Lander et al., 2001;Venter et al., 2001) has enabled the systematic large-scale discovery of polymorphisms in targeted regions. The approach generally involves PCR primer design, the amplification of the targeted region from whole genomic DNA using PCR, followed by Sanger sequencing and the analysis of the targeted sequences (Figure 7.4). T
I
R
O Consensus
reads
2 reads with 5 bp insertion ATTCT
Figure 7.3 Display and detection of a 5 bp insertion/deletion polymorphism in a human genome assembly of a single DNA donor. The consensus sequence displays the “short” allelic form whilst the “long” form (ATTCT) is found in 2 of the 8 sequence assembled reads. T type of sequence read, either assemblyed contig or individual read, I unique identifer, R co-ordinates of alignment, O strand orientation that aligns. “” gap inserted into read to permit correct alignment.
PCR primer design
Figure 7.4
PCR reaction and clean-up
PCR directed sequencing process flow.
Sequencing reaction and clean-up
Electrophoresis and trace analysis
DNA Sequencing
Primer Design The goal of directed sequencing is to provide reliable amplification of a single genomic region, and this is readily achieved via the design of unique pairs of oligomers for priming of the PCR. The resulting amplicon encompassing the targeted region can vary in size; however reproducible and robust assays range from 350 bp to 1000 bp in length per amplicon. In order to provide coverage of a larger genomic locus with this amplicon size, it becomes necessary to tile PCR products with some overlap (typically 100 bp) between amplicons. The primer design process attempts to generate oligomers for products with similar melting temperatures whilst avoiding primer-dimerization forming interfering secondary structures. Other factors to permit specific amplification would be the GC content of the region, the proximity of repeat regions, the uniqueness of the oligomer sequences in the genome and the absence of known DNA variants in the primer binding sites. These considerations can be embodied in a computational pipeline as illustrated in Figure 7.5. The sequence context, in this case percent GC quantity, are considered before identifying a candidate genomic region for primer design. In a high-throughput directed sequencing pipeline, it is frequently desirable to employ few, or preferable one, PCR protocol, and in our experience GC content allows a distinction between at least two desired PCR conditions (high and low GC content). Once the PRIMER3 software (Rozen and Skaletsky, 2000) generates candidate primer pairs for a desired target region, these are further
Gene/region of interest
Pre-Compute/Load GC content
Generate 1st tile
Decrease tile size change 5 tile position Fail Generate next tile position
Fail
Primer3
Critique primer pair for genome specificity. Check for: Homopolymers Known SNP/indels Repeat regions Pass Order primers
Figure 7.5 Primer design pipeline for PCR based directed sequencing.
■
93
assessed for their product uniqueness (low and high copy repeat regions) and to ensure that resulting amplicons contains only one type of complex sequence, i.e. homopolymer or known indel polymorphism. The last requirement is important since sequence detection is performed using capillary-based detection methods and chromatograms potentially contain a mixture of alternate alleles at any one mobility position. This renders the process of allelic definition potentially difficult in the case of multiple indel events in a single amplicon. PCR Set-Up and Clean-Up As described in the primer design section, due to the variable GC content within the genome, application of different PCR primer design criteria, as well as different PCR protocols might be necessary when targeting different regions. In our laboratory, we have two fully validated high-throughput amplification protocols (Rand et al., 2005), one for regions of normal GC content and another for amplicons with melting temperatures (Tm) 82°C. AmpliTaq Gold® DNA polymerase (Applied Biosystems) is used to perform hot-start PCR to minimize the formation of primer–dimers and spurious amplicons and ensure efficient, consistent and specific amplification results in a highthroughput environment. A Shrimp Alkaline Phosphatase/Exonuclease I (SAP/Exo I) mix is used to digest excess dNTPs and amplification primers. (While SAP is commonly used to remove 5 phosphate groups from DNA, it also degrades dNTPs.) The SAP/Exo I method provides an effective, single tube approach to PCR clean-up that is easily automated and scalable. Sequencing Reaction Set-Up and Clean-Up In order to simplify the downstream sequencing process, the PCR primers are frequently tagged with M13 forward and reverse sequences on their 5 ends. The resulting amplicons can then be sequenced using universal M13 sequencing primer sets and a standard sequencing protocol (see previous section). Reaction products can be purified with sodium acetate and ethanol prior to analysis on the sequencers. Electrophoresis and Trace Analysis The state-of-the-art technology for the electrophoresis-based separation of Sanger sequenced DNA fragments and the detection of individual bases can be accomplished by a combination of trace filtering, trace separation and mixed base calling. Electropherogram peaks are described as “mixed” in that a primary peak is detected with a secondary but significantly present sub-peak due to two distinct and different allelic forms contributing to the locus, We have developed a trace analysis pipeline that applies digital signal filters to reduce the occurrence of secondary, typically noisy, sequence peaks (Figure 7.6), thus enabling the accurate distinction of a single position with mixed peaks. This approach has enabled a substantial improvement of common artifacts observed in diploid genomic regions sequenced after PCR amplification that include dye-blob, PCR slippage
94
CHAPTER 7
■
DNA Sequencing for the Detection of Human Genome Variation and Polymorphism
Unfiltered trace
Filtered trace
Mixed base Figure 7.6 Sequence chromatograms of a locus revealing a SNP before and after digital signal filtering were applied to remove PCR slippage artifacts due to the presence of proximal (not seen) homopolymeric DNA.
and primer synthesis length errors, all of which confound mixed peak detection. After generating either a raw (unfiltered) or digital signal filtered chromatogram, it is then possible to identify mixed peaks that vary significantly in height, area and width to characterize departures from the reference sequence. This general approach can be performed with many different software tools in the public domain (Polyphred, Mutation Surveyor, novoSNP and SNPdetector, InSNP), the most extensively tested of which is Polyphred. In the most recent version of this software, detailed parameterization was performed using sequence reads from PCR-amplified diploid samples and a statistical framework was developed to report evidence of genotypes being called (Stephens et al., 2006). Heterozygous indel polymorphisms result in significant changes in the chromatograms from diploid DNA samples where typically traces contain mixed peaks from the point of the insertion/deletion onwards. One can develop computational methods that essentially separate the traces to enable the definition of the long and short alleles (Bhangale et al., 2005; Bhangale et al., 2006). Polyphred, InSNP (Manaster et al., 2005) and Mutation Surveyor are all capable of detecting heterozygous indels, but as of this writing an unbiased detailed comparison of performance has not yet been attempted. Sequencing by Synthesis Recently implemented sequencing by synthesis technologies can employ PCR (Mullis et al., 1986) to achieve clonal amplification of single DNA molecules. In addition, the sequencing by synthesis principle in a massively parallel format has been developed by Rothberg and colleagues (Margulies et al., 2005) and by Church and colleagues (Shendure et al., 2005). In each of these new technologies, different chemistries are applied to interrogate, tag and detect the sequence bases. For example, utilizing nanotechnology and pyrosequencing chemistry (Nyren et al., 1993), coupled with an enzymatic luminometric inorganic pyrophosphate detection technique, Rothberg and colleagues at 454 Life Sciences recently developed a novel and highly parallel system capable of sequencing approximately 100 million bp in a 7-hour period by sampling over 400,000 sequence clones. They were able to illustrate this technological
advance by sequencing and assembling the Mycoplasma genitalium and Streptococcus pneumoniae genomes (0.6 and 2.1 Mb respectively in size). Compared to the current state-of-the-art Sanger DNA sequencing and capillary-based electrophoresis platform, their system generates “raw” data with approximately 100 times higher throughput. However, compared to the Sanger/electrophoresis-based sequencing approach, it still has several limitations, such as the short sequence read lengths which, averaging 200–250 bases per read, are only a quarter of Sanger read lengths. Another problem is that the accuracy of base calling on individual reads is low especially for genomic regions containing homopolymers. This leads to the increased likelihood of spurious polymorphism detection. In addition, the 454 sequencing methodology currently does not incorporate the use of mate pairs that come from large insert libraries (10 kb), which for the successful assembly of repeat rich genomes, like those of most eukaryotes, are important in order to span large repeat regions. Nonetheless, this technology has been employed to generate some sequence data from DNA extracted from fossil remains (Green et al., 2006) and has been employed most recently to generate sixfold coverage of an individual genome (Wheeler et al., 2008). Some of these problems can be circumvented by adopting a hybrid sequencing approach, that is one that employs, in addition to the sequencing-by-synthesis approach (454), a small number of large insert clones (typically fosmid libraries) sequenced by Sanger technology to provide order and orientation in the final genome assembly.While this approach has been successfully employed for a number of bacterial genomes (Goldberg et al., 2006), establishing the most efficient mix of the sequence-by-synthesis contribution and fosmid libraries will be the challenge for the completion of larger mammalian genomes. Each of the sequencing-by-synthesis approaches mentioned above share similar advantages as well as limitations. However, they do show great promise for the discovery of polymorphisms in a rapid, cost effective manner. The massively parallel sequencing-by-synthesis approach enabled by the 454 picoliter plate technology can also be utilized for targeted polymorphism discovery. In this manner, PCR amplicons that contain the targeted genomic regions are extensively sampled since it is possible to generate currently up to 400,000 clonal amplifications of the PCR product per experiment. This
Other Methodologies for Polymorphism Detection
approach has potentially great value for the discovery and detection of DNA variations present in low frequency in the samples. This is especially applicable for mutation detection in tumor tissue, in which potentially only a few cells carry activating mutations and the surrounding cell population contributes to the “contamination” of the tumor DNA. The lowest detection level thus far obtained has been for alleles present in the sample as low as 9% for indel mutations in the EGFR gene in lung adenocarcinoma specimens (Thomas et al., 2006) and at a similar frequency for non-synonymous point mutations in the FGFR1 gene from glioblastoma samples (Strausberg et al., personal communication). In both of these cases the corresponding PCRamplified, Sanger sequence chromatograms lack the necessary secondary peak signal to detect a variant above the background noise level, where the limit of detection of a secondary allele is thought to be between 15–25%. Sequencing by Hybridization The sequencing by hybridization on microarray approach has also become a widely utilized tool for SNP detection and analysis (Kukita et al., 2005; Patil et al., 2001; Maraganore et al., 2005; Kozal, 1996; Cronin et al., 1996; Chee et al., 1996; Hacia et al., 1996; Kapranov et al., 2002) due to the technological advances that make the manufacturing of high-density DNA arrays possible The genomic DNA is generally prepared by PCR to amplify the regions of interest or by using a whole genome amplification methodology (Dean et al., 2002) and tagged with fluorescent dyes for detection prior to hybridizing to the oligonucleotide array. This approach allows the detection of sequence present in a test sample via direct hybridization and was able to detect the unique (non-repeat) contiguous sequence from human DNA chromosome 21 conserved with mouse and dog (Frazer et al., 2001) and the haplotype block structure of chromosome 21 in 24 human DNA samples (Patil et al., 2001). In both cases, this was accomplished by creating a tiling array where each position was interrogated for all possible bases using a set of eight distinct oligonucleotide probes comprising the position in question and 12 bp flanking to provide for binding specificity. The interspecies hybridization experiments revealed additional putative exonic region of current genes and identified other regions of significant sequence conservation that could represent other forms of functional sequence. The human population study revealed that 80% of the DNA samples could be described by three common haplotypes. Such knowledge of chromosomal block structure and inheritance patterns enables the choice of a smaller set of variants with which one can perform disease-gene association studies (see Chapter 8). The sequencing-by-hybrization approach clearly necessitates that all the sequence being tested is present on the microarray. While this may be a limitation for discovering novel sequence in some genomes, even for the multiple, yet to be sequenced, human genomes, the methodology is well suited and has been extensively employed for the genotyping of known SNPs in disease conditions and deletions in arrayed sequence. Recent whole genome association studies
■
95
employing large-scale SNP assays performed on microarrays have identified genes as either biomarkers or therapeutic targets in Parkinson’s disease (200,000 loci) (Maraganore et al., 2005) and copy number variations and deletions in mental retardation (100,000 loci) (Friedman et al., 2006).
OTHER METHODOLOGIES FOR POLYMORPHISM DETECTION Many other methodologies have been developed for genomewide discovery and characterization of DNA polymorphisms. Although each of the methodologies has their own unique capability in detecting various types of polymorphisms, they tend to have limited utility and are often employed at a lower scale than the approaches described above. Restriction Fragment Length Polymorphism One of the most widely utilized techniques is Restriction Fragment Length Polymorphisms (RFLP). RFLPs are polymorphisms in the lengths of particular restriction fragments. It was one of the first marker systems developed and has been used in creating genetic maps of chromosomes (Botstein et al., 1980; Rothschild et al., 1993), tracking changes occurring in the cancer genome (Jenkins et al., 2002), as well as serving as a starting points for chromosome walks (Bakker et al., 1985). The technology is based on using restriction endonucleases that cleave DNA at highly specific sites; therefore mutations or polymorphisms change the enzymatic activity at precise genomic loci. Thus by measuring the differences in length of the cleaved DNA strands (accomplished by electrophoresis separation and probing with cloned fragments of the genome), it is possible to monitor the presence of genetic variations. While reliable, this method’s detection power is limited to the small number of the bases around the restriction sites and can only detect the presence and not the identities of the sequences variations. In the simplest and most common type of RFLP, the polymorphism results from a single nucleotide difference that provides a recognition site for a restriction enzyme in one allelic form and not the other. A polymorphism of this type, once identified and characterized by sequencing, can be detected readily among a population of individuals in a targeted fashion by first amplifying the region around the polymorphic site from each sample, digesting the amplified material with the appropriate restriction enzyme and distinguishing the undigested PCR product from the smaller digested fragments by gel electrophoresis (Saiki et al., 1985). This PCR-based protocol provides results rapidly and is a significant improvement over the Southern blot alternative which requires blotting, probe labeling, hybridization and autoradiography. Simple Sequence Repeat Polymorphisms Simple Sequence Repeat Polymorphisms (SSRPs), generally referred to as “microsatellite” markers, are tandem repeats units of short sequences of nucleotides (generally 2–6 units in
96
CHAPTER 7
■
DNA Sequencing for the Detection of Human Genome Variation and Polymorphism
length). The detection of these markers can be readily accomplished by probing the genomic DNA libraries with synthetic oligonucleotides followed by confirmation sequencing of the positive clones with the Sanger technique. PCR primers can be designed to amplify the markers once the sequences flanking the SSRP markers are known. The variations in the length of PCR products can then be separated by electrophoresis and detected (Cornall et al., 1991; Love et al., 1990). This method is relatively easy to automate at a low cost for large scale genotyping. It can be very informative for fingerprinting applications; however, its utility is obviously limited to the detection of length polymorphisms. Chromatography or Electrophoresis Separation Methods for Conformation Polymorphism Detection The natural intra- or inter-molecular structural distortions (conformational changes) of the DNA double helix induced by mispairing of the bases could result in detectable or measurable physical properties changes that can be used to detect the presence and the approximate locations of polymorphisms. These detectable or measurable physical property changes include melting temperature and mobility shift. These methods generally utilize separation techniques such as electrophoresis or chromatography to resolve conformational differences among DNA samples. For example, the Denaturing Gradient Gel Electrophoresis (Myers et al., 1987; Burmeister et al., 1991) method takes advantage of the fact that the denaturing of double stranded DNA is highly dependent on its sequence. One base mispairing could induce conformational changes and alter the melting temperature of a DNA molecule, thereby suggesting the presence of an allelic variant in a particular locus. The DNA molecules with different melting temperatures show distinguishable motility shifts under denaturing conditions for electrophoresis. Other methods utilizing these properties include Denaturing High-Performance Liquid Chromatography (Huber et al., 2001; Underhill et al., 1997), Single Strand Conformation Polymorphism (Nataraj et al., 1999; Sheffield et al., 1993) and Heteroduplex Analysis (Nataraj et al., 1999; Thomas et al., 2001). The main drawback of these methods is that they can only be used to detect the presence and not the identities of the polymorphisms. Other Enzymatic and Chemical Cleavage Methods Cleavage techniques also exploit the conformational changes of the DNA double helix caused by mispairing of the bases. Several approaches utilizing the different chemical and enzymatic properties of the perfectly matched homoduplexes and the mismatched heteroduplexes for targeted polymorphism detection in specific regions have been developed. These methods include chemical cleavage of mismatched DNA (Curiel et al., 1990; Lambrinakos et al., 1999; Tabone et al., 2006), ribonuclease cleavage of mismatched DNA (Myers et al., 1985), cleavase fragment length polymorphism analysis (Nomura et al., 2001; Uehara et al., 1999), T4 Endonuclease VII cleavage of heteroduplex DNA (Babon et al., 1999), mismatch repair detection (Faham et al., 2005),
UNG-Mediated T-Sequencing (Hawkins and Hoffman, 1997), and RNA-mediated fingerprinting with MALDI MS Detection (Krebs et al., 2003; Stanssens et al., 2004). The detection sensitivity and accuracy of these techniques highly depend on the specificity and efficiency of the particular enzyme activity or chemical used. Ideally, little or no cleavage would be seen in a perfectly matched DNA fragment and a near complete digestion or modification of all distortions of the helix generated by base mismatches would be accomplished. However, none of the techniques mentioned above are able to detect both the matched and mismatched DNA states equally well. The utility and the choice of a technique becomes a balance between ease of use, sensitivity and specificity. Some of these methods can be applied directly to extracted genomic DNA if the quantity of the DNA is not an issue. However, PCR is commonly used to select the targeted genomic region and generate enough DNA material for the assays. The separation and detection of the digested or modified DNA fragments is generally accomplished through electrophoretic techniques.
FUTURE DIRECTIONS The generation of a first draft of the human genome sequence permitted the global identification of small scale variation such as SNPs and heterozygous indels. However, it also set the stage for detection of large scale genome-to-genome comparisons (Istrail et al., 2004). These and other such studies (Khaja et al., 2006) have identified large-scale difference (1–3 Mb ranges) between human individuals that have been confirmed in larger populations (Redon et al., 2006). Recent studies have also implicated such large scale polymorphism (copy number variants) in developmental verbal dyspraxia (Feuk et al., 2006) and Crohn’s disease (Fellermann et al., 2006), to name but a few. The complete human genome sequence has also fueled the development of newer sequencing technologies that employ short sequence reads but generate large quantities of sequence data at increasingly lower cost without the need of elaborate production facilities. These approaches that involved the interrogation of single DNA molecules via clonal amplication (454, Solexa, ABI-Agencourt) will permit de novo sequencing of distinct human individuals. However de novo human genome assembly is currently not attainable due to the short read length, the current absence of well developed methodologies for generating large insert clone mate sequence and computational approaches that account for both the error profile and the input characteristics of reads generated. These new sequencing platforms do permit the generation of sequence data from defined genomic regions of multiple individuals via PCR and other methods, to allow the targeted amplification or separation of genomic loci (Thomas et al., 2005;Thomas et al., 2006). The goal for sequencing in the next decade will be to achieve de novo assembly of multiple human genomes, since the mapping of newly generated reads to an existing human genome reference constrain new sequence data to an existing, likely unique, genome structure. Therefore establishing de novo assembly approaches in the analysis of future genomes will create
References
an unbiased understanding of human chromosomal structure and the underlying individual genotypes that define each person uniquely in the context of growing phenotypic information. Lastly, any human DNA variation will need to be interpreted in the context of the functional sequence that it potentially modulates. This appears to be an easier problem to address when considering a SNP in a protein-coding domain (Livingston et al., 2004; Ng and Henikoff, 2006), and computational prediction methods can rank which putative functional changes are to be tested experimentally. These prediction methods are less defined for polymorphisms larger than 1 bp (2–100 bp) that occur distal to known protein-coding domains, and clearly much of the impact on human disease may be modulated by these variants (Ball et al., 2005; Chuzhanova et al., 2003). More detailed recent analyses of 1% of the human genome sequence revealed that many transcriptional signals and cues reside in non-coding DNA and a more significant fraction of the human genome is functional than thought previously
■
97
(Birney et al., 2007). This indicates that a large amount of DNA variation that has been identified in non-coding DNA (Levy et al., 2007) may elicit a variety of phenotypic effects, whose elucidation will be a rich source for future research. Much of the human genetic and genomic research since the completion of the human genome sequence has led us to conclude that much still remains to be discovered in primary sequence space as a predecessor to our understanding of basic functional changes in cells and tissues. Such knowledge is crucial so that we can bring genomic data to bear in a useful manner to the medical context.
ACKNOWLEDGEMENTS The authors would like to thank Drs Robert Strausberg, Jiaqi Huang and Pauline Ng for their suggestions and critical review of the manuscript.
REFERENCES Babon, J.J., McKenzie, M. and Cotton, R.G. (1999). Mutation detection using fluorescent enzyme mismatch cleavage with T4 endonuclease VII. Electrophoresis 20, 1162–1170. Bakker, E., Hofker, M.H., Goor, N., Mandel, J.L., Wrogemann, K., Davies, K.E., L.M., K., Willard, H.F., Fenton, W.A., Sandkuyl, L. et al. (1985). Prenatal diagnosis and carrier detection of Duchenne muscular dystrophy with closely linked RFLPs. Lancet 1, 655–658. Ball, E.V., Stenson, P.D., Abeysinghe, S.S., Krawczak, M., Cooper, D.N. and Chuzhanova, N.A. (2005). Microdeletions and microinsertions causing human genetic disease: Common mechanisms of mutagenesis and the role of local DNA sequence complexity. Hum Mutat 26, 205–213. Bashkin, J., Marsh, M., Barker, D. and Johnston, R. (1996a). DNA sequencing by capillary electrophoresis with a hydroxyethylcellulose sieving buffer. Appl Theor Electrophor 6, 23–28. Bashkin, J.S., Bartosiewicz, M., Roach, D., Leong, J., Barker, D. and Johnston, R. (1996b). Implementation of a capillary array electrophoresis instrument. J Capillary Electrophor 3, 61–68. Behr, S., Matzig, M., Levin, A., Eickhoff , H. and Heller, C. (1999). A fully automated multicapillary electrophoresis device for DNA analysis. Electrophoresis 20, 1492–1507. Bhangale, T.R., Rieder, M.J., Livingston, R.J. and Nickerson, D.A. (2005). Comprehensive identification and characterization of diallelic insertion-deletion polymorphisms in 330 human candidate genes. Hum Mol Genet 14, 59–69. Bhangale, T.R., Stephens, M. and Nickerson, D.A. (2006). Automating resequencing-based detection of insertion-deletion polymorphisms. Nat Genet 38, 1457–1462. Birney, E., Stamatoyannopoulos, J.A., Dutta, A., Guigo, R., Gingeras, T.R., Margulies, E.H., Weng, Z., Snyder, M., Dermitzakis, E.T., Thurman, R.E. et al. (2007). Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447, 799–816. Botstein, D., White, R., Skolnick, M. and Davis, R.W. (1980). Construction of a genetic linkage map in man using restriction fragment length polymorphisms. Am J Hum Genet 32, 314–331.
Burmeister, M., diSibio, G., Cox, D.R. and Myers, R.M. (1991). Identification of polymorphisms by genomic denaturing gradient gel electrophoresis: Application to the proximal region of human chromosome 21. Nucleic Acid Res 19, 1475–1481. Chee, M., Yang, R., Hubbell, E., Berno, A., Huang, X.C., Stern, D., Winkler, J., Lockhart, D.J., Morris, M.S. and Fodor, S.P. (1996). Accessing genetic information with high-density DNA arrays. Science 274, 610–614. Chuzhanova, N.A., Anassis, E.J., Ball, E.V., Krawczak, M. and Cooper, D.N. (2003). Meta-analysis of indels causing human genetic disease: Mechanisms of mutagenesis and the role of local DNA sequence complexity. Hum Mutat 21, 28–44. Consortium, The International HapMap (2005). A haplotype map of the human genome. Nature 437, 1299–1320. Cornall, R.J., Aitman, T.J., Hearne, C.M. and Todd, J.A. (1991). The generation of a library of PCR-analyzed microsatellite variants for genetic mapping of the mouse genome. Genomics 10, 874–881. Cox, D.G., Kraft, P., Hankinson, S.E. and Hunter, D.J. (2005). Haplotype analysis of common variants in the BRCA1 gene and risk of sporadic breast cancer. Breast Cancer Res 7, R171–R175. Crawford, D.C., Carlson, C.S., Rieder, M.J., Carrington, D.P., Yi, Q., Smith, J.D., Eberle, M.A., Kruglyak, L. and Nickerson, D.A. (2004). Haplotype diversity across 100 candidate genes for inflammation, lipid metabolism, and blood pressure regulation in two populations. Am J Hum Genet 74, 610–622. Cronin, M.T., Fucini, R.V., Kim, S.M., Masino, R.S., Wespi, R.M. and Miyada, C.G. (1996). Cystic fibrosis mutation detection by hybridization to light-generated DNA probe arrays. Hum Mutat 7, 244–255. Curiel, D.T., Buchhagen, D.L., Chiba, I., D’Amico, D., Takahashi, T. and Minna, J.D. (1990). A chemical mismatch cleavage method useful for the detection of point mutations in the p53 gene in lung cancer. Am J Respir Cell Mol Biol 3, 405–411. Davies, J., Alton, E. and Griesenbach, U. (2005). Cystic fibrosis modifier genes. J R Soc Med 98 Suppl 45, 47–54.
98
CHAPTER 7
■
DNA Sequencing for the Detection of Human Genome Variation and Polymorphism
Dean, F.B., Hosono, S., Fang, L., Wu, X., Faruqi, A.F., Bray-Ward, P., Sun, Z., Zong, Q., Du, Y., Du, J. et al. (2002). Comprehensive human genome amplification using multiple displacement amplification. Proc Natl Acad Sci USA 99, 5261–5266. Faham, M., Zheng, J., Moorhead, M., Fakhrai-Rad, H., Namsaraev, E., Wong, K., Wang, Z., Chow, S.G., Lee, L., Suyenaga, K. et al. (2005). Multiplexed variation scanning for 1,000 amplicons in hundreds of patients using mismatch repair detection (MRD) on tag arrays. Proc Natl Acad Sci USA 102, 14717–14722. Fan, J.B., Chee, M.S. and Gunderson, K.L. (2006). Highly parallel genomic assays. Nat Rev Genet 7, 632–644. Fellermann, K., Stange, D.E., Schaeffeler, E., Schmalzl, H., Wehkamp, J., Bevins, C.L., Reinisch, W., Teml, A., Schwab, M., Lichter, P. et al. (2006). A chromosome 8 gene-cluster polymorphism with low human beta-defensin 2 gene copy number predisposes to Crohn disease of the colon. Am J Hum Genet 79, 439–448. Feuk, L., Kalervo, A., Lipsanen-Nyman, M., Skaug, J., Nakabayashi, K., Finucane, B., Hartung, D., Innes, M., Kerem, B., Nowaczyk, M.J. et al. (2006). Absence of a paternally inherited FOXP2 gene in developmental verbal dyspraxia. Am J Hum Genet 79, 965–972. Fleischmann, R.D.,Adams, M.D.,White, O., Clayton, R.A., Kirkness, E.F., Kerlavage,A.R., Bult, C.J.,Tomb, J.F., Dougherty, B.A., Merrick, J.M. et al. (1995). Whole-genome random sequencing and assembly of Haemophilus influenzae Rd. Science 269, 496–512. Frazer, K.A., Sheehan, J.B., Stokowski, R.P., Chen, X., Hosseini, R., Cheng, J.F., Fodor, S.P., Cox, D.R. and Patil, N. (2001). Evolutionarily conserved sequences on human chromosome 21. Genome Res 11, 1651–1659. Friedman, J.M., Baross, A., Delaney, A.D., Ally, A., Arbour, L., Asano, J., Bailey, D.K., Barber, S., Birch, P., Brown-John, M. et al. (2006). Oligonucleotide microarray analysis of genomic imbalance in children with mental retardation. Am J Hum Genet 79, 500–513. Gibbs, R.A., Weinstock, G.M., Metzker, M.L., Muzny, D.M., Sodergren, E.J., Scherer, S., Scott, G., Steffen, D., Worley, K.C., Burch, P.E. et al. (2004). Genome sequence of the Brown Norway rat yields insights into mammalian evolution. Nature 428, 493–521. Goldberg, S.M., Johnson, J., Busam, D., Feldblyum, T., Ferriera, S., Friedman, R., Halpern, A., Khouri, H., Kravitz, S.A., Lauro, F.M. et al. (2006). A Sanger/pyrosequencing hybrid approach for the generation of high-quality draft assemblies of marine microbial genomes. Proc Natl Acad Sci USA 103, 11240–11245. Green, R.E., Krause, J., Ptak, S.E., Briggs, A.W., Ronan, M.T., Simons, J.F., Du, L., Egholm, M., Rothberg, J.M., Paunovic, M. et al. (2006). Analysis of one million base pairs of Neanderthal DNA. Nature 444, 330–336. Hacia, J.G., Brody, L.C., Chee, M.S., Fodor, S.P. and Collins, F.S. (1996). Detection of heterozygous mutations in BRCA1 using high density oligonucleotide arrays and two-colour fluorescence analysis. Nat Genet 14, 441–447. Havlak, P., Chen, R., Durbin, K.J., Egan, A., Ren, Y., Song, X.Z., Weinstock, G.M. and Gibbs, R.A. (2004). The Atlas genome assembly system. Genome Res 14, 721–732. Hawkins, G.A. and Hoffman, L.M. (1997). Base excision sequence scanning. Nat Biotech 15, 803–804. Huber, C.G., Premstaller, A., Xiao, W., Oberacher, H., Bonn, G.K. and Oefner, P.J. (2001). Mutation detection by capillary denaturing high-performance liquid chromatography using monolithic columns. J Biochem Biophys Meth 47, 5–19.
Istrail, S., Sutton, G.G., Florea, L., Halpern, A.L., Mobarry, C.M., Lippert, R., Walenz, B., Shatkay, H., Dew, I., Miller, J.R. et al. (2004). Whole-genome shotgun assembly and comparison of human genome assemblies. Proc Natl Acad Sci USA 101, 1916–1921. Jenkins, G.J.,Williams, G.L., Beynon, J.,Ye, Z., Baxter, J.N. and Parry, J.M. (2002). Restriction enzymes in the analysis of genetic alterations responsible for cancer progression. Br J Surg 89, 8–20. Ju, J., Ruan, C., Fuller, C.W., Glazer, A.N. and Mathies, R.A. (1995). Fluorescence energy transfer dye-labeled primers for DNA sequencing and analysis. Proc Natl Acad Sci USA 92, 4347–4351. Kapranov, P., Cawley, S.E., Drenkow, J., Bekiranov, S., Strausberg, R.L., Fodor, S.P. and Gingeras, T.R. (2002). Large-scale transcriptional activity in chromosomes 21 and 22. Science 296, 916–919. Khaja, R., Zhang, J., Macdonald, J.R., He, Y., Joseph-George, A.M., Wei, J., Rafiq, M.A., Qian, C., Shago, M., Pantano, L. et al. (2006). Genome assembly comparison identifies structural variants in the human genome. Nat Genet 38, 1413–1418. Kirkness, E.F., Bafna, V., Halpern, A.L., Levy, S., Remington, K., Rusch, D.B., Delcher, A.L., Pop, M., Wang, W., Fraser, C.M. et al. (2003). The dog genome: Survey sequencing and comparative analysis. Science 301, 1898–1903. Klein, R.J., Zeiss, C., Chew, E.Y., Tsai, J.Y., Sackler, R.S., Haynes, C., Henning, A.K., SanGiovanni, J.P., Mane, S.M., Mayne, S.T. et al. (2005). Complement factor H polymorphism in age-related macular degeneration. Science 308, 385–389. Kozal, M.J. et al. (1996). Extensive polymorphisms observed in HIV-1 clade B protease gene using high-density oligonucleotide arrays. Nat Med 2, 753–759. Krebs, S., Medugorac, I., Seichter, D. and Förster, M. (2003). RNaseCut: A MALDI mass spectrometry-based method for SNP discovery. Nucleic Acids Res 31, e37. Kukita, Y., Miyatake, K., Stokowski, R., Hinds, D., Higasa, K.,Wake, N., Hirakawa, T., Kato, H., Matsuda, T., Pant, K. et al. (2005). Genomewide definitive haplotypes determined using a collection of complete hydatidiform moles. Genome Res 15, 1511–1518. Kwok, P.Y. and Chen, X. (2003). Detection of single nucleotide polymorphisms. Curr Issues Mol Biol 5, 43–60. Laitinen, T., Polvi, A., Rydman, P., Vendelin, J., Pulkkinen, V., Salmikangas, P., Makela, S., Rehn, M., Pirskanen, A., Rautanen, A. et al. (2004). Characterization of a common susceptibility locus for asthma-related traits. Science 304, 300–304. Lambrinakos, A., Humphrey, K.E., Babon, J.J., Ellis, T.P. and Cotton, R.G. (1999). Reactivity of potassium permanganate and tetraethylammonium chloride with mismatched bases and a simple mutation detection protocol. Nucleic Acids Res 27, 1866–1874. Lander, E.S., Linton, L.M., Birren, B., Nusbaum, C., Zody, M.C., Baldwin, J., Devon, K., Dewar, K., Doyle, M., FitzHugh, W. et al. (2001). Initial sequencing and analysis of the human genome. Nature 409, 860–921. Lee, L.G., Spurgeon, S.L., Heiner, C.R., Benson, S.C., Rosenblum, B.B., Menchen, S.M., Graham, R.J., Constantinescu, A., Upadhya, K.G. and Cassel, J.M. (1997). New energy transfer dyes for DNA sequencing. Nucleic Acids Res 25, 2816–2822. Levy, S., Sutton, G., Ng, P.C., Feuk, L., Halpern, A.L., Walenz, B.P., Axelrod, N., Huang, J., Kirkness, E.F., Denisov, G. et al. (2007). The Diploid Genome Sequence of an Individual Human. PLoS Biol 5, e254.
References
Livingston, R.J., von Niederhausern, A., Jegga, A.G., Crawford, D.C., Carlson, C.S., Rieder, M.J., Gowrisankar, S., Aronow, B.J., Weiss, R.B. and Nickerson, D.A. (2004). Pattern of sequence variation across 213 environmental response genes. Genome Res 14, 1821–1831. Love, J.M., Knight, A.M., McAleer, M.A. and Todd, J.A. (1990). Towards construction of a high resolution map of the mouse genome using PCR-analysed microsatellites. Nucleic Acids Res 18, 4123–4130. Manaster, C., Zheng, W.,Teuber, M., Wachter, S., Doring, F., Schreiber, S. and Hampe, J. (2005). InSNP: A tool for automated detection and visualization of SNPs and InDels. Hum Mutat 26, 11–19. Maraganore, D.M., de Andrade, M., Lesnick, T.G., Strain, K.J., Farrer, M.J., Rocca, W.A., Pant, P.V., Frazer, K.A., Cox, D.R. and Ballinger, D.G. (2005). High-resolution whole-genome association study of Parkinson disease. Am J Hum Genet 77, 685–693. Margulies, M., Egholm, M., Altman, W.E., Attiya, S., Bader, J.S., Bemben, L.A., Berka, J., Braverman, M.S., Chen, Y.J., Chen, Z. et al. (2005). Genome sequencing in microfabricated high-density picolitre reactors. Nature 437, 376–380. McCombie, W.R., Heiner, C., Kelley, J.M., Fitzgerald, M.G. and Gocayne, J.D. (1992). Rapid and reliable fluorescent cycle sequencing of double-stranded templates. DNA Seq 2, 289–296. Mullikin, J.C. and Ning, Z. (2003). The phusion assembler. Genome Res 13, 81–90. Mullis, K.B., Faloona, F.A., Scharf, S.J., Saiki, R.K., Horn, G.T. and Erlich, H.A. (1986). Specific enzymatic amplification of DNA in vitro: The polymerase chain reaction. Cold Spring Harbor Symp Quant Biol 51, 263–273. Myers, E.W., Sutton, G.G., Delcher, A.L., Dew, I.M., Fasulo, D.P., Flanigan, M.J., Kravitz, S.A., Mobarry, C.M., Reinert, K.H., Remington, K.A. et al. (2000). A whole-genome assembly of Drosophila. Science 287, 2196–2204. Myers, R.M., Larin, Z. and Maniatis, T. (1985). Detection of single base substitutions by ribonuclease cleavage at mismatches in RNA: DNA duplexes. Science 230, 1242–1246. Myers, R.M., Maniatis, T. and Lerman, L.S. (1987). Detection and localization of single base pair changes by denaturing gradient gel electrophoresis. Meth Enzymol 155, 501. Nataraj, A.J., Olivos-Glander, I., Kusukawa, N. and Highsmith, W.E.J. (1999). Single-strand conformational polymorphism and heteroduplex analysis for gel-based mutation detection. Electrophoresis 20, 1177–1185. Ng, P.C. and Henikoff , S. (2006). Predicting the effects of amino acid substitutions on protein function. Annu Rev Genomics Hum Genet 7, 61–80. Nomura, Y., Kamimura, J., Yoshinaga, M., Nishi, J., Kono, Y. and Miyata, K. (2001). Cleavase fragment length polymorphism analysis in a family with long QT syndrome. Pediatr Cardiol Card Surg 17, 20–25. Nyren, P., Pettersson, B. and Uhlen, M. (1993). Solid phase DNA minisequencing by an enzymatic luminometric inorganic pyrophosphate detection assay. Anal Biochem 208, 171–175. Osoegawa, K., Mammoser, A.G., Wu, C., Frengen, E., Zeng, C., Catanese, J.J. and de Jong, P.J. (2001). A bacterial artificial chromosome library for sequencing the complete human genome. Genome Res 11, 483–496. Patil, N., Berno, A.J., Hinds, D.A., Barrett, W.A., Doshi, J.M., Hacker, C.R., Kautzer, C.R., Lee, D.H., Marjoribanks, C., McDonough, D.P. et al. (2001). Blocks of limited haplotype diversity
■
99
revealed by high-resolution scanning of human chromosome 21. Science 294, 1719–1723. Prober, J.M., Trainor, G.L., Dam, R.J., Hobbs, F.W., Robertson, C.W., Zagursky, R.J., Cocuzza, A.J., Jensen, M.A. and Baumeister, K. (1987). A system for rapid DNA sequencing with fluorescent chain-terminating dideoxynucleotides. Science 238, 336–341. Rand, V., Huang, J., Stockwell, T., Ferriera, S., Buzko, O., Levy, S., Busam, D., Li, K., Edwards, J.B., Eberhart, C. et al. (2005). Sequence survey of receptor tyrosine kinases reveals mutations in glioblastomas. Proc Natl Acad Sci USA 102, 14344–14349. Redon, R., Ishikawa, S., Fitch, K.R., Feuk, L., Perry, G.H., Andrews, T.D., Fiegler, H., Shapero, M.H., Carson, A.R., Chen, W. et al. (2006). Global variation in copy number in the human genome. Nature 444, 444–454. Rothschild, C.B.,Akots, G., Hayworth, R., Pettenati, M.J.,Nagesh Rao, P., Wood, P., Stolz, F.-M., Hansmann, I., Serino, K.K.T.P., Fajans, S.S. et al. (1993). A genetic map of chromosome 20q12–13.1: Multiple highly polymorphic microsatellite and RFLP markers linked to the maturity-onset diabetes of the young (MODY) locus. Am J Hum Genet 52, 110–123. Rowntree, R.K. and Harris, A. (2003). The phenotypic consequences of CFTR mutations. Ann Hum Genet 67, 471–485. Rozen, S. and Skaletsky, H. (2000). Primer3 on the WWW for general users and for biologist programmers. Methods Mol Biol 132, 365–386. Saiki, R.K., Scharf, S., Faloona, F., Mullis, K.B., Horn, G.T., Erlich, H.A. and Arnheim, N. (1985). Enzymatic amplification of beta-globin genomic sequences and restriction site analysis for diagnosis of sickle cell anemia. Science 230, 1350–1354. Sambrook, J., Fritsch, E.F. and Maniatis, T. (1989). Molecular Cloning: A Laboratory Manual, 2nd ed. Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY. Sanger, F., Nicklen, S. and Coulson, A.R. (1977). DNA sequencing with chain-terminating inhibitors. Proc Natl Acad Sci USA 74, 5463–5467. Sanger, F., Coulson, A.R., Hong, G.F., Hill, D.F. and Petersen, G.B. (1982). Nucleotide sequence of bacteriophage lambda DNA. J Mol Biol 162, 729–773. Sebat, J., Lakshmi, B., Troge, J., Alexander, J., Young, J., Lundin, P., Maner, S., Massa, H., Walker, M., Chi, M. et al. (2004). Large-scale copy number polymorphism in the human genome. Science 305, 525–528. Sheffield,V.C., Beck, J.S., Kwitek, A.E., Sandstrom, D.W. and Stone, E.M. (1993). The sensitivity of single-strand conformation polymorphism analysis for the detection of single base substitutions. Genomics 16, 325–332. Shendure, J., Mitra, R.D.,Varma, C. and Church, G.M. (2004). Advanced sequencing technologies: Methods and goals. Nat Rev Genet 5, 335–344. Shendure, J., Porreca, G.J., Reppas, N.B., Lin, X., McCutcheon, J.P., Rosenbaum, A.M., Wang, M.D., Zhang, K., Mitra, R.D. and Church, G.M. (2005). Accurate multiplex polony sequencing of an evolved bacterial genome. Science 309, 1728–1732. Sladek, R., Rocheleau, G., Rung, J., Dina, C., Shen, L., Serre, D., Boutin, P., Vincent, D., Belisle, A., Hadjadj, S. et al. (2007). A genome-wide association study identifies novel risk loci for type 2 diabetes. Nature 445, 881–885. Smith, L.M., Sanders, J.Z., Kaiser, R.J., Hughes, P., Dodd, C., Connell, C.R., Heiner, C., Kent, S.B. and Hood, L.E. (1986). Fluorescence detection in automated DNA sequence analysis. Nature 321, 674–679.
100
CHAPTER 7
■
DNA Sequencing for the Detection of Human Genome Variation and Polymorphism
Stanssens, P., Zabeau, M., Meersseman, G., Remes, G., Gansemans, Y., Storm, N., Hartmer, R., Honisch, C., Rodi, C.P., Böcker, S. et al. (2004). High-throughput MALDI-TOF discovery of genomic sequence polymorphisms. Genome Res 14, 126–133. Stephens, M., Sloan, J.S., Robertson, P.D., Scheet, P. and Nickerson, D.A. (2006). Automating sequence-based detection and genotyping of SNPs from diploid samples. Nat Genet 38, 375–381. Strausberg, R.L., Simpson, A.J. and Wooster, R. (2003). Sequence-based cancer genomics: Progress, lessons and opportunities. Nat Rev Genet 4, 409–418. Tabone, T., Sallmann, G., Webb, E. and Cotton, R.G. (2006). Detection of 100% of mutations in 124 individuals using a standard UV/Vis microplate reader: A novel concept for mutation scanning. Nucleic Acids Res 34, e45. Thomas, G.A., Williams, D.L. and Soper, S.A. (2001). Capillary electrophoresis-based heteroduplex analysis with a universal heteroduplex generator for detection of point mutations associated with Rifampin resistance in tuberculosis. Clin Chem 47, 1195–1203. Thomas, R.K., Greulich, H., Yuza, Y., Lee, J.C., Tengs, T., Feng, W., Chen, T.H., Nickerson, E., Simons, J., Egholm, M. et al. (2005). Detection of oncogenic mutations in the EGFR gene in lung adenocarcinoma with differential sensitivity to EGFR tyrosine kinase inhibitors. Cold Spring Harb Symp Quant Biol 70, 73–81. Thomas, R.K., Nickerson, E., Simons, J.F., Janne, P.A., Tengs, T., Yuza, Y., Garraway, L.A., LaFramboise, T., Lee, J.C., Shah, K.
et al. (2006). Sensitive mutation detection in heterogeneous cancer specimens by massively parallel picoliter reactor sequencing. Nat Med 12, 852–855. Uehara, H., Kawano, H., Kataoka, H., Sameshima, T., Moriyama, T., Nakano, S. and Wakisaka, S. (1999). Mutational analysis of human p53 gene in human gliomas by Cleavase® fragment length polymorphism. Neuropathology 19, 267. Underhill, P.A., Jin, L., Lin, A.A., Mehdi, S.Q., Jenkins, T., Vollrath, D., Davis, R.W., Cavalli-Sforza, L.L. and Pefner, P.J. (1997). Detection of numerous Y chromosome biallelic polymorphisms by denaturing high-performance liquid chromatography. Genome Res 7, 996–1005. Venter, J.C., Smith, H.O. and Hood, L. (1996). A new strategy for genome sequencing. Nature 381, 364–366. Venter, J.C., Adams, M.D., Myers, E.W., Li, P.W., Mural, R.J., Sutton, G.G., Smith, H.O., Yandell, M., Evans, C.A., Holt, R.A. et al. (2001). The sequence of the human genome. Science 291, 1304–1351. Wheeler, D.A., Srinivasan, M., Egholm, M., Shen, Y., Chen, L., McGuire, A., He, W., Chen, Y. -J., Makhijani, V., Roth, G.T. et al. (2008). The complete genome of an individual by massively parallel DNA sequencing. Nature 452, 872–876.
CHAPTER
8 Genome-Wide Association Studies and Genotyping Technologies Kevin V. Shianna
INTRODUCTION Since the completion of the International HapMap Project (International HapMap Consortium, 2005; The International HapMap Consortium, 2003), a wealth of genetic information within the human genome has been uncovered. Over 12 million single nucleotide polymorphisms (SNPs) have been identified (Wheeler et al., 2007). This dense map of common genetic variation has revolutionized the field of human genetics. Genomewide linkage scans can now be performed using efficient and cost-effective high-throughput SNP genotyping technologies, while at the same time achieving a fine map at a density not previously possible when performing earlier-generation genome scans based on microsatellite polymorphisms. The SNP map, along with recent technological gains, has also made it possible to comprehensively study complex diseases and traits by allowing for the genotyping of hundreds of thousands of SNPs for many samples. Also, with the identification of a specific type of SNP, called tagging SNPs, almost complete genomic coverage of common variation can be attained. This chapter will provide an introduction to genome-wide association studies, followed by an overview of two commonly used genome-wide genotyping platforms, the Illumina Infinium Assay and the Affymetrix GeneChip. The methods and concepts discussed here complement those described elsewhere in this book.
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
PRINCIPLES OF GENOME-WIDE ASSOCIATION STUDIES Before performing a genome-wide association study, it is essential to begin with a fully developed plan.The key issues to consider are the kind of study to perform, the required sample size, the importance of a well-defined phenotype, the need to correct for population stratification and the technologies available for genome-wide genotyping. This section will provide an introduction to the basic experimental design for a genome-wide association study. Type of Study First, the study should be focused on a key question based on a specific phenotype or disease. As an example, a study could be established to identify genetic variants associated with epilepsy. The key would be to collect a large cohort of individuals that have a definitive and well-defined diagnosis of epilepsy (cases) and compare these individuals to a group that does not have epilepsy (controls). In the past it was always necessary to collect matched controls, but with an abundance of control samples containing data from the various genome-wide array-based studies, it is now possible to use control genotyping data from the available databases. One example is the database made available by Illumina (iControlDB) where individual users have deposited genotyping data for control samples from previously processed samples. It has been demonstrated that the use of general control
Copyright © 2009, Elsevier Inc. All rights reserved. 101
102
CHAPTER 8
TABLE 8.1
■
Genome-Wide Association Studies and Genotyping Technologies
Examples of successful genome-wide association studies
Disease
Chip type
Gene/locus
HIV
Illumina HumanHap 550
HLA-C, HCP5/HLA-B*5701
486
(Fellay et al., 2007)
Type 2 Diabetes
Illumina HumanHap 300
TCF7L2, SLC30A8, HHEX, FTO, PPARG and KCNJ11
1161/1174
(Scott et al., 2007)
Affymetrix Human Mapping 500
CDKN2A, CDKN2B, IGF2BP2, CDKAL1
1464/1467
(Saxena et al., 2007)
Illumina HumanHap 300
TCF7L2, AHI1-LOC441171 region
500/497
(Salonen et al., 2007)
Illumina HumanHap 300
TCF7L2, IDE–KIF11–HHEX and EXT2–ALX4
1363
(Sladek et al., 2007)
Illumina HumanHap 550
KIAA0350
563/1146
(Hakonarson et al., 2007)
Affymetrix Human Mapping 500
Chromosome regions 18q22, 12q24, 12q13, 16p13 and 18p11
4000/5000
( Todd et al., 2007)
Affymetrix Human Mapping 100
IFIH1
2029/1775
(Smyth et al., 2006)
Restless leg syndrome
Affymetrix Human Mapping 500
MEIS1, BTBD9
401/1644
( Winkelmann et al., 2007)
Colorectal cancer
Illumina HumanHap 550
8q24.21 locus
930/960
( Tomlinson et al., 2007)
Affymetrix Human Mapping 100
8q24 locus
1257/1336
(Zanke et al., 2007)
Illumina HumanHap 300
TCF2
1501/11290
(Gudmundsson et al., 2007)
Illumina HumanHap 300
8q24
1453/3064
(Gudmundsson et al., 2007)
Illumina HumanHap 300
8q24
1172/1157
( Yeager et al., 2007)
Illumina HumanHap 300
Chromosome region 4q25 near PITX2
550/4476
(Gudbjartsson et al., 2007)
Type 1 Diabetes
Prostate cancer
Atrial fibrillation
Sample size (cases/controls)
Reference
Celiac disease
Illumina HumanHap 300
IL2, IL21
778/1422
(van Heel et al., 2007)
Crohn’s disease
Illumina HumanHap 300
IL23R
2877/1345
( Tremelling et al., 2007)
Illumina HumanHap 300
IL23R
567/571
(Duerr et al., 2006)
Macular degeneration
Affymetrix Human Mapping 100
CFH
96/50
(Klein et al., 2005)
Asthma
Illumina HumanHap 300
ORMDL3
994/1243
(Moffatt et al., 2007)
samples across studies is feasible (Wellcome Trust Case Control Consortium, 2007).This greatly decreases the cost to run a project and potentially allows for the inclusion of more cases (if available). A study could also be arranged to look at a specific continuous phenotype or quantitative trait within a group of individuals. Using the previous example, one could study how individuals with epilepsy responded to a specific drug or series of drugs. Sample Size The sample sizes required for a genome-wide association study are dependent on several factors. The first is the effect size of an unknown genetic variable. Of course, this is impossible to know in advance; however, power calculations can be performed to determine how many samples would be required to identify a variant with a specific effect size. These calculations can serve as a guide when deciding how many samples to include in the study. Purcell and colleagues (2003) have created a web-based genetic power calculator (http://pngu.mgh.harvard.edu/~purcell/gpc/) that can be used to determine power of a study. Another factor
that is important when determining sample size is the type of study: case/control or a continuous phenotype. The current view for case/control studies is that large sample sizes are needed (1,000 of each), but this is not always true. A few studies have identified associated variants with only 150–200 samples (Table 8.1). Again, it is impossible to know the effect size of an unknown variant so it is best to include a large number of samples when designing a study. For a continuous phenotype, studies have had success while using fewer samples (Table 8.1). Importance of Phenotype The most important aspect when designing a genome-wide association study is the quality and discrete nature of the phenotype. It is essential that the studied phenotype is evaluated carefully for accuracy and is well defined across all samples. The inclusion of incorrectly phenotyped samples in a study could lead to a decrease in the statistical power to identify associated variants. This could consist of control samples that are in fact cases, but have yet to be identified with the specific phenotype/diagnosis or a misdiagnosis
Platform Overview
of an individual within the cases. For some complex diseases, the phenotypic characteristics overlap with other closely related diseases. This can make it difficult when choosing which samples should be included in the study and, because of genetic heterogeneity, will decrease the power of the study. Population Stratification When analyzing the data from genome-wide association studies, it is essential to evaluate and account for population stratification between cases and controls. Cryptic population structure can generate false positive and/or negative associations. This can happen if allele frequencies are different in any unknown subgroups. One solution to correct for this difference in ancestry is described by Price and colleagues (2006). Left uncorrected, population stratification can lead directly to false associations.
Address
■
103
Probe
Bead
Figure 8.1 An example of a bead on the Illumina Infinium BeadChip. A schematic of a single bead, containing a 75nucleotide oligomer consisting of an address and a specific probe sequence, as described in the text.
Genomic DNA A T Extended nucleotides emit signal
Bead 1
Genomic DNA
Genotyping Technologies, Tagging SNPs and Genomic Coverage Another necessity for genome-wide association studies is to have a method to genotype hundreds of thousands of genetic variants around the genome. Fortunately, there are now standard products that allow for screening of very large numbers of SNPs. Depending on the chosen platform, the SNPs on the arrays have been selected based either on even spacing or on the basis of certain SNPs being able to represent multiple SNPs.The SNPs in the second approach are termed “tagging SNPs” and greatly increase the genomic coverage that can be obtained compared to SNPs that are evenly spaced across the genome (Barrett and Cardon, 2006; Pe’er et al., 2006). Reflecting now the well-established patterns of linkage disequilibrium (Hinds et al., 2005; International HapMap Consortium, 2005), these tagging SNPs work by acting as proxies for other nearby associated SNPs, making it possible to genotype only the tagging SNP and yet capture the content of other associated SNPs in the region. There are currently commercially available arrays (SNP chips) that contain close to 1 million tagging SNPs, representing most of the common variation in the human genome. Products where the SNPs have not been specifically chosen based on the close association with other SNPs will have much less genomic coverage, because the SNPs have not been optimally chosen for the ability to tag. For example, Illumina’s HumanHap300 array, containing around 300,000 tagging SNPs, has essentially the same statistical genomic coverage as Affymetrix’s Mapping 500 K Array Set, which contains 500,000 evenly spaced SNPs (Barrett and Cardon, 2006; Pe’er et al., 2006). Further background information on genome-wide association studies and tagging SNPs has been elegantly presented elsewhere (Hirschhorn and Daly, 2005).
PLATFORM OVERVIEW Illumina Infinium Assay The Illumina genotyping technology is based on 3 silicon beads that have hundreds of thousands of DNA oligomers attached to
A
(a)
G
Bead 2
Genomic DNA A T*
(b)
Bead
G
Figure 8.2 Infinium I and II assay. (a) A graphical representation of the Infinium I assay. This assay uses an ASPE and requires two bead types to detect a single SNP. (b) A depiction of an SBE reaction as used for the Infinium II assay. Only one bead type is required to detect a single SNP. The diagram shows a single dideoxy base (T) being added to the 50-base oligomer on the bead.
the surface (Shen et al., 2005). For Illumina’s Infinium wholegenome genotyping BeadChip, the attached oligomers consist of a stretch of 75 nucleotides (Gunderson et al., 2005; Steemers et al., 2006) (Figure 8.1). The first 25 nucleotides (the address) serve as a barcode to identify the location of specific beads on the array, and the next 50 bases (the probe) are complementary to a genomic region containing a specific SNP (or copy number variant). On average, there is an 18-fold redundancy of each bead type on the BeadChip. The placement of the beads on the BeadChip is a random process, thus requiring a decoding step. The decoding involves multiple hybridizations to the 25 oligomer address. This not only identifies the location of each bead, but performs a functional quality control for each bead on the BeadChip. There are two Infinium chemistries, type I and II (Figure 8.2). The Infinium I assay uses an allele-specific primer extension (ASPE) reaction in which fluorescently labeled nucleotides are inserted at the 3 end of the primer. With this assay it takes two bead types to determine the genotype at a single SNP. The Infinium II assay involves a single-base extension (SBE) reaction that occurs on a single bead type for each SNP. Most of Illumina’s off-the-shelf BeadChips use the Infinium II assay.
104
CHAPTER 8
■
Genome-Wide Association Studies and Genotyping Technologies
rs2381194
Restriction site
Restriction site
6000
Intensity (B)
Enzyme digestion 4000
2000
Adapter ligation 0
(a)
0
2000
4000 6000 Intensity (A)
8000
10000
PCR
rs2381194 1.20 1
Fragment and label
Norm R
0.80 0.60 0.40 0.20 0 0.20 (b)
26 0
227 0.20
0.40 0.60 Norm Theta
696 0.80
1
Figure 8.3 Raw and normalized data for SNP genotyping, using the Illumina platform.(a) Raw data from genotype analysis of one SNP (rs2381194) from 949 individuals. Each dot corresponds to a single individual. Blue, homozygote for one allele (n 696); Purple, heterozygous (n 227); Red, homozygote for other allele (n 26). (b) Same SNP and individuals as in (a), but after normalization using Illumina’s BeadStudio software.
The steps for the Infinium II assay are as follows: (1) 750 ng of intact genomic DNA is whole-genome amplified (WGA); (2) the amplified product is fragmented; (3) the fragmented DNA is hybridized to a BeadChip; (4) an SBE reaction is performed directly on the chip with the insertion of a labeled (2,4-dinitrophenol or biotin) dideoxy base; (5) the inserted base is stained with a fluorescently labeled antibody and (6) fluorescence on the beads on the BeadChip is detected by scanning with the Illumina BeadStation. An average of the raw intensity per bead type is then imported into the Illumina BeadStudio software, where the data can be efficiently managed and individual SNPs can be automatically or manually called. The quality of the raw and normalized genotype calls can be visually confirmed within this software (Figure 8.3).
Figure 8.4 Overview of DNA processing steps for the AffymetrixGeneChip. The diagram demonstrates the steps implemented to reduce the genomic complexity allowing for efficient allele-specific hybridization.
Affymetrix GeneChip The Affymetrix genotyping GeneChip is manufactured using the same principles as for the expression GeneChip, using a technology called photolithography. This has been described in great detail (Dalma-Weiszhausz et al., 2006; Matsuzaki et al., 2004), including elsewhere in this book. However, the genotyping GeneChip contains additional probes for each interrogated SNP, to allow for differentiation of true signal over noise (DalmaWeiszhausz et al., 2006). The steps required to process a DNA sample in preparation for hybridization to the GeneChip lead to a reduction in genome complexity allowing for allele-specific hybridization (Figure 8.4). The following is the procedure for processing a sample: (1) perform separate genomic DNA (250 ng) digestions using the NspI and StyI restriction enzymes; (2) ligate specific adapters containing a single universal primer binding site to the digested DNA; (3) pool the DNA fragments from the separate digestions and amplify using a single universal PCR primer specific to the primer site within the adaptor; (4) fragment the amplified products; (5) end label the DNA; (6) hybridize to a GeneChip and (7) scan the chip. The raw data can then be converted into genotype calls by using the appropriate algorithms (from Affymetrix).
Platform Overview
Platform Comparison Numerous laboratories worldwide have successfully used both the Illumina and Affymetrix platforms to identify genetic variants involved with complex disease (Table 8.1). This section will describe potential benefits of each platform, paying particular attention to (1) differences in the basic design of the chips; (2) statistical tagging power (genomic coverage); (3) ability to detect copy number variation (CNV); (4) processing issues and (5) overall quality of the final raw genotyping data. The length of the probes to detect an SNP during hybridization constitutes one of the major differences between the platforms. The probe length on the Affymetrix GeneChip is 25 nucleotides, while on the Illumina BeadChip the probes contain 50 nucleotides. This difference in length for hybridization provides for higher specificity for probes on the Illumina BeadChip. The other main difference is the content of the probe for each SNP. For Affymetrix, each SNP is represented by a series of probes consisting of mismatches and perfect homology, allowing for differentiation between signal and noise. The Illumina BeadChips do not require specific mismatch probes for every SNP because of the increased length of the probes for hybridization, but contain control beads that detect the overall level of background noise. The probe length constitutes the major difference between the two platforms and is most likely directly linked to the differences in quality between the platforms. Most of the SNPs on the Illumina Infinium BeadChips were selected as tagging SNPs, resulting in excellent genomic coverage for most populations (based on HapMap populations) (International HapMap Consortium, 2005). On the other hand, the Affymetrix GeneChips are based on SNPs that are evenly spaced throughout the genome. The positive aspect of even spacing is better physical coverage for detecting CNVs (Redon et al., 2006). One disadvantage of this approach is less statistical power to capture the known common variation when compared to arrays that contain tagging SNPs. A multimarker or haplotype-tagging approach may gain increased power/coverage, but the downstream statistical analyses become more difficult. Multiple groups (Barrett and Cardon, 2006; Magi et al., 2007; Pe’er et al., 2006) have performed statistical comparisons of the Illumina HumanHap300 and the HumanHap550 BeadChips with the Affymetrix Mapping 500 K Set. From these analyses it is evident that the Illumina array with only around 300,000 tagging SNPs has essentially the same statistical genomic coverage as the Affymetrix Mapping 500 K Set, while the Illumina HumanHap550 array is statistically superior for genome-wide association studies. Overall, both platforms provide good quality results (Table 8.1). However, a comparison of the SNP conversion rate (percentage of SNPs that work) between the two technologies reveals a significant difference. In our hands, for example, when using the Illumina pre-defined cluster, an average sample call rate of 99.6% is achieved, with an average reproducibility of over 99.99%. With the Affymetrix platform, the average sample call rate is variable and depends greatly on the chosen stringency within the calling algorithm, but a typical call rate would be in the range of 95–98% (personal communications). As discussed above, the increased SNP conversion rate with the Illumina BeadChip is most likely related
■
105
to the size of the probe (50 nucleotides) which increases the specificity during the hybridization step. Recently, there has been a push to detect CNVs throughout the human genome by using genome-wide genotyping chips (Komura et al., 2006; Redon et al., 2006) (see Chapter 9). Using these SNP-based arrays is an efficient method to detect large deletions or duplications. The Affymetrix GeneChip with its even spacing of SNPs has great genomic coverage to detect large CNVs. Most of the BeadChips from Illumina have specific SNPs or non-polymorphic probes added in regions where there are known CNVs. In general, both arrays are capable of detecting large CNVs. However, with the current SNP densities, the detection of small CNVs is difficult with the use of SNP-based arrays. Raw Data Quality Control Genetic association studies using genome-wide genotyping chips produce very large amounts of data. A typical project with 2,000 samples, each containing over 500,000 SNPs per sample, will result in over one billion data points. With this much data, the probability for false positives becomes much greater. A few reasons that can attribute to this are samples being prepared using different purification methods and the processing of the chips in different labs. These slight variations can lead to minor differences in intensity, thus making it more difficult to accurately call the genotypes. To prevent false positives from clouding the results from the statistical analyses, it is of utmost importance to begin with a clean or curated data set. This is accomplished by establishing a high threshold of sample and SNP quality. The number of SNPs that are obtained for each sample (the “call rate”) and the raw fluorescent intensity are the main variables that can be analyzed to determine whether a sample should be included or excluded. If any sample’s call rate is much lower (1–2%) than the average call rate for the project, the sample should be excluded. A 1% difference can represent as many as 5,000 fewer SNPs being called in a sample when compared with the average number of SNP calls in a population of samples. There are many variables that can be used to assess SNP quality. Metrics such as call frequency (number of samples that are being called for an SNP), cluster separation (degree of split between the heterozygote grouping and homozygotes, e.g. Figure 8.3), extreme deviation from Hardy–Weinberg proportions and overall raw signal intensity should be evaluated for all SNPs. A very aggressive approach to maintaining a clean data set is to adhere to a rigorous level of acceptable SNP call frequency. We have established a high threshold for SNP inclusion and have coined the phrase the “1% rule”; we delete any SNP that has greater than 1% of the samples not being called or called ambiguously. This rule was established from personal experiences when attempting to extract as much genotype data as possible. After performing statistical analyses, we noticed a large number of false positives due entirely to poor genotyping calls. For the 1% rule to work effectively, each new experiment needs to be reclustered using its own raw data. With the 1% rule, the use of predetermined cluster files will not work, because slight variations from experiment to experiment can result in slight shifts of intensity and cluster separation (i.e., separation of heterozygote
106
CHAPTER 8
■
Genome-Wide Association Studies and Genotyping Technologies
and homozygote clusters), resulting in incomplete calling of truly callable samples for a large number of SNPs. When applying the 1% rule, it is a balance between including a greater number of samples or a greater number of SNPs. Attempting to keep good but lower quality samples can result in the deletion of many SNPs that fail to meet the 1% rule standard. By deleting a few extra samples across the entire data set, it is possible to keep many of the SNPs that would have violated the 1% rule. In practice, in our experience, a typical project will have between 1% and 3% samples deleted (not all because of 1% rule) across all SNPs and between 1% and 2% of SNPs deleted because of violating the 1% rule. We have successfully used this approach for multiple studies (Fellay et al., 2007). After completing the first pass of statistical tests, the SNPs with the lowest 100–200 p-values should be manually checked for genotyping data quality. Visualization of the raw/normalized genotype call can increase one’s confidence in any potentially associated SNP with a low p-value. Inspecting the data is especially important if a data set has not been stringently curated under the 1% rule. For a given SNP, if only 95% of samples are called, over 50 samples from a sample set of 1,000 samples will not have a genotype call. Across a very large data set (over 500,000 SNPs), this loosened threshold can result in many false positives solely because of the increased opportunity for an imbalance of the case/control ratio in the uncalled samples. Sample Collection, Processing and Throughput It is now feasible for institutional core facilities or small- to medium-sized labs to run hundreds of samples per week using genome-wide genotyping chips. For each platform described here, there are off-the-shelf kits allowing for easy processing of many chips/samples at once with minimal variation across experiments. Also, with the implementation of automation developed specifically for each assay, sample throughput and tracking have all been simplified.
Genome-wide genotyping chips require less than 1 g of DNA, a very low amount considering the amount of data generated from this sample. In general, high-quality genomic DNA should be used. DNA of lesser quality or WGA DNA can be used, but will result in a lower sample call rate. The Affymetrix platform is more amenable to the use of WGA DNA, because of the lack of an initial WGA step as in the Illumina Infinium assay. However, WGA DNA will work for the Infinium platform, although lower call rates will be achieved (personal communication). Formalin-fixed, paraffin-embedded (FFPE) tissue samples represent a rich and abundant source of well-annotated material, but DNA isolated from FFPE samples using currently available methods is not amenable to genome-wide genotyping assays. The main reason for this is that the interaction of the formalin with the DNA prevents efficient enzymatic amplification of the DNA. Creating methods to successfully perform whole-genome genotyping on FFPE samples will be of great importance. This will benefit current projects by allowing for the potential increase in sample sizes. Also, precious samples obtained only via FFPE tissues could then be included in genome-wide association studies.
CONCLUSIONS This chapter has provided a basic overview of genome-wide association studies and the current technologies used for genome-wide genotyping. Since the completion of the International HapMap Project in October 2005 (International HapMap Consortium, 2005), there have been great technological advances that now allow for the genotyping of over one million SNPs within many individual samples in a matter of days. The ability to achieve genomewide SNP coverage has lead to the identification of multiple variants associated with disease (Table 8.1). These sorts of studies will continue in the coming years until it becomes feasible to simply and cost-effectively sequence the entire genome of individuals.
REFERENCES Barrett, J.C. and Cardon, L.R. (2006). Evaluating coverage of genomewide association studies. Nat Genet 38, 659–662. Dalma-Weiszhausz, D.D.,Warrington, J.,Tanimoto, E.Y. and Miyada, C.G. (2006). The affymetrix GeneChip platform: an overview. Methods Enzymol 410, 3–28. Duerr, R.H., Taylor, K.D., Brant, S.R., Rioux, J.D., Silverberg, M.S., Daly, M.J., Steinhart, A.H., Abraham, C., Regueiro, M., Griffiths, A., Dassopoulos, T., Bitton, A., Yang, H., Targan, S., Datta, L.W., Kistner, E.O., Schumm, L.P., Lee, A.T., Gregersen, P.K., Barmada, M.M., Rotter, J.I., Nicolae, D.L. and Cho, J.H. (2006). A genomewide association study identifies IL23R as an inflammatory bowel disease gene. Science 314, 1461–1463. Fellay, J., Shianna, K.V., Ge, D., Colombo, S., Ledergerber, B., Weale, M., Zhang, K., Gumbs, C., Castagna, A., Cossarizza, A. et al. (2007). A Whole-Genome Association Study of Major Determinants for Host Control of HIV-1. Science 317, 944–947. Gudbjartsson, D.F., Arnar, D.O., Helgadottir, A., Gretarsdottir, S., Holm, H., Sigurdsson, A., Jonasdottir, A., Baker, A., Thorleifsson, G., Kristjansson, K. et al. (2007). Variants conferring risk of atrial fibrillation on chromosome 4q25. Nature 448, 353–357.
Gudmundsson, J., Sulem, P., Manolescu, A., Amundadottir, L.T., Gudbjartsson, D., Helgason, A., Rafnar, T., Bergthorsson, J.T., Agnarsson, B.A., Baker, A. et al. (2007). Genome-wide association study identifies a second prostate cancer susceptibility variant at 8q24. Nat Genet 39, 631–637. Gudmundsson, J., Sulem, P., Steinthorsdottir, V., Bergthorsson, J.T., Thorleifsson, G., Manolescu, A., Rafnar, T., Gudbjartsson, D., Agnarsson, B.A., Baker, A. et al. (2007). Two variants on chromosome 17 confer prostate cancer risk, and the one in TCF2 protects against type 2 diabetes. Nat Genet 39, 977–983. Gunderson, K.L., Steemers, F.J., Lee, G., Mendoza, L.G. and Chee, M.S. (2005). A genome-wide scalable SNP genotyping assay using microarray technology. Nat Genet 37, 549–554. Hakonarson, H., Grant, S.F., Bradfield, J.P., Marchand, L., Kim, C.E., Glessner, J.T., Grabs, R., Casalunovo, T., Taback, S.P., Frackelton, E.C. et al. (2007). A genome-wide association study identifies KIAA0350 as a type 1 diabetes gene. Nature 448, 591–594. Hinds, D.A., Stuve, L.L., Nilsen, G.B., Halperin, E., Eskin, E., Ballinger, D.G., Frazer, K.A. and Cox, D.R. (2005). Whole-genome
References
patterns of common DNA variation in three human populations. Science 307, 1072–1079. Hirschhorn, J.N. and Daly, M.J. (2005). Genome-wide association studies for common diseases and complex traits. Nat Rev Genet 6, 95–108. International HapMap Consortium (2003). The International HapMap Project. Nature 426, 789–796. International HapMap Consortium (2005). A haplotype map of the human genome. Nature 437, 1299–1320. Klein, R.J., Zeiss, C., Chew, E.Y., Tsai, J.Y., Sackler, R.S., Haynes, C., Henning, A.K., SanGiovanni, J.P., Mane, S.M., Mayne, S.T. et al. (2005). Complement factor H polymorphism in age-related macular degeneration. Science 308, 385–389. Komura, D., Shen, F., Ishikawa, S., Fitch, K.R., Chen, W., Zhang, J., Liu, G., Ihara, S., Nakamura, H., Hurles, M.E. et al. (2006). Genomewide detection of human copy number variations using high-density DNA oligonucleotide arrays. Genome Res 16, 1575–1584. Magi, R., Pfeufer, A., Nelis, M., Montpetit, A., Metspalu, A. and Remm, M. (2007). Evaluating the performance of commercial whole-genome marker sets for capturing common genetic variation. BMC Genomics 8, 159. Matsuzaki, H., Loi, H., Dong, S., Tsai, Y.Y., Fang, J., Law, J., Di, X. et al. (2004). Parallel genotyping of over 10,000 SNPs using a oneprimer assay on a high-density oligonucleotide array. Genome Res 14, 414–425. Moffatt, M.F., Kabesch, M., Liang, L., Dixon, A.L., Strachan, D., Heath, S., Depner, M., von Berg, A., Bufe, A., Rietschel, E. et al. (2007). Genetic variants regulating ORMDL3 expression contribute to the risk of childhood asthma. Nature 448, 470–473. Pe’er, I., de Bakker, P.I., Maller, J., Yelensky, R., Altshuler, D. and Daly, M.J. (2006). Evaluating and improving power in whole-genome association studies using fixed marker sets. Nat Genet 38, 663–667. Price, A.L., Patterson, N.J., Plenge, R.M., Weinblatt, M.E., Shadick, N.A. and Reich, D. (2006). Principal components analysis corrects for stratification in genome-wide association studies. Nat Genet 38, 904–909. Purcell, S., Cherny, S.S. and Sham, P.C. (2003). Genetic Power Calculator: design of linkage and association genetic mapping studies of complex traits. Bioinformatics 19, 149–150. Redon, R., Ishikawa, S., Fitch, K.R., Feuk, L., Perry, G.H., Andrews, T.D., Fiegler, H. et al. (2006). Global variation in copy number in the human genome. Nature 444, 444–454. Salonen, J.T., Uimari, P., Aalto, J.M., Pirskanen, M., Kaikkonen, J., Todorova, B., Hypponen, J., Korhonen, V.P., Asikainen, J., Devine, C. et al. (2007). Type 2 Diabetes Whole-Genome Association Study in Four Populations: The DiaGen Consortium. Am J Hum Genet 81, 338–345. Saxena, R., Voight, B.F., Lyssenko, V., Burtt, N.P., de Bakker, P.I., Chen, H., Roix, J.J., Kathiresan, S., Hirschhorn, J.N., Daly, M.J. et al. (2007). Genome-wide association analysis identifies loci for type 2 diabetes and triglyceride levels. Science 316, 1331–1336. Scott, L.J., Mohlke, K.L., Bonnycastle, L.L., Willer, C.J., Li, Y., Duren, W.L., Erdos, M.R., Stringham, H.M., Chines, P.S., Jackson, A.U. et al. (2007). A genome-wide association study of type 2 diabetes in Finns detects multiple susceptibility variants. Science 316, 1341–1345.
■
107
Shen, R., Fan, J.B., Campbell, D., Chang, W., Chen, J., Doucet, D., Yeakley, J. et al. (2005). High-throughput SNP genotyping on universal bead arrays. Mutat Res 573, 70–82. Sladek, R., Rocheleau, G., Rung, J., Dina, C., Shen, L., Serre, D., Boutin, P., Vincent, D., Belisle, A., Hadjadj, S. et al. (2007). A genome-wide association study identifies novel risk loci for type 2 diabetes. Nature 445, 881–885. Smyth, D.J., Cooper, J.D., Bailey, R., Field, S., Burren, O., Smink, L.J., Guja, C., Ionescu-Tirgoviste, C., Widmer, B., Dunger, D.B. et al. (2006). A genome-wide association study of nonsynonymous SNPs identifies a type 1 diabetes locus in the interferon-induced helicase (IFIH1) region. Nat Genet 38, 617–619. Steemers, F.J., Chang, W., Lee, G., Barker, D.L., Shen, R. and Gunderson, K.L. (2006). Whole-genome genotyping with the single-base extension assay. Nat Methods 3, 31–33. Todd, J.A., Walker, N.M., Cooper, J.D., Smyth, D.J., Downes, K., Plagnol, V., Bailey, R., Nejentsev, S., Field, S.F., Payne, F. et al. (2007). Robust associations of four new chromosome regions from genome-wide analyses of type 1 diabetes. Nat Genet 39, 857–864. Tomlinson, I., Webb, E., Carvajal-Carmona, L., Broderick, P., Kemp, Z., Spain, S., Penegar, S., Chandler, I., Gorman, M., Wood, W. et al. (2007). A genome-wide association scan of tag SNPs identifies a susceptibility variant for colorectal cancer at 8q24.21. Nat Genet 39, 984–988. Tremelling, M., Cummings, F., Fisher, S.A., Mansfield, J., Gwilliam, R., Keniry, A., Nimmo, E.R., Drummond, H., Onnie, C.M., Prescott, N.J. et al. (2007). IL23R variation determines susceptibility but not disease phenotype in inflammatory bowel disease. Gastroenterology 132, 1657–1664. van Heel, D.A., Franke, L., Hunt, K.A., Gwilliam, R., Zhernakova, A., Inouye, M., Wapenaar, M.C., Barnardo, M.C., Bethel, G., Holmes, G.K. et al. (2007). A genome-wide association study for celiac disease identifies risk variants in the region harboring IL2 and IL21. Nat Genet 39, 827–829. Wellcome Trust Case Control Consortium (2007). Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 447, 661–678. Wheeler, D.L., Barrett, T., Benson, D.A., Bryant, S.H., Canese, K., Chetvernin, V., Church, D.M. et al. (2007). Database resources of the National Center for Biotechnology Information. Nucleic Acids Res 35, D5–D12. Winkelmann, J., Schormair, B., Lichtner, P., Ripke, S., Xiong, L., Jalilzadeh, S., Fulda, S., Putz, B., Eckstein, G., Hauk, S. et al. (2007). Genome-wide association study of restless legs syndrome identifies common variants in three genomic regions. Nat Genet 39, 1000–1006. Yeager, M., Orr, N., Hayes, R.B., Jacobs, K.B., Kraft, P., Wacholder, S., Minichiello, M.J., Fearnhead, P.,Yu, K., Chatterjee, N. et al. (2007). Genome-wide association study of prostate cancer identifies a second risk locus at 8q24. Nat Genet 39, 645–649. Zanke, B.W., Greenwood, C.M., Rangrej, J., Kustra, R., Tenesa, A., Farrington, S.M., Prendergast, J., Olschwang, S., Chiang, T., Crowdy, E. et al. (2007). Genome-wide association scan identifies a colorectal cancer susceptibility locus on chromosome 8q24. Nat Genet 39, 989–994.
CHAPTER
9 Copy Number Variation and Human Health Charles Lee, Courtney Hyland, Arthur S. Lee, Shona Hislop and Chunhwa Ihm
INTRODUCTION Early cytogenetic studies recognized that microscopically visible aberrations such as duplications or deletions of entire chromosomes (aneuploidy) were associated with specific congenital developmental disorders. For example, an extra copy of chromosome 21 (i.e., trisomy 21) has long been established as being correlative with the mongoloid phenotype that was first recognized in 1866 by John Langdon Down (Down 1866; Lejeune et al., 1959). However, recent development of high-resolution assays, capable of detecting small segmental genetic alterations in a genome-wide fashion, have led to the detection of widespread copy number variation (CNV) among the genomes of healthy individuals (Iafrate et al., 2004; Sebat et al., 2004). This finding has now sparked intense efforts to identify and characterize the extent of this type of genetic variation in human populations and to understand its impact on human health.
BASIC PRINCIPLES OF CNVS CNVs have been operationally defined as genomic gains and losses of 1 kb or larger (Freeman et al., 2006; Feuk et al., 2006). This definition helps to differentiate it from other forms of
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 108
polymorphism and/or repeated DNAs in the human genome (see Chapter 1), which include indels, microsatellites, minisatellites, simple repeats (e.g., dinucleotide repeats, trinucleotide repeats, etc.), telomeric and centromeric repetitive DNAs, and most interspersed repetitive elements (although LINES and other long interspersed repeats have repeat elements greater than 1 kb in size). In addition, since CNVs are considered subchromosomal imbalances, they are differentiated from whole chromosomal aneuploidies, such as trisomy 21 or monosomy X. When CNV data are available for many individuals within or across different human populations, CNVs can be categorized into biallelic or multiallelic states (Figure 9.1). Biallelic CNVs have only two alleles and thus produce three different genotypes. For example, a CNV that exists solely in the form of a 1-copy allele or a 2-copy allele (Figure 9.1a, right) can produce diploid copy numbers of 2 (1 allele1 allele), 3 (1 allele2 allele), or 4 (2 allele2 allele) (Figure 9.1a, left). CNVs with greater than two alleles are considered multiallelic and result in more than three different genotypes (Figure 9.1b and Figure 9.2). Heritable CNVs are thought to arise from germline genomic rearrangements (or in some cases, possibly very early somatic events).The genomic rearrangements (mutational events) that are thought to cause CNVs can be broadly categorized as arising via one of two mechanisms: (i) Non-allelic homologous recombination (NAHR) – where homologous recombination
Copyright © 2009, Elsevier Inc. All rights reserved.
Basic Principles of CNVs
■
109
Alleles
Reference = 2 copies (40% frequency)
Duplication CNV = 3 copies (50% frequency)
Frequency (%)
100 Duplication CNV = 4 copies (10% frequency)
80 60 40 20 0
1
2 3 4 Copy Number
5
6 Hypothetical Gene
(a) Bi-allelic CNV Alleles
Deletion CNV = 3 copies (10% frequency)
Duplication CNV = 5 copies (20% frequency)
Frequency (%)
100 80 60 Reference = 4 copies (60% frequency) 40 20 0
1
3 4 2 Copy Number
5
6
SULT1A1 Gene SULT1A1 Gene Deletion
(b) Multi-allelic CNV
Figure 9.1 Examples of bi-allelic and multi-allelic CNVs. a) An example of a bi-allelic CNV that has a 1-copy allele and a 2-copy allele. The reference individual has two 1-copy alleles but 50% of individuals in this population have a total of three copies of this gene per cell. All bi-allelic CNVs have three genotypes per diploid cell, and in this case, copy numbers of 2, 3, and 4 per diploid cell. b) An example of a multi-allelic CNV that has a 0, 1, 2 and 3 copy alleles, resulting in six genotypes in this population. Only the allelic combinations for the three most common genotypes are shown.
Figure 9.2 The amylase gene is considered a multiallelic CNV with diploid copy number estimates ranging from 2 to 15 among humans (Perry et al., 2007). Here, a two-color fiber-FISH experiment of the amylase gene (each copy of the gene is represented by a redgreen combination) shows three copies on a DNA fiber from a single chromosome. It is interesting to note here that the second amylase gene copy is inverted relative to the others.
occurs between highly identical sequences in the genome, such as segmental duplications or related interspersed repetitive elements (Figure 9.3) and (ii) non-homologous end joining (NHEJ) – a repair mechanism whereby double-strand breaks
that occur in the genome are ligated together with the assistance of specific protein complexes that form at the sites of the double-strand breaks (Figure 9.4). The rate for NHEJ is likely influenced significantly by environmental factors and localized DNA
110
CHAPTER 9
■
Copy Number Variation and Human Health
(gene a and b deleted)
(gene a and b duplicated)
Segmental duplication – allele 1 Segmental duplication – allele 2 Gene a Gene b
Figure 9.3 Non-allelic homologous recombination (NAHR) is a mechanism for generating CNVs where recombination between non-allelic repeats with 90% sequence homology (indicated by black- and gray-colored DNA segments). Intervening DNA sequences are deleted and duplicated on different chromatids.
conformations, but in general has been estimated to occur at a rate of less than 107 per generation, similar to the 108 per generation estimated mutation rate of single nucleotide polymorphisms (SNPs) (Conrad and Hurles, 2007). NAHR events are believed to occur more frequently, with estimates of up to 104 per locus per gamete per generation (Shaffer and Lupski 2000). Since NAHR events lead to duplication and deletion of DNA sequences that lie between highly identical sequences in the genome, this mutational event tends to be associated with larger CNVs (Redon et al., 2006). CNVs are scattered throughout the human genome (Figure 9.5) with at least 6% of a chromosome’s total DNA content being potentially copy number variable (Redon et al., 2006). Taken together, current estimates are that more than 500 Mb (or 18.8%) of the reference human genome is copy number variable (Scherer et al., 2007). In fact, one recent study suggested that structural
genomic variation (i.e., non-SNP variation that is predominantly in the form of CNVs) account for as much as 22% of all genetic variable events found in a given individual and total 74% of all variant bases (Levy et al., 2007). Many of the earlier CNV discovery projects relied on array comparative genomic hybridization (aCGH) technologies that utilized large-insert clones (on the order of 120kb–150kb each) as individuals probles and therefore yielded CNV data with ill-defined boundaries. This has likely led to a size over-estimation for many currently identified CNVs. Documented CNVs are being cataloged and collated in several databases. For example, CNV data from healthy individuals can be found in the Database of Genomic Variants (http:// projects.tcag.ca/variation) and the human paralogy database (http://humanparalogy.gs.washington.edu/structuralvariation), while CNVs from patients affected with a neurodevelopmental disorder are being collated in databases such as DECIPHER
Basic Principles of CNVs
b
c
b
■
111
c
(a)
(b)
Acentric fragment Gene a
Gene c
Gene b
Gene d
Double-strand DNA break
Figure 9.4 Non-homologous end joining (NHEJ) can result in a genomic rearrangement that immediately or eventually results in the gain or loss of genetic material after multiple double-strand DNA breaks occur and are repaired. (a) A balanced rearrangement has occurred and there is no immediate net gain or loss of genomic material within the cell, but it will result in gains or losses of DNA in subsequent generations depending on which chromosome is inherited. (b) In an unbalanced rearrangement, genomic material can be lost immediately (e.g., in this case, the acentric fragment containing one copy of genes b and c will be lost in subsequent cell divisions).
112
CHAPTER 9
■
Copy Number Variation and Human Health
1
2 3
4
5
6
7
8
9
10
11 12 Length scaled log(length)
Presence of CNV freq
X 13 14
15 16 17 18 19 20 21 22
Y Length log (heterozygosity)
Figure 9.5 Genomic distribution of CNVs identified in one study (Redon et al., 2006). No large stretches of the genome appear to be exempt from CNVs and the proportion of any given chromosome that is susceptible to CNVs conservatively varies from 6% to 19%.
(http://www.sanger.ac.uk/PostGenomic/decipher/), the Chromosome Abnormality Database (http://www.ukcad. org.uk/cocoon/ukcad/) and the European Cytogenetics Association Register of Unbalanced Chromosome Aberrations (http://www.ecaruca.net). Most CNVs found in healthy individuals are biased away from genes and reside within intergenic regions (i.e., DNA sequences between genes) (Conrad et al., 2006; Nguyen et al., 2006). Nevertheless, the remaining CNVs have been shown to overlap some 3,000 RefSeq genes and 300 genes implicated in genetic disease (as listed in the Online Mendelian Inheritance in Man database, http://www.ncbi.nlm.nih.gov/sites/ entrez?dbOMIM). Ontology analyses of the genes thought to be copy number variable demonstrate a substantial number that can be classified as “environmental sensor/interaction” genes (Tuzun et al., 2005). These are genes that are involved in sensory perception, neurophysiological processes, drug detoxification, immunity and inflammation, as well as cell surface integrity and cell surface antigens. Clearly, such genes are not critical for early development but rather are involved in our perception and interaction with external stimuli, helping us to adapt to our ever-changing environment. The functional impact of certain CNVs can be relatively straightforward. For example, reduced copy number of a given gene can often be correlated with reduced expression levels, while additional copies of a gene could lead to increased expression levels of the CNV gene (McCarroll et al., 2006). CNVs that involve parts of a gene could result in fusion gene products or aberrant proteins with addition or loss of specific protein domains. Some CNVs in intergenic regions could overlap regulatory elements that affect the expression of genes as far away as 4 Mb (Stranger et al., 2007), and correlations of CNVs with transcriptional levels do not necessarily have to be positive. For example, deletion of a repressor element may cause upregulation of an associated gene.
DETECTING CNVS IN A GENOME-WIDE MANNER Array-Based Comparative Genomic Hybridization There are different genome-wide methods for detecting CNVs. By far, the most widely used method has been array-based comparative genomic hybridization (aCGH). This technology was first introduced as “matrix-CGH” (Solinas-Toldo et al., 1997) and later referred to as “array CGH” (Pinkel et al., 1998). In aCGH, the “test” genome being interrogated is labeled with one type of fluorescent molecule (e.g., Cy5), and a “normal” or “reference” control genome is labeled with another type of fluorescent molecule (e.g., Cy3) (Figure 9.6). The labeled DNAs are combined, denatured and hybridized to an array of DNA fragments or oligonucleotides on a microscope slide, with each DNA fragment or oligonucleotide representing a unique part of the human genome. The labeled DNAs are then allowed to hybridize to their complementary DNA sequences on the array, in a stoichiometric fashion, such that – by measuring the fluorescence ratio of the two fluorescent dyes at each spot on the array – one can infer the relative copy number of that particular DNA sequence in the genome being tested, with respect to the reference genome (Figure 9.6). There have been many advances in aCGH technology over the past 5 years. Originally, aCGH platforms contained several hundred or even up to a thousand, large-insert DNA clones (e.g., BAC clones that have an average insert size of about 120–150 kb), which recapitulated the human genome with a clone per 3 Mb. More recently, the trend has been to manufacture arrays that use smaller DNA sequences as hybridization targets but with increasing number of targets on an array. On such arrays, targets can be oligonucleotides of 45–75 bases in length that have been designed to have similar annealing temperatures (isothermic), based primarily on the length of the oligonucleotide
Detecting CNVs in a Genome-Wide Manner
■
113
Gain
Microarray with DNA segments No change
Loss Reference DNA
Test DNA (a)
Log2 ratio
1.2 1.6 0
0.6 1.2
Position along the genome
(b)
Figure 9.6 (a) A schematic of an aCGH assay where a test genome (labeled with Cy5, denoted in green) is cohybridized with a reference genome (labeled with Cy3, denoted in red). The DNA probes are mixed and allowed to hybridize to its complementary sequences on the array in a stoichiometric fashion. Fluorescence intensities of the spots on the microarray (each containing a specific DNA sequence) are measured and DNA sequences occurring in greater copy number in the test than in the reference will result in more green fluorescence for those spots on the microarray. A lower copy number of the same DNA sequences will result in more red fluorescence. (b) Typically the log2 of the fluorescence ratios for each DNA segment on the array is then plotted from one end of a chromosome to the other. A gain is indicated in green and a loss in red. DNA segments having no significant change in DNA copy number in the test sample (with respect to the reference DNA) is indicated in yellow.
and its GC base pair content. Two companies that produce such arrays are NimbleGen Systems, Inc. (www.nimblegen.com) and Agilent, Inc. (www.agilent.com/chem/goCGH). NimbleGen uses a programmable mirror array to synthesize 385,000 oligonucleotides (more recently 2.1 million targets) directly on a glass surface using photolithography. Agilent, on the other hand, uses ink-jet technology to synthesize 244,000 oligonucleotides (more recently 1.1 million targets), on a spot-by-spot basis. When using oligonucleotide-based arrays, it is important to note that their primary disadvantage is that each oligonucleotide probe tends to have lower signal-to-noise ratios than a large-insert
genomic clone target. This results in more experimental “noise” for oligonucleotide-based aCGH assays. Indeed, the typical standard deviation of log2 ratios for an oligonucleotide-based array is approximately 0.25–0.3, five times the standard deviation of log2 ratios obtained for BAC-based arrays. However, the inclusion of hundreds of thousands to millions of targets on a given oligonucleotide-based platform provides an assay with an increased resolution. The “effective resolution” of one of these platforms depends largely on the minimum number of consecutive probes needed to confidently call a CNV, which in turn is a function of how well the target sequences were chosen to accurately and consistently
114
CHAPTER 9
■
Copy Number Variation and Human Health
respond to copy number changes. Hence, a particular platform that has 500,000 targets and requires only three consecutive probes to make a confident CNV call actually has a higher effective resolution than an array platform with one million targets, but requiring 10 consecutive probes to make a CNV call, assuming that both platforms distribute targets evenly throughout the genome. Genotyping Arrays High-throughput array technologies for identifying SNPs can also be used to identify CNVs (see Chapter 8). In general, these arrays contain short targets (20–30 base oligonucleotides) that make them ideal for detecting single base alterations, but less ideal for identifying CNVs (especially when compared to long oligonucleotide-based arrays). For genotyping arrays, only a single labeled DNA source (“test sample”) is hybridized, and the signal intensities obtained at each hybridized target appear to have a linear relationship with respect to copy numbers of that particular DNA sequence in the test genome. For example, if a given DNA sequence has four copies in test genome “X” and only two copies in test genome “Y,” the signal intensity obtained for that DNA sequence when test genome “X” is hybridized would be twice that of when test genome “Y” is hybridized. In general, genotyping platforms can detect larger CNVs (especially when there is a higher level of copy number change) or smaller CNVs (that are detectable by numerous targets on the array platform). Affymetrix (www.affymetrix.com) SNP arrays contain targets that are ~25 bases long and the predecessor to the Affymetrix 500K array was recently used to identify CNVs in 270 HapMap individuals (Redon et al., 2006). Illumina Inc. (www.illumina.com) has designed a genotyping platform that uses 50 base oligonucleotides attached to indexed beads on a glass slide. Labeled test DNAs are hybridized to this slide, followed by primer extension and then immunofluorescence detection. Peiffer et al. (2006) recently showed that the ability of Illumina genotyping platforms have also been used to detect CNVs in both constitutional and tumor samples. Clearly, there would be significant benefits to having platforms that are capable of accurately determining both SNP and CNV genotypes, including reduced reagent costs and expenditure of minimal amounts of DNA. However, the fluorescence intensity data obtained from these genotyping platforms typically have more noise when trying to obtain copy number information than do long oligonucleotide-based arrays. Hence, both Affymetrix and Illumina are now designing “next generation” arrays that incorporate thousands of non-polymorphic probes (i.e., probes that do not target known SNPs) that fall within known CNV regions. It is thought that a CNV can be confidently detected if enough targets are strategically chosen for a given CNV region and included in the array. In other words, one probe may not be able to reliably detect a single copy loss in a CNV region but cumulative data from one hundred targets in the same CNV region may result in a consistent and confident CNV call. Whole-Genome Sequence Comparisons CNVs can also be detected via whole-genome sequence comparison analyses. The main advantage of this method for
identifying CNVs is that the acquisition of fine-scale genomic architecture of CNVs (i.e., accurate CNV sizes and breakpoint information). A major disadvantage of this approach is the limited number of individuals for which whole-genome sequence data are currently available. The first human whole-genome sequence was made available from the Human Genome Project. This reference human genome is actually a compilation of DNA sequences from over ten different individuals with ~67% of the DNA sequences originating from the RPCI-11 DNA library, derived from a single individual male. Tuzun et al. (2005) aligned end-sequence data from thousands of fosmid clones from the G248 DNA library (derived from a single North American female) and compared these with the human reference genome sequence. Taking advantage of the tight size restriction of fosmid clones, they were able to identify genomic gains/losses in the reference genome, when pairs of end-sequences aligned with intervening spacing significantly greater or less than the expected 40 kb. In this manner, 241 CNVs were identified in one of these two healthy and presumably normal individuals. More recently, sequencing technologies have advanced to a point where complete human genome sequences can be obtained more efficiently and cost-effectively than previously possible. For example, using Solexa or 454 DNA sequencing approaches, an individual’s complete genome sequence can now be obtained within months and at costs of under $250,000. Undoubt-edly, these advances will be used to comprehensively identify and define CNVs at a DNA sequence resolution, provided that a minimal fold-coverage of the person’s genome is achieved. Recent studies that have obtained complete genome sequences for different individuals have reported thousands of CNVs in a given individual, encompassing hundreds of millions of bases of DNA (Korbel et al., 2007; Levy et al., 2007). A major advantage of utilizing DNA sequence comparison strategies is the ability to identify balanced chromosomal rearrangements that cannot be detected by aCGH-based methods. For example, Tuzun et al. (2005) found evidence for 56 inversion breakpoints in their comparative analysis of the two genomes of two individuals, and Korbel et al. (2007) found 132 inversion breakpoints when comparing the genomes of two different individuals.
ASSOCIATION OF CNVS WITH DISEASE AND DISEASE SUSCEPTIBILITY Genomic imbalances, including CNVs, can contribute to human diseases in at least two ways. First, certain genomic imbalances appear to directly cause neurodevelopmental diseases, that can occur at birth or even later on in life. These genomic imbalances (referred by some as “pathogenic” CNVs) are usually de novo in nature and recent estimates have associated specific genomic imbalances with as many as 50 such genetic syndromes (http://www.sanger.ac.uk/PostGenomics/decipher/). Most of the remaining known genomic imbalances – sometimes referred
Association of CNVs to Disease and Disease Susceptibility
to as “benign” CNVs because of their identification in healthy individuals – actually have more subtle consequences on human health. Fcgr3 Variation in Glomerulonephritis Glomerulonephritis is a major contributor to human kidney failure. Fcgr3 is a gene that encodes for a receptor found on the surfaces of macrophages and that has low-affinity binding properties to immunoglobulin G. The copy number of Fcgr3 can vary in humans and among rat strains from 0 to 4 per diploid cell. Individuals with fewer copies of this gene, due to deletions of paralogous Fcgr3 genes (which appear to have a negative regulatory effect on the “full length” Fcgr3 gene/gene products), demonstrate increased macrophage activity and an autoimmune response (Aitman et al., 2006; Fanciulli et al., 2007). DEFB4 CNV in Crohn Disease Human -defensins are a family of genes predominantly secreted from leukocytes and epithelial tissues. -defensins are small proteins (15–20 residues) that function in antimicrobial defense by penetrating a microbe’s cell membrane and cause microbial death in a manner similar to that of antibiotics. In the presence
■
115
of interleukin 1-alpha (IL-1), which is secreted by macrophages and other immunologically relevant cell types at the site of tissue inflammation, the expression levels of the -defensin gene, DEFB4, increases (O’Neil et al., 1999), to protect the tissue from further microbial invasion. Therefore, individuals with a lower copy number of this -defensin have decreased immunity against microbes and increased susceptibility to Crohn disease (Inflammatory bowel disease 1) (Fellermann et al., 2006; Naser et al., 2004). UGT2B17 Variation in Prostate Cancer Park and colleagues (2006) found that CNV of the UGT2B17 gene (Murata et al., 2003; Redon et al., 2006; Wilson et al., 2004) confers differential susceptibility to prostate cancer (Figure 9.7). Testosterone is normally processed into dihydrotestosterone (DHT) and other androgens within human cells. In the presence of functioning UGT2B17 gene product, excess DHT molecules are converted into water-soluble glucuronic acid which can be subsequently eliminated from the cell. When UGT2B17 is deleted, this process becomes less efficient and results in greater amounts of DHT within the cells. Increased intracellular DHT concentrations lead to increased interactions with androgen
Dissolution / expulsion
DHT Testosterone Glucuronic acid Androgen receptor (a) Normal copy number of UGT2B17 gene
DHT Testosterone Androgen receptor Deleted UGT2B17 (b) Two copy deletion of UGT2B17 gene
Figure 9.7 Potential mechanism of the UGT2B17 CNV in prostate cancer pathogenesis. (a) Testosterone is normally processed into DHT and other androgens. When there are two copies of UGT2B17, the gene product converts excess molecules of DHT into the water-soluble glucuronic acid and is subsequently eliminated from the cell. (b) When there is a homozygous deletion of UGT2B17, the DHT-sequestration pathway no longer exists, increasing endogenous interaction with androgen receptors (AR) and leading to elevated cell proliferation, which in some cases can result in the development of prostate cancer (Park et al., 2006).
116
CHAPTER 9
■
Copy Number Variation and Human Health
receptors that cause elevated cell proliferation. When this occurs in the prostate cells, it can result in prostatic tumorigenesis. Interestingly, while UGT2B17 null deletions significantly increases prostate cancer susceptibility as a whole, stratified analysis revealed that this correlation to prostate cancer susceptibility is only significant in the Caucasian populations studied, but not among African-Americans. The CNV of Complement Component C4 in Systemic Lupus Erythematosus (SLE) Although a link between the complement component C4 (and its isotypes, C4A and C4B) and SLE has been previously reported (Hauptmann et al., 1974a,b), a recent study suggests that the gene’s variable copy number actually serves as a significant risk factor for the disease (Yang et al., 2007). The complement system is comprised of over 20 proteins/protein fragments that normally circulate in the blood but, when activated, cause a biochemical cascade that clears pathogens from the human body, often by forming new transmembrane channels in the pathogen and causing osmotic lysis of the target cell. The median copy number of complement component C4 is 4 but can range from 0 to 5 copies among humans (Yang et al., 2007). Low copy numbers of this gene are correlated with increased risk for SLE (Fanciulli et al., 2007;Yang et al., 2007). CCL3L1 CNVs in HIV/AIDS A number of studies on HIV/AIDS have focused on the gene copy number and protein levels of CCL3L1 and CCR5 (Gonzalez et al., 2001, 2005) in diseased and healthy individuals. CCR5 is a major co-receptor for the HIV-1 virus as well as the CCL3L1 chemokine. It was recently found that the combination of low CCL3L1 gene copy number, which is populationspecific, and high protein levels of CCR5 showed the greatest increase in susceptibility to AIDS (Gonzalez et al., 2005). It is speculated that increased CCR5 protein levels lead to greater number of CD4CCR5 receptor complexes on the surfaces of T cells and decreased copy number of CCL3L1 leads to decreased expression (and protein levels) of this chemokine, leaving more CD4CCR5 receptor complexes available for binding with the HIV-1 virus.
IMPLICATIONS OF CNVS Disease Association Studies SNPs have become powerful markers for identifying important disease loci in genetic association studies. However, recent analysis of the complete DNA sequence from a single human individual has shown that CNVs (and other structural genomic variants including small balanced chromosomal rearrangements) account for some 22% of all genetic variation events in the individual and 74% of the total DNA sequence variation, when compared to the human reference genome (Levy et al., 2007). Although it is premature to predict the relative contribution of
CNVs to the etiology of common, complex diseases (compared to SNPs), it is clear that CNVs represent a substantial component of human genetic variation that should not be ignored in future disease association studies. Indeed, the presence of CNVs in the human genome has actually reduced the power of certain SNPs. For example, SNPs that lie within CNV regions are difficult to genotype, result in Hardy–Weinberg equilibrium distortions, and are often erroneously calculated as having reduced linkage disequilibrium to nearby causative genomic regions. Moreover, CNVs are far more complex in nature than SNPs and demand appreciation of several factors in order to use this form of genetic variation appropriately in disease association studies. First, the absolute copy number (rather than relative copy number obtained from aCGH-based experiments) needs to be established for each CNV. Second, the exact genomic location of duplicated CNVs should also be considered. For example, the presence of a third copy of a gene may have dramatically different phenotypic effects depending on where the third copy occurs in the genome. Finally, the precise boundaries (at the DNA sequence level) of each CNV should also be known, as should the specific allelic state of a CNV (e.g., two copies of a gene could be distributed as one copy per chromosome or both copies on a single chromosome). Without this level of information, the power of CNVs in genetic association studies diminishes. Indeed, of the several thousand CNVs that have been identified to date, only a small percent have actually been annotated and genotyped to this precision. Some efforts have been made to determine if CNVs are in linkage disequilibrium with nearby SNPs. If so, this would allow for specific CNV alleles to be assayed indirectly with a subset of well-characterized SNPs (i.e., “tagging” SNPs) (see Chapter 2). Initial observations suggested that a subset of CNVs did appear to be ancestral in nature and therefore taggable by specific SNPs (McCarroll et al., 2006). Larger CNVs, especially those that are in segmental, duplication-rich areas of the genome, appear to be less taggable by known SNPs (Locke et al., 2006). Part of the reason for this may be that the mutation rate for such CNVs is higher than the mutation rates for nearby SNPs (recall that NAHR mutations rates are estimated to be substantially higher than that of SNPs). Overall, this suggests that, for association studies, a substantial number of CNVs may need to be genotyped directly. Pharmacogenetics Since many CNV genes are involved in metabolism and drug detoxification, it has been speculated that CNVs may also make significant contributions to future pharmacogenomic studies (Ouahchi et al., 2006). For example, the CYP2D6 genes code for enzymes that are involved in the metabolism of more than 30 medications that include antiarrhythmics, antihypertensives, beta-blockers, etc. CYP2D6 CNVs have been identified and shown to result in gene products with differential metabolism efficiency (Rotger et al., 2007). Similarly, CNVs of genes involved in metabolism may explain some cases of interindividual variation in drug toxicity. McCarroll et al. (2005) and Conrad et al. (2006) found more than 120 CNV genes that were
Implications of CNVs
■
117
14p13 14p12
RP11-98N22 14p11.2-11.1 14q11.1 14q11.2 14q12
RP11-557015
13.1 14q13.3 14q21.2 14q21.3 14q22.1 14q22.2 14q23.1 14q23.2 14q24.1 14q24.2 14q24.3 14q31.1 14q31.3 14q32.12 14q32.2 14q32.32 14q32.33
4
5
5
3.
3
5
2.
2
Human 14
1.
08 1. 3 9 0.
67
0.
5
0.
4 0. 33 0. 29 0. 25
(a)
(b)
RP11-98N22 RP11-557015
Figure 9.8 Example of a de novo genomic imbalance that is likely causative of the clinical phenotype of a patient seen in the clinical genetic clinic. (a) A chromosome 14 profile from a 1 Mb resolution aCGH assay that has been run in dye swap. Deviation of the red line to the right of the expected 1:1 ratio and the blue line to the left of the expected 1:1 ratio is indicative of a genomic loss in the patient DNA. Here, nine BAC clones (each clone represented by a dot on the array profile) are within the deleted region, suggesting a 9–10 Mb sized deletion. The position of BAC clones used in FISH confirmation studies are indicated in red and green. (b) A metaphase spread from the patient after a two-color FISH confirmation assay using a clone within the deletion interval (RP11-557O15) and a control clone outside of the deletion interval (RP11-98N22). The absence of the green signal on one of the chromosome 14s (indicated with a white arrowhead) confirms the aCGH findings. Parental FISH studies were normal (not shown), indicating that this chromosomal aberration was a de novo event.
homozygously deleted. Individuals harboring such homozygous deletions presumably have low toxicity tolerance to medications that depend heavily on the homozygously deleted CNV gene product for proper metabolism. Rapid and accurate identification of these individuals should be made a priority in pharmacogenomic research (see also Chapter 27). Clinical Cytogenetic Diagnostics Array CGH-based techniques are now being more widely used in the clinical cytogenetic diagnostic arena to identify smaller genomic imbalances that may be associated with neurodevelopmental disorders. Indeed, it has been estimated that aCGHbased assays are now detecting apparently pathogenic genomic imbalances in as much as 20% of cases that have had normal results from chromosome-banded karyotyping tests (Figure 9.8). However, the recent recognition of widespread CNVs among the genomes of healthy individuals has made interpretation of aCGH-based assays more difficult [for a recent review on this, see Lee et al. (2007)]. Specifically, all CNVs detected in an
aCGH clinical test need to be assessed for their potential to contribute to the clinical presentation. One of the primary means for assessing pathogenicity of a CNV is by determining whether or not the CNV is de novo. Indeed, 2–3% of children are born with a major birth defect that is sporadic in nature, and the genetic alteration causing these birth defects is similarly assumed to be sporadic in nature. This strategy for determining the pathogenicity of a genomic imbalance in “constitutional” genetic testing is not new and has been applied by cytogeneticists for decades (albeit at the chromosomal banding analysis level). Therefore, for genomic imbalances detected by aCGH assays, it must be determined if the same genomic imbalance is observed in either of the biological parents, and close attention must be paid to the absolute copy number of the CNV (i.e., not just the relative copy number that is obtained from aCGH experiments), breakpoint information when available and genomic distribution of the CNV DNA segment. Similarly, determining whether a given CNV is present in affected (or unaffected) relatives can also be highly informative.
118
CHAPTER 9
■
Copy Number Variation and Human Health
Another factor to consider is whether the CNV has been found in other unrelated but affected individuals with similar clinical phenotypes (using CNV databases such as DECIPHER) or among healthy individuals (using CNV databases such as the Database of Genomic Variants). This information is only as useful as the accuracy of the CNV entries in these databases (i.e., minimal false positives and precise definition of the architecture of each CNV entry). Moreover, the gene content within a given CNV will also dictate the clinical interpretation of that CNV. Finally, the overall size of the CNV can be a minor indicator of its pathogenicity, with the caveat that “benign” CNVs, as large as 10 Mb in size in gene-poor regions of the genome, have been documented (Hansson et al., 2007). Accurate clinical interpretation of CNVs will improve as our knowledge of the structure and function of the human genome increases. Clearly, CNVs cannot be considered as isolated genomic aberrations and eventually should be interpreted in the context of all genomic variants, including other CNVs as well as SNPs. For example, the same deletion CNV in a healthy individual may uncover a recessive SNP mutation on the other allele in an unrelated individual – leading to a drastically different phenotype. Since we are at a stage in clinical cytogenetic diagnostics where the technology and the data being obtained are beyond our understanding of human genome function, caution continues to be advised in the application of genome-wide aCGH assays in prenatal diagnostics. Instead, targeted aCGH platforms that primarily utilize probes with well-defined associations with genetic disorders have been used for prenatal diagnoses.
CONCLUSIONS It was previously touted that the genomes of healthy individuals were 99.9% similar (Lander et al., 2001). Over the last 3 years, however, it has become obvious that there is more genetic variation than had been appreciated and that a substantial portion is in the form of structural genomic variation, including genomic imbalances of DNA segments greater than 1 kb in length (CNVs). As a result, CNVs are now requiring special consideration in many aspects of genomic medicine, including disease association studies, pharmacogenomics and clinical cytogenetic diagnostics. While our understanding of the locations, structures and frequencies of CNVs remains rudimentary, it is anticipated that there will be rapid growth in CNV knowledge over the next few years. This growth will be essential for accurately integrating this type of genetic variation with SNPs (and other forms of genetic alterations) for each human being. New findings will lead researchers and clinicians toward a greater appreciation of the genetic factors that make each person so unique, which is essential to the practice of genomic and personalized medicine.
ACKNOWLEDGEMENTS The authors would like to express their gratitude to George Perry for providing the image for Figure 9.2 and Marc Listewnik for his assistance in generating Figure 9.8.
REFERENCES Aitman, T.J., Dong, R., Vyse, T.J., Norsworthy, P.J., Johnson, M.D., Smith, J. et al. (2006). Copy number polymorphism in Fcgr3 predisposes to glomerulonephritis in rats and humans. Nature 439, 851–855. Conrad, D.F., Andrews, T.D., Carter, N.P., Hurles, M.E. and Pritchard, J.K. (2006). A high-resolution survey of deletion polymorphism in the human genome. Nat Genet 38, 75–81. Conrad, D.F. and Hurles, M.E. (2007). The population genetics of structural variation. Nat Genet 39, 530–536. Down, J.L.H. (1866). Observations on an ethnic classification of idiots. London Hosp Clin Lect Rep 3, 259. Fanciulli, M., Norsworthy, P.J., Petretto, E., Dong, R., Harper, L., Kamesh, L., Heward, J.M. et al. (2007). FCGR3B copy number variation is associated with susceptibility to systemic, but not organ-specific, autoimmunity. Nat Genet 39, 721–723. Fellermann, K., Stange, D.E., Schaeffeler, E., Schmalzl, H., Wehkamp, J., Bevins, C.L., Reinisch, W. et al. (2006). A chromosome 8 genecluster polymorphism with low human beta-defensin 2 gene copy number predisposes to Crohn disease of the colon. Am J Hum Genet 79, 439–448. Feuk, L., Marshall, C.R., Wintle, R.F. and Scherer, S.W. (2006). Structural variants: Changing the landscape of chromosomes and design of disease studies. Hum Mol Genet 15, R57–R66. Freeman, J.L., Perry, G.H., Feuk, L., Redon, R., McCarroll, S.A., Altshuler, D.M. et al. (2006). Copy number variation: new insights in genome diversity. Genome Res 16, 949–961.
Gonzalez, E., Dhanda, R., Bamshad, M., Mummidi, S., Geevarghese, R., Catano, G., Anderson, S.A. et al. (2001). Global survey of genetic variation in CCR5, RANTES, and MIP-1alpha: Impact on the epidemiology of the HIV-1 pandemic. Proc Natl Acad Sci USA, 5199–5204. Gonzalez, E., Kulkarni, H., Bolivar, H., Mangano, A., Sanchez, R., Catano, G. et al. (2005).The influence of CCL3L1 gene-containing segmental duplications on HIV-1/AIDS susceptibility. Science 307, 1434–1440. Hansson, K., Szuhai, K., Knijnenburg, J., van Haeringen, A. and de Pater, J. (2007). Interstitial deletion of 6q without phenotypic effect. Am J Med Genet A 143, 1354–1357. Hauptmann, G., Grosshans, E. and Heid, E. (1974a). Lupus erythematosus syndrome and complete deficiency of the fourth component of complement. Boll Ist Sieroter Milan 53(Suppl), 228. Hauptmann, G., Grosshans, E., Heid, E., Mayer, S. and Basset, A. (1974b). Acute lupus erythematosus with total absence of the C4 fraction of complement. Nouv Presse Med 3, 881–882. Iafrate, A.J., Feuk, L., Rivera, M.N., Listewnik, M.L., Donahoe, P.K., Qi,Y. et al. (2004). Detection of large-scale variation in the human genome. Nat Genet 36, 949–951. Korbel, J.O., Urban, A.E., Affourtit, J.P., Godwin, B., Grubert, F., Simons, J.F. et al. (2007). Paired-end mapping reveals extensive structural variation in the human genome. Science 318, 420–426.
Recommended Resources
Lander, E.S., Linton, L.M., Birren, B., Nusbaum, C., Zody, M.C., Baldwin, J. et al. (2001). Initial sequencing and analysis of the human genome. Nature 409, 860–921. Lee, C., Iafrate, A.J. and Brothman, A.R. (2007). Copy number variations and clinical cytogenetic diagnosis of constitutional disorders. Nat Genet 39, S48–S54. Lejeune, J., Gautier, M. and Turpin, R. (1959). Etude des chromosomes somatiques de neuf enfants mongoliens. C R Hebd Seances Acad Sci 248, 1721–1722. Levy, S., Sutton, G., Ng, P.C., Feuk, L., Halpern, A.L., Walenz, B.P. et al. (2007). The diploid genome sequence of an individual human. PLoS Biol 5, e254. Locke, D.P., Sharp, A.J., McCarroll, S.A., McGrath, S.D., Newman, T.L., Cheng, Z. et al. (2006). Linkage disequilibrium and heritability of copy-number polymorphisms within duplicated regions of the human genome. Am J Hum Genet 79, 275–290. McCarroll, S.A., Hadnott, T.N., Perry, G.H., Sabeti, P.C., Zody, M.C., Barrett, J.C. et al. (2006). Common deletion polymorphisms in the human genome. Nat Genet 38, 86–92. Murata, M., Warren, E.H. and Riddell, S.R. (2003). A human minor histocompatibility antigen resulting from differential expression due to a gene deletion. J Exp Med 197, 1279–1289. Naser, S.A., Ghobrial, G., Romero, C. and Valentine, J.F. (2004). Culture of Mycobacterium avium subspecies paratuberculosis from the blood of patients with Crohn’s disease. Lancet 364, 1039–1044. Nguyen, D.Q., Webber, C. and Ponting, C. (2006). Bias of selection on human copy-number variants. PLoS Genet 2, e20. O’Neil, D.A., Porter, E.M., Elewaut, D., Anderson, G.M., Eckmann, L., Ganz, T. and Kagnoff , M.F. (1999). Expression and regulation of the human beta-defensins hBD-1 and hBD-2 in intestinal epithelium. J Immunol 163, 6718–6724. Ouahchi, K., Lindeman, N. and Lee, C. (2006). Copy number variants and pharmacogenomics. Pharmacogenomics 7, 25–29. Park, J., Chen, L., Ratnashinge, L., Sellers, T.A., Tanner, J.P., Lee, J.H. et al. (2006). Deletion polymorphism of UDP-glucuronosyltransferase 2B17 and risk of prostate cancer in African American and Caucasian men. Cancer Epidemiol Biomarkers Prev 15, 1473–1478. Perry, G.H., Dominy, N.J., Claw, K.G., Lee, A.S., Fiegler, H., Redon, R., Werner, J.,Villanea, F.A., Mountain, J.L., Misra, R., Carter, N.P., Lee, C. and Stone, A.C. (2007). Diet and the evolution of human amylase gene copy number variation. Nat Genet 39, 1256–1260. Peiffer, D.A., Le, J.M., Steemers, F.J., Chang, W., Jenniges, T. and Garcia, F.K. (2006). High-resolution genomic profiling of
■
119
chromosomal aberrations using Infinium whole-genome genotyping. Genome Res 16, 1136–1148. Pinkel, D., Segraves, R., Sudar, D., Clark, S., Poole, I., Kowbel, D. et al. (1998). High resolution analysis of DNA copy number variation using comparative genomic hybridization to microarrays. Nat Genet 20, 207–211. Redon, R., Ishikawa, S., Fitch, K.R., Feuk, L., Perry, G.H., Andrews,T.D. et al. (2006). Global variation in copy number in the human genome. Nature 444, 444–454. Rotger, M., Saumoy, M., Zhang, K., Flepp, M., Sahli, R., Decosterd, L. et al. (2007). The Swiss HIV Cohort Study. Partial deletion of CYP2B6 owing to unequal crossover with CYP2B7. Pharmacogenet Genomics 17, 885–890. Scherer, S.W., Lee, C., Birney, E., Altshuler, D.M., Eichler, E.E., Carter, N.P. et al. (2007). Challenges and standards in integrating surveys of structural variation. Nat Genet 39, S7–S15. Sebat, J., Lakshmi, B., Troge, J., Alexander, J., Young, J., Lundin, P., Maner, S. et al. (2004). Large-scale copy number polymorphism in the human genome. Science 305, 525–528. Shaffer, L.G. and Lupski, J.R. (2000). Molecular mechanisms for constitutional chromosomal rearrangements in humans. Annu Rev Genet 34, 297–329. Solinas-Toldo, S., Lampel, S., Stilgenbauer, S., Nickolenko, J., Benner, A., Dohner, H. et al. (1997). Matrix-based comparative genomic hybridization: Biochips to screen for genomic imbalances. Genes Chromosomes Cancer 20, 399–407. Stranger, B.E., Forrest, M.S., Dunning, M., Ingle, C.E., Beazley, C., Thorne, N. et al. (2007). Relative impact of nucleotide and copy number variation on gene expression phenotypes. Science 315, 848–853. Tuzun, E., Sharp, A.J., Bailey, J.A., Kaul, R., Morrison, V.A., Pertz, L.M. et al. (2005). Fine-scale structural variation of the human genome. Nat Genet 37, 727–732. Wilson, W., III, Pardo-Manuel de Villena, F., Lyn-Cook, B.D., Chatterjee, P.K., Bell, T.A., Detwiler, D.A. et al. (2004). Characterization of a common deletion polymorphism of the UGT2B17 gene linked to UGT2B15. Genomics 84, 707–714. Yang, Y., Chung, E.K., Wu, Y.L., Savelli, S.L., Nagaraja, H.N., Zhou, B. et al. (2007). Gene copy-number variation and associated polymorphisms of complement component c4 in human systemic lupus erythematosus (SLE): Low copy number is a risk factor for and high copy number is a protective factor against SLE susceptibility in European Americans. Am J Hum Genet 80, 1037–1054.
RECOMMENDED RESOURCES Affymetrix, Inc. (http://www.affymetrix.com) Agilent, Inc. (www.agilent.com/chem/goCGH) Chromosome Abnormality Database (http://www.ukcad.org. uk/cocoon/ukcad/) Database of Genomic Variants (http://projects.tcag.ca/variation) DECIPHER (http://www.sanger.ac.uk/PostGenomic/decipher/) Ensembl Genome Browser (http://www.ensembl.org/index.html) European Cytogenetics Association Register of Unbalanced Chromosome Aberrations (http://www.ecaruca.net)
Human Paralogy Database (http://humanparalogy.gs.washington.edu/ structuralvariation) Human Segmental Duplication Database (http://projects.tcag.ca/ humandup/) Illumina Inc. (www.illumina.com) NimbleGen Systems, Inc. (www.nimblegen.com) UCSC Genome Browser (http://www.genome.ucsc.edu/)
CHAPTER
10 Inter-Species Comparative Sequence Analysis: A Tool for Genomic Medicine Anthony Antonellis and Eric D. Green
INTRODUCTION In 2001, the Human Genome Project completed a draft sequence of the human genome (International Human Genome Sequencing Consortium, 2001). Since then, the human genome sequence has been refined (International Human Genome Sequencing Consortium, 2004) and attention has turned to generating the genome sequences of a number of biomedically relevant vertebrate species, including mouse (Mouse Genome Sequencing Consortium, 2002), rat (Rat Genome Sequencing Project Consortium, 2004), dog (Lindblad-Toh et al., 2005), chicken (International Chicken Genome Sequencing Consortium, 2004), chimpanzee (Chimpanzee Sequencing and Analysis Consortium, 2005), and others (www.intlgenome.org). A major rationale for elucidating these additional sequences is that their comparison with the human genome sequence would reveal important details about the functional landscape of all human chromosomes, which in turn would empower studies aiming to understand the genetic basis of human disease (Collins et al., 2003). In reality, our current understanding of the functional sequence elements residing within the 3 billion base pairs that constitute the human genome is relatively nascent. Needed are approaches that can identify such functional elements – including both sequences that encode protein (coding sequences) and those that function in other ways (non-coding Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 120
functional sequences). Importantly these functional sequences may harbor disease-associated mutations. Comparative sequence analysis (CSA) involves crossanalyzing related nucleotide or protein sequences, so as to identify similarities and differences among them. Here, we limit our discussion to one type of CSA – that involving the comparison of genomic sequences from different species. Such interspecies CSA can be used to identify genomic sequences that are highly conserved among species. When applied to the analysis of sequences from distantly related species, conserved sequences can be identified that reflect genomic regions that have not changed substantially over millions of years of evolutionary time; such sequences represent strong candidates for being functionally important. For example, protein-coding sequences are typically highly conserved among vertebrates, even between organisms as distantly related as human and fish (Aparicio et al., 2002). Similarly, many non-coding functional sequences, such as those that regulate the expression of genes, also tend to be highly conserved (e.g., Loots et al., 2000; Prabhakar et al., 2006a). In addition to identifying similarities, CSA can be used to identify differences between species. Indeed, comparing genomic sequences among closely related species can reveal sequence changes that are specific to certain lineages (e.g., human-specific sequences). Importantly, such analyses can provide insight into the evolution
Performing Comparative Sequence Analysis: Resources and Methods
of human-specific traits (e.g., Pollard et al., 2006; Prabhakar et al., 2006b). Because of its ability to provide clues about functional sequences in the genome, CSA can facilitate the identification and characterization of genes associated with human disease. Specifically, CSA can benefit such studies in several ways. First, CSA can be used to identify and prioritize the genomic regions that are most likely to harbor disease-causing mutations. For example, it is estimated that roughly 5% of the human genome sequence is functionally important (Lindblad-Toh et al., 2005; Mouse Genome Sequencing Consortium, 2002; Rat Genome Sequencing Project Consortium, 2004), but the precise nucleotides constituting that 5% are, at present, not known; CSA can be used to help hone in on that small fraction of functional sequence, prioritizing it for more detailed study in patients with genetic disease. Second, CSA can be used to characterize disease-associated sequence variants once they are identified, helping to classify them as disease-causing mutations versus innocent polymorphisms. For example, demonstrating that a variant occurs at a highly conserved nucleotide (or amino-acid residue) provides support for its relevance in the disease. Finally, CSA can aid in the design of functional analyses aiming to characterize a gene of interest in another species. For example, elucidating the function of a newly discovered disease gene often requires studying the functional consequences of mutating that gene in a model organism, such as yeast, worm, fruit fly, zebrafish, or mouse. Such studies allow rapid, in vivo analyses of gene function and mutational consequences that are not possible in patient populations. In this setting, CSA can help to identify the presence or absence of the corresponding gene (i.e., the orthologous gene) in another species as well as the conservation of affected nucleotides or amino-acid residues; if present, that gene can then be studied in greater detail using that species as a model system. These applications of CSA have implications for the practice of genomic medicine. Specifically, CSA can directly aid the identification and characterization of sequence variants, both common and rare, that are responsible for disease onset (or severity) and differential response to therapeutics. In the long run, such information will be valuable for providing patients with accurate diagnoses and prognoses, and for directing a course for proper care. In this chapter, we describe how to access and utilize resources for performing CSA, demonstrate how such studies can help identify and characterize disease-associated mutations, and discuss the role of this approach in genomic medicine.
PERFORMING COMPARATIVE SEQUENCE ANALYSIS: RESOURCES AND METHODS Step 1: Species Selection and Sequence Acquisition CSA generally involves a series of sequential steps, with the typical ones illustrated in Figure 10.1. Different resources and methodologies are utilized in each of these steps. The first step
■
121
involves acquiring genomic sequences from species of interest (Figure 10.1, Step 1), with the choice of species often dictated by the biological question(s) being asked. For example, the study of the basic genomic elements that are relevant for vertebrate development requires the acquisition of sequences from distantly related vertebrate species (Plessy et al., 2005). In contrast, examination of mammalian- or primate-specific genomic features requires sequences from discrete lineages (Prabhakar et al., 2006a). Often, the initial questions are quite broad and not associated with a priori expectations, in which case it is prudent to collect all available (i.e., already generated) genomic sequence data for subsequent CSA. Once the species have been selected, the relevant sequences for a genomic region of interest must be collected. Genomic sequences can be identified and obtained from several public databases and Web sites. These include the National Center for Biotechnology Information (NCBI; this is the home of GenBank), the University of California at Santa Cruz (UCSC) Genome Browser (Kent et al., 2002), and Ensembl (Stalker et al., 2004) (Table 10.1). Each of these sites has different routines for identifying sequences of interest. For example, genomic sequences can be obtained from the UCSC Genome Browser by selecting a species and performing a query using a gene name, marker, or chromosome band; this site is particularly useful because its graphical interface is tailored for CSA. In a similar fashion, protein sequences can be collected (e.g., for determining the conservation of an amino-acid residue implicated in a human genetic disease). In this case, a text query using the name or symbol of the human protein may be performed using either the HomoloGene or protein databases at NCBI (Table 10.1); this will yield a list of related sequences from various species that can be easily downloaded. A critical aspect of sequence acquisition is ensuring that appropriate sequences are actually obtained. Genomic regions that originate from a common ancestor are considered homologous, and these subdivide based on their origins. Homologous regions generated by genome-duplication events in a given species are considered paralogs. Homologous regions generated by speciation are considered orthologs. For the applications described here, CSA involves the study of orthologous sequences – that is, those from corresponding genomic regions among a set of species. There are two general approaches to ensure that orthologous sequences are obtained. First, alignment tools such as BLAST (Basic Local Alignment Search Tool; Altschul et al., 1990) and BLAT (BLAST-Like Alignment Tool; Kent, 2002) (Table 10.1) can be used to search nucleotide databases, requiring that a “reciprocal best hit” is obtained for a pair of sequences. For example, a human sequence is used to query the dog genome sequence; the most closely related dog sequence is then used to query the human genome sequence. If the most closely related sequence identified in this second query is the same as the starting human sequence, then it can be assumed that those human and dog sequences are orthologous. Second, the content and order of genes within related sequences from two species can be compared, with strong similarity providing support that the sequences are orthologous.
122
CHAPTER 10
■
Comparative Sequence Analysis
Step 1. Acquire Sequence Non-Primate Mammals Non-Mammals Primates
Pig
Human
Cat
Rat
Dog
Mouse
Chicken
Chimp Fugu Cow
Zebrafish
Step 2. Align Sequences ~30 kb
Human Conservation Chimp Mouse Rat Dog Cow Chicken Zebrafish Fugu
Step 3. Identify and Localize Conserved Regions 1
2
3
4
Exons Conserved Regions Coding
Step 4a. Computational Analysis Predict Transcription Factor-Binding Sites
Non-Coding
Step 4b. Experimental Analysis In Vitro
In Vivo
Expression
LacZ Expression
Conserved Predicted Binding Sites Search for Common Motifs
Patient Screening Mutation in Conserved Region
T G C Y C A C Common Motifs
Figure 10.1
Figure Caption on Page 123.
Performing Comparative Sequence Analysis: Resources and Methods
Repetitive elements are a broad class of DNA sequences that occur more than once in a genome. Two types of these sequences that form a substantial fraction of most vertebrate genomes are tandemly repeated repetitive elements and interspersed repetitive
TABLE 10.1 Bioinformatic resources for comparative sequence analysis Sequence Databases and Sequence-Analysis Tools National Center for Biotechnology Information
www.ncbi.nlm.gov
Ensembl Genome Browser
www.ensembl.org
University of California at Santa Cruz (UCSC) Genome Browser
genome.ucsc.edu
RepeatMasker
www.repeatmasker.org
Sequence-Alignment Tools BLAST
www.ncbi.nlm.nih. gov/BLAST
BLAT
genome.ucsc. edu/cgi-bin/hgBlat
ClustalW
www.ebi.ac.uk/clustalw
PipMaker/MultiPipMaker
pipmaker.bx.psu. edu/pipmaker
VISTA
genome.lbl.gov/vista
Programs for Conservation and Motif Detection WebMCS
zoo.nhgri.nih.gov/mcs
ExactPlus
research.nhgri.nih. gov/projects/exactplus
TRANSFAC
www.gene-regulation. com/pub/databases.html
MEME
meme.sdsc.edu/meme
■
123
elements (see Chapter 1). When analyzing and comparing genomic sequences, it is important to consider whether or not to include repetitive elements. Including repetitive sequences can lead to the spurious alignment of non-orthologous sequences; excluding such sequences can result in completely missing functional elements residing therein, and, indeed, repetitive elements can harbor functional sequences (Kamal et al., 2006; Nishihara et al., 2006). For most initial CSA studies, it is reasonable to “mask” (or remove) repetitive sequences, and this can be accomplished in one of two ways. First, masking repetitive elements can be performed while downloading the sequence using tools available at the UCSC Genome Browser and Ensembl (Table 10.1). Alternatively, this step can be accomplished by subjecting already obtained sequence to processing by the program RepeatMasker (Table 10.1). Step 2: Generating Multi-Sequence Alignments The next step in CSA is to generate multi-sequence alignments (Figure 10.1, Step 2), a process in which the orthologous genomic sequences from different species are mapped (or organized) relative to one another. Various computational tools for aligning sequences have been developed and refined in recent years (Table 10.1). There are two general types of alignment methods: those that produce local alignments and those that produce global alignments. Local-alignment methods [e.g., PipMaker and MultiPipMaker (Schwartz et al., 2000; Schwartz et al., 2003)] generate alignments based on similarity of subregions within the sequence, placing more importance on local sequence identity and less on the long-range pattern of alignment across the entire sequence. As such, these methods can produce alignments of orthologous regions that are inverted or otherwise shuffled between species. In contrast, global-alignment methods [e.g., ClustalW (Thompson et al., 1994) and VISTA (Frazer et al., 2004)] generate alignments based on the similarity over the entire length of the sequence, placing more importance on maintaining the relative long-range organization of the genomic region among species and less on local sequence identity. It is generally accepted that local alignments are more appropriate for analyzing genomic regions where the long-range organization may not be well-conserved among species, while
Figure 10.1 Utilization of comparative sequence data. A generic set of steps for using comparative sequence data in human genetic studies are represented. In Step 1, sequences of a genomic region of interest are acquired from relevant (often highly diverged) species. Next (Step 2), these sequences are aligned relative to one another, revealing regions that are identical between a given species and the human reference sequence (indicated by vertical lines within the “track” for each species at the UCSC Genome Browser). Analysis of these sequences allows the overall level of conservation to be established across the region. Levels of conservation above a defined threshold can be used to designate “conserved regions” (Step 3), and these typically reflect both coding and non-coding sequences. Note the hypothetical model for a four-exon gene (in Step 3 and at the top of Step 2). In Step 4, various analyses can be performed to characterize the conserved regions in greater detail. Computational analyses (Step 4a) can include searching for predicted sites of transcription factor binding (based on the sequences of known transcription factor-binding sites) within the conserved regions as well as searching for common motifs (arrow in the upper panel indicates the transcription start site of a gene of interest). Experimental analyses (Step 4b) can include in vitro assays to assess if any or all of a conserved region influences expression of a reporter gene, in vivo testing to see if a conserved region can drive the expression of a marker gene (e.g., LacZ), and screening patients of interest for mutations in a conserved region.
124
CHAPTER 10
■
Comparative Sequence Analysis
global alignments are more appropriate for analyzing genomic regions where the long-range organization is likely to be wellconserved (Frazer et al., 2003). For aligning nucleotide sequences, PipMaker generates alignments between two orthologous sequences, while MultiPipMaker and VISTA generate alignments of up to 20 and 100 sequences, respectively (Table 10.1). For example, to identify sequences conserved in human, mouse, and rat across a genomic region of interest, an investigator would typically submit the human sequence (referred to as the “reference sequence”) to the MultiPipMaker web server followed by the orthologous sequences from mouse and rat. The program then produces a text-based multi-sequence alignment as well as a graphical representation of conservation levels across the region. Importantly, all of the output is indexed relative to the reference sequence, thereby organizing and facilitating downstream analyses. As an aside, multi-sequence protein alignments are typically generated using CLUSTALW (Table 10.1). Briefly, CLUSTALW analyzes a set of text-formatted protein sequences and produces an alignment of the amino-acid sequences. Individual regions of interest can then be examined for conservation (e.g., Figure 10.2A, left panel). Step 3: Identifying Conserved Sequences In contrast to examining relatively simple protein-sequence alignments, analyzing a multi-sequence alignment of a genomic region of interest typically involves scanning across many kilobase pairs of sequence with highly variable levels of conservation. As a result, identifying the most highly conserved regions cannot be accomplished by manual inspection of the alignment. Rather, additional computational analysis is required; indeed, some of the available alignment programs perform this step as well. For example, PipMaker and VISTA will identify the subset of regions with a defined level of conservation between two sequences in a pair-wise fashion (e.g., 70% identity over at least 100 bp). Other stand-alone programs [e.g., WebMCS (Margulies et al., 2003) and ExactPlus (Antonellis et al., 2006); Table 10.1] analyze multi-sequence alignments (e.g., generated by tools such as MultiPipMaker), and identify regions that are highly conserved across sequences from multiple species. It is important to then localize such conserved regions relative to other annotated genomic landmarks (e.g., genes); for example, classifying conserved regions into coding versus non-coding is particularly important (see Figure 10.1, Step 3). The UCSC Genome Browser and Ensembl (Table 10.1) both allow the results of such conservation analyses to be uploaded, thereby facilitating the direct assimilation of user-produced data with all other available annotations. Above, we have emphasized performing CSA in a customized, step-wise fashion. An alternative is to use the results of genome-wide CSA studies that are available on most of the genome browsers. For example, the UCSC Genome Browser provides various graphical representations of sequence conservation across the entire human genome (e.g., Figure 10.1, Step 2), often computed using multiple different methods; further,
such analyses are frequently updated and refined. The data about conserved regions of interest can be readily downloaded from such sites and studied in greater detail. One caution is that while such pre-computed analyses can be rapidly and easily accessed, the genome-wide nature of these CSA studies is associated with inherent limitations, and thus more focused and customized studies of a region of interest typically yield more informative results. Step 4: Computational and Experimental Analyses of Conserved Sequences Once sequence conservation is assessed for a genomic region of interest, computational and experimental analyses are typically performed to further investigate the role of specific sequences. In the case of conserved coding regions, the amino-acid sequence of the encoded protein can be deduced in various species. Mutation screening might reveal a human disease-associated sequence variant that changes an amino acid, and such a change can then be analyzed within the context of knowing the conservation level of that particular residue (e.g., Figure 10.2A, left panel). In addition, computational predictions can be made about mutation-associated structural changes in the encoded protein, and assays can be developed to test the function of the mutant protein. The analysis of conserved non-coding sequences is less straightforward (Figure 10.1, Steps 4a and 4b). This is because we know very little about the “language” of functional noncoding sequences, especially compared to our understanding of how coding sequences encode information about proteins. As a result, the biological relevance of newly encountered disease-associated variants in non-coding sequence is almost never immediately obvious (e.g., compared to a variant that clearly changes an amino-acid residue in the case of coding sequence). As a first step, conserved non-coding sequences can be analyzed for the presence of transcription factor-binding sites using the TRANSFAC database (Matys et al., 2006) and/or commonly occurring sequence motifs using programs such as MEME (Figure 10.1, Step 4a and Table 10.1) (Bailey and Elkan, 1994). Such analyses can help identify sequences involved in transcriptional regulation. In parallel to computational studies, conserved non-coding sequences can also be tested for regulatory activity using relevant in vitro or in vivo assays. Such studies can provide evidence for function that is directly associated with sequence conservation, thereby helping to prioritize further analyses. Finally, conserved regions can be screened for disease-associated mutations in relevant patients; it is then important to assess the extent of conservation at the position of any disease-associated variant that is encountered (e.g., Figure 10.3B, middle panel).
COMPARATIVE SEQUENCE ANALYSIS AND HUMAN GENETIC DISEASE CSA can play a valuable role in establishing the relevance of disease-associated sequence variants. Specifically, human genetic studies often lead to the identification of multiple nucleotide
Comparative Sequence Analysis and Human Genetic Disease
A. CFTR Protein-Coding Mutations Conservation of CFTR G551 Human Chimp Cow Pig Dog Mouse Rat Chicken Frog Salmon Dogfish
TLSGGQRAR TLSGGQRAR TLSGGQRAR TLSGGQRAR TLSGGQRAR TLSGGQRAR TLSGGQRAR ILSGGQRAR TLSGGQRAR TLSGGQRAR TLSGGQRAR
Function of CFTR G551
Cl TMD
*
NBD
* G551
B. CFTR Splice-Site Mutations 6211 G to T 6212 T to G 6213 A to G
6213 A to G
CFTR Intron 4
6211 G to T
AAG GTA
AAGAAG GTAATACTT... AAGAAG GTAATACTT... AAGAAG GTAATACTT... AAAAAG GTAATACTT... AAAAAG GTAATACTT... AAGAAG GTAATGCTT... AAGAAG GTAATACTT... AAGAAG GTAATACTT... AAGAAG GTAATACTT... AAGAAG GTAACTGTT... AAGAAG GTAAACAGA... AAGAAG GTAAGGGGA...
Wt CFTR mRNA
CFTR Exon 4 Human Chimp Cow Pig Sheep Dog Hedgehog Fruit Bat Mouse Chicken Zebrafish Tetraodon
Figure 10.2 Conserved sequences are important for protein function and mRNA processing. Mutations in coding and non-coding regions of the CFTR gene can cause cystic fibrosis. (A) The position of a coding mutation (G551D) is indicated in a multi-species protein-sequence alignment (left panel, arrow). The importance of this amino-acid residue in CFTR function is underscored by its conservation in species ranging from human to fish. The G551D change is shown relative to a schematic model of the CFTR protein within a plasma membrane (right panel). The G551 residue resides within a nucleotide-binding domain (NBD) that is critical for activation of the chloride channel, and the G551D mutation severely impairs this process [(Anderson and Welsh, 1992; Logan et al., 1994); note that the two green areas represent transmembrane domains (TMD)]. (B) The positions of three non-coding CFTR mutations are shown, with at least two (621+1 G to T and 621+3 A to G) resulting in improper splicing and deletion of exon 4 in the mature CFTR transcript (Hull et al., 1993; Tzetis et al., 2001). Inset: illustration of improper CFTR splicing caused by two mutations, as detected by RT-PCR. The importance of these nucleotides in CFTR mRNA processing is again underscored by the striking conservation of this noncoding sequence in species ranging from human to fish (indicated in red).
■
125
differences that correlate with the inheritance of a disease. Sorting out which of these, if any, is responsible for the disease can be challenging. CSA can establish the extent of evolutionary conservation at the genomic position of each disease-associated variant, with evidence of high conservation providing support that the variant is functionally relevant to the disease. Also note that this process can occur in a slightly reversed order, in which CSA is first used to identify conserved regions of interest, and these in turn are examined closely in patients to search for diseaseassociated sequence variants (see below). In this section, we provide illustrative examples where CSA has been used to implicate specific mutations in human genetic diseases. Conserved Sequences Important for Protein Function CSA is often used as a validation step to help implicate a protein-coding sequence variant with a genetic disease. Specifically, the presence of a disease-associated amino-acid variant at a residue that is highly conserved among widely diverged species is supportive evidence that the mutation plays a role in the disease. Cystic fibrosis (CF) is an autosomal recessive disease characterized by the production of abnormally viscous mucus, which frequently obstructs the pulmonary airways and leads to chronic, life-threatening respiratory infections. Loss-of-function mutations in the cystic fibrosis transmembrane conductance regulator gene (CFTR), which encodes a chloride channel, cause CF (Kerem et al., 1989; Riordan et al., 1989; Rommens et al., 1989). One of the most common missense mutations found in patients with CF is G551D (genet.sickkids.on.ca/cftr). Examination of the affected amino-acid residue by CSA reveals that G551, as well as its adjacent amino acids, are highly conserved among species ranging from human to fish (Figure 10.2A, left panel). This indicates that G551 is likely to be important for CFTR function. Indeed, G551 resides within the nucleotide-binding domain (NBD; Figure 10.2A, right panel), which is critical for proper function of the encoded chloride channel (Anderson and Welsh, 1992). Furthermore, the G551D mutation appears to cause abnormal channel activity (Logan et al., 1994). Conserved Sequences Important for mRNA Processing Identifying disease-associated mutations in protein-coding sequences is relatively straightforward. However, screening a gene or genomic region implicated in a disease does not always reveal mutations in coding sequence, making it necessary to search for mutations in the surrounding non-coding sequence. A common strategy is to then analyze conserved non-coding nucleotides – especially those at positions with a known or expected role in gene function. Proper splicing of heteronuclear RNA to form mature messenger RNA (mRNA) is essential for proper protein translation. The splicing process is carried out by a complex set of interactions between intron sequences and the nuclear spliceosome
126
CHAPTER 10
■
Comparative Sequence Analysis
A. CX32 Promoter Mutations
B. RET Enhancer Mutation Conserved Disease-Associated SNP in RET Intron 1
SOX10-Binding Sites
RET Intron 1 CX32 Exon 1 SNPs
CATTGTGCA CAGTGTGCA CATTCTGCA
Mutation Analysis 528 T to G 526 G to C
Experimental Analysis Wt Promoter
526 G to C
35% of Wt
528 T to G
50% of Wt Expression
RET3 C to T
Conserved Regions
MCS9.7
Sequence Conservation of RET+3 C Allele RET+3 C to T Human Chimp Baboon Cow Pig Cat Dog Rat Mouse
GGGGGCC-AGTGACCCTTACACGGTCATCCACAGGCCACTTGG GGGGGCC-AGTGACCCTTACACGGTCATCCACAGGCCACTTGG GGGGGCC-AGTGACCCTTACACGGTCATCCATAGGCCACTTGG GGGAGCCTGGTGACCTGCACACAGTCATCAGCAGGCCACTTGG GCGGACTTTGTGACCCTCACACGGTCATCATCAGGCCACTTGG GTGGGCCTGGTGACCCCCACACAGTCATCAGCAGGCCACTTGG GCGGGCCTGGTGACCCACACACAGTCATCAGCAGGCCACTTGG AGGGGCCTGGTCACTCACACGCACTCATCCACAGGCCACTTGG AGAGGTCTGGTCACTCACACACGCTCATCCCCAGGCCACTTGG Predicted Retinoic Acid Response Elements
Conservation Analysis 528 T to G 526 G to C Human Mouse Rat Rabbit Dog Opossum Fugu
CATTGTGCA CATTGTACA CATTGTACA TATTGTGCA CATTGTGCA TATTGTGTA CATTGTATG
Experimental Analysis of RET+3 Alleles Promoter Only MCS9.7 / RET3 C Allele 6.3-fold change MCS9.7 / RET3 T Allele 0
10 20 30 Expression
40
Figure 10.3 Conserved sequences are important for gene expression.Two examples of non-coding, disease-causing mutations in regulatory elements are schematically illustrated. (A) Mutations in the CX32 promoter (arrow indicates transcription start site) are associated with one form of Charcot-Marie-Tooth disease. Specifically, mutations in a SOX10-binding site (indicated in red) have been identified in patients with Charcot-Marie-Tooth disease. Experimental analysis of the wild-type (Wt) and mutant (526 G to C and 528 T to G) promoters, involving in vitro expression of a reporter gene, demonstrates mutation-associated decreases in promoter activity (based on Houlden et al. 2004). Conservation analysis reveals that the mutated nucleotides are conserved in vertebrates (highlighted in red). (B) Within intron 1 of the RET gene, a specific single-nucleotide polymorphism (SNP) – RET+3 C to T – is highly associated with Hirschsprung disease in a large subset of patients. That SNP resides within a conserved region – MCS+9.7 (grey box in upper panel). The RET+3 C allele is conserved among mammals (middle panel; highlighted in red), with that nucleotide position residing between two predicted transcription factor-binding sites involved in the response to retinoic acid (grey shading). Experimental analysis, which involved testing the MCS+9.7 segment for enhancer activity, revealed that the disease-associated RET+3 T allele shows greatly reduced expression compared to the wild-type RET+3 C allele. These studies demonstrate that a mutation in a RET-associated enhancer is responsible for Hirschsprung disease in some patients (based on Emison et al. 2005).
machinery. Specifically, three intron sequences are required for splicing to occur: (1) a donor site at the 5-end of the intron beginning with the sequence GT(A/G)ATG; (2) an acceptor site at the 3-end of the intron ending with the sequence CAG; and (3) an internal branch site with the sequence CT(A/G)A(C/T). Indeed, the identification of disease-causing mutations in these sequences underscores their importance in gene function. In a similar fashion to the analysis of amino-acid residues, the functional importance of nucleotides involved in mRNA processing can be established by CSA. In addition to missense mutations, CF can be caused by splice-site mutations in the CFTR gene. For example, three
mutations in the CFTR intron 4 donor splice site (621+1 G to T, 621+2 T to G, and 621+3 A to G) have been identified in patients with CF (genet.sickkids.on.ca/cftr), and these likely lead to the deletion of exon 4 in mature CFTR transcripts (Figure 10.2B, inset) (Hull et al., 1993; Tzetis et al., 2001). Comparing the genomic sequence of this region in highly diverged species reveals complete conservation of the 621+1 G, 621+2 T, and 621+3 A nucleotides (Figure 10.2B). Interestingly, while the donor-site consensus sequence GT(A/G)ATG appears to allow for an A to G nucleotide change at position 621+3, the G allele appears to cause CF and the A allele is conserved to fish (Figure 10.2B). Thus, CSA supports a pathogenic role for all three of
Comparative Sequence Analysis and Human Genetic Disease
these mutations and illustrates the importance of donor-site nucleotides for proper mRNA processing. Conserved Sequences Important for Promoter Function Transcriptional regulation is mainly controlled by trans-acting proteins (e.g., transcription factors) and cis-acting sequences (e.g., promoters, enhancers, and repressors). Sequence variants affecting either of these components can alter levels of transcription and lead to disease. Other non-coding sequences typically screened for diseaseassociated mutations are those corresponding to putative cis-acting transcriptional regulatory elements. This frequently begins by analyzing the region thought to harbor a gene’s promoter; specifically, the segment immediately upstream (roughly 500–2000 bases) of a gene’s known or predicted transcription start site (Figure 10.3A, arrow). Charcot-Marie-Tooth (CMT) disease consists of a heterogeneous group of inherited peripheral neuropathies that are generally characterized by muscle weakness and wasting in the distal extremities. One form of CMT, CMTX1, is associated with peripheral nerve demyelination and is caused by mutations in the connexin 32 gene (CX32) (Ionasescu et al., 1994). CX32 encodes a gap-junction protein expressed in Schwann cells that facilitates the radial transport of material through the concentric layers of the myelin sheath. One transcription factor involved in CX32 expression is SRY-related HMG-box 10 (SOX10), which binds to two sequences in the CX32 promoter (Bondurand et al., 2001). Interestingly, two mutations (526 G to C and 528 T to G) have been identified in these SOX10binding sites in patients with CMTX1 (Figure 10.3A, top) (Houlden et al., 2004). To establish the functional consequences of these disease-associated mutations, the promoter activities of fragments bearing each allele were tested. Specifically, CX32promoter fragments bearing the wild-type sequence, 526 G to C, or 528 T to G were cloned upstream of a luciferase reporter gene; in turn, each construct was transfected into cultured cells. There is a notable decrease in reporter-gene expression with 526 G to C and 528 T to G compared to the wild-type promoter sequence (Figure 10.3A, middle) (Houlden et al., 2004). Finally, CSA studies reveal that the wild-type nucleotides (526 G and 528 T) are conserved across a diverse array of vertebrates. Thus, CSA adds to a body of evidence supporting a pathogenic role for two CX32-promoter mutations in CMT disease. Conserved Sequences Important for the Function of Distal Regulatory Elements In the above examples, CSA was used to help implicate identified amino-acid and nucleotide variants with a human genetic disease. These cases all involved regions within genes that are relatively “obvious” to search for such mutations (i.e., coding, splice-site, and promoter sequences). A current obstacle in human genetics is finding disease-causing variants in other types of non-coding functional sequences, which are more difficult to
■
127
identify (i.e., less “obvious”). For example, distal transcriptional regulatory elements can reside within large introns or many kilobase pairs away from their associated gene. Yet, such sequences are believed to harbor mutations that cause human genetic disease. CSA can play a major role in identifying such elements in the first place, and these in turn can be screened for mutations (Figure 10.1, Step 4b). Hirschsprung disease is a genetically heterogeneous developmental disorder of the enteric nervous system, which is characterized by the absence of enteric ganglia in the distal colon. To date, eight loci have been implicated in the disease, with the majority of mutations identified residing in the RET gene, which encodes the ret proto-oncogene (Edery et al., 1994). However, mutations in these genes can only be identified in less than 30% of patients with Hirschsprung disease, indicating that there are other disease-associated genes or novel alleles of already identified genes. A common Hirschsprung disease-associated haplotype was identified involving RET intron 1, with no associated RET coding mutation (Carrasquillo et al., 2002). This haplotype contains multiple single-nucleotide polymorphisms (SNPs; red triangles in Figure 10.3B, top panel) that are associated with the disease, rendering it difficult to determine which one, if any, is pathogenic. CSA was used to identify highly conserved regions in and around RET intron 1 (Emison et al., 2005). One such region (MCS+9.7) overlapped one of the disease-associated SNPs (RET+3; Figure 10.3B, top panel). Importantly, the T allele of this SNP is over-transmitted to affected individuals, and the C allele is highly conserved among mammals (Figure 10.3B, middle panel). Analysis of the MCS+9.7 sequence for transcription factor-binding sites revealed two predicted, highly conserved retinoic acid-binding sites that flank the RET+3 SNP (Figure 10.3B, middle panel). Retinoic acid is known to regulate RET expression in certain tissues (Batourina et al., 2001; Shoba et al., 2002), suggesting that these sites are biologically relevant. Finally, experimental analysis of each RET+3 allele revealed enhancer activity with the C allele and significantly lower activity, with the T allele (Figure 10.3B, bottom panel), suggesting that this genomic region may function as an enhancer to regulate RET expression (Emison et al., 2005). In this example, CSA played a central role in identifying candidate functional non-coding sequences that were then screened for disease-associated variants. Without the evidence of sequence conservation, it would have been far more difficult to ultimately establish that this region functions as an enhancer or that one of the variants residing therein is disease-causing. The Utility of CSA in the Study of Complex Genetic Diseases A key promise of the Human Genome Project was to enhance the study of complex genetic diseases. This could involve using the generated human genomic sequence for identifying genetic variants in human populations. In turn, these variants can be studied in genome-wide association studies to identify risk alleles for
128
CHAPTER 10
■
Comparative Sequence Analysis
common genetic diseases (see Chapter 8), including, for example, type II diabetes (e.g., Scott et al., 2007 and Zeggini et al., 2007). A major challenge with such studies is that it typically is unclear whether a given identified variant is conferring disease risk, or if it is just closely linked to the true risk variant. CSA can be used to evaluate the degree of evolutionary conservation of a given region harboring identified variants. Those variants of particular interest (e.g., perhaps because they reside in a highly conserved region) can then be tested in experimental assays (see above for examples) to study their functional consequences. Another important consideration for interpreting the results of CSA in studying complex genetic diseases is the expected molecular pathology of the disease in question. If a loss of gene function is expected in disease pathogenesis, then CSA can be quite informative since conservation will often imply function. Multiple examples of this are provided above. Alternatively, if a gain of gene function is expected in disease pathogenesis (e.g., certain neurodegenerative diseases and some forms of cancer), then CSA may be less informative. On the one hand, a gainof-function mutation may mutate a highly conserved nucleotide or amino acid to generate higher levels of a gene product (e.g., an increase in affinity between a promoter and a transcription factor) or a gene product with a higher activity (e.g., an increase in affinity for a binding partner); here, CSA can provide information for implicating a disease-causing variant. On the other hand, a gain-of-function mutation may change a lessconserved nucleotide or amino acid to one that creates a novel or toxic function of the encoded protein; here, CSA would not help to support this variant as a probable disease-causing mutation.
CSA AND THE FUTURE OF HUMAN GENETICS AND GENOMIC MEDICINE The availability of sequences of the human genome and that of multiple other evolutionarily diverse species is greatly advancing the study of human genetic diseases. In this chapter, we have focused on how CSA can facilitate the identification and study of disease-causing mutations. The identification of genes defective in diseases that are inherited in a Mendelian fashion is now (almost) a routine process. Aided by CSA, investigators can now screen a more-complete catalog of conserved protein-coding and non-coding functional elements for mutations in such disease-associated genes. In a similar fashion, these same conserved sequences can be carefully scrutinized for the presence of mutations
causing more-common, but genetically complex, diseases (see section “The Utility of CSA in the Study of Complex Genetic Diseases” for a discussion of this issue). A related utility of CSA will be in more-completely characterizing the functional landscape of the human genome. Specifically, by establishing which sequences have been conserved over various evolutionary time frames (e.g., between human and mouse or between human and fish) and by linking those sequences to specific biological functions, an enhanced view of human genome function should emerge. Such a comprehensive examination of sequence conservation and sequence function is the primary goal of the ENCODE (ENCyclopedia Of DNA Elements) Project (www.genome.gov/ENCODE) (ENCODE Project Consortium, 2007). For this Project, a consortium of investigators is performing myriad computational (including CSA) and experimental studies (some similar to the ones described above) to catalog all functional sequences in the human genome, initially focusing on a targeted set of regions that comprise 1% of the genome. This last application of CSA has important implications for genomic medicine. It is predicted that the human genome harbors 3 to 5 million sites of SNPs that differ between any two individuals (Altshuler and Clark, 2005). While many of these SNPs have no functional consequence, a subset of them reside in functional regions of the genome and are functionally important. These latter SNPs may contribute to disease onset (as described above), modify the phenotypes caused by mutations at other loci (Nadeau, 2001), or modulate a patient’s response to medication (Goldstein et al., 2003). CSA can aid the identification of SNPs that are most likely to have functional consequences. For example, a subset of patients may experience adverse side effects when treated with a given drug. These side effects may be caused by an allele of a SNP within an enhancer that increases the expression of a gene in the pathway that metabolizes that drug. Through the use of CSA and other studies, this SNP can be identified and a test developed that determines which patients are most likely to experience the adverse side effects upon treatment with that drug (see Chapter 27). In summary, CSA continues to hold enormous potential as a tool for research in genome function, human genetics, and genomic medicine. One of the remaining challenges is to establish which species are most relevant for each biological question (Figure 10.1, Step 1). In the coming years, the continued collection and analysis of genomic sequences from multiple species should make this less of an issue, thereby enhancing the utility of CSA and its applications.
REFERENCES Altschul, S.F., Gish, W., Miller, W., Myers, E.W. and Lipman, D.J. (1990). Basic local alignment search tool. J Mol Biol 215, 403–410. Altshuler, D. and Clark, A.G. (2005). Harvesting medical information from the human family tree. Science 307, 1052–1053.
Anderson, M.P. and Welsh, M.J. (1992). Regulation by ATP and ADP of CFTR chloride channels that contain mutant nucleotide-binding domains. Science 257, 1701–1704. Antonellis, A., Bennett, W.R., Menheniott, T.R., Prasad, A.B., Lee-Lin, S.Q., Green, E.D., Paisley, D., Kelsh, R.N., Pavan, W.J.
References
and Ward, A. (2006). Deletion of long-range sequences at Sox10 compromises developmental expression in a mouse model of Waardenburg-Shah (WS4) syndrome. Hum Mol Genet 15, 259–271. Aparicio, S., Chapman, J., Stupka, E., Putnam, N., Chia, J.M., Dehal, P., Christoffels, A., Rash, S., Hoon, S., Smit, A. et al. (2002). Wholegenome shotgun assembly and analysis of the genome of Fugu rubripes. Science 297, 1301–1310. Bailey, T.L. and Elkan, C. (1994). Fitting a mixture model by expectation maximization to discover motifs in biopolymers. Proc Int Conf Intell Syst Mol Biol 2, 28–36. Batourina, E., Gim, S., Bello, N., Shy, M., Clagett-Dame, M., Srinivas, S., Costantini, F. and Mendelsohn, C. (2001). Vitamin A controls epithelial/mesenchymal interactions through Ret expression. Nat Genet 27, 74–78. Bondurand, N., Girard, M., Pingault, V., Lemort, N., Dubourg, O. and Goossens, M. (2001). Human Connexin 32, a gap junction protein altered in the X-linked form of Charcot-Marie-Tooth disease, is directly regulated by the transcription factor SOX10. Hum Mol Genet 10, 2783–2795. Carrasquillo, M.M., McCallion, A.S., Puffenberger, E.G., Kashuk, C. S., Nouri, N. and Chakravarti, A. (2002). Genome-wide association study and mouse model identify interaction between RET and EDNRB pathways in Hirschsprung disease. Nat Genet 32, 237–244. Chimpanzee Sequencing and Analysis Consortium (2005). Initial sequence of the chimpanzee genome and comparison with the human genome. Nature 437, 69–87. Collins, F.S., Green, E.D., Guttmacher, A.E. and Guyer, M.S. (2003). A vision for the future of genomics research. Nature 422, 835–847. Edery, P., Lyonnet, S., Mulligan, L.M., Pelet, A., Dow, E., Abel, L., Holder, S., Nihoul-Fekete, C., Ponder, B.A. and Munnich, A. (1994). Mutations of the RET proto-oncogene in Hirschsprung’s disease. Nature 367, 378–380. Emison, E.S., McCallion, A.S., Kashuk, C.S., Bush, R.T., Grice, E., Lin, S., Portnoy, M.E., Cutler, D.J., Green, E.D. and Chakravarti, A. (2005). A common sex-dependent mutation in a RET enhancer underlies Hirschsprung disease risk. Nature 434, 857–863. ENCODE Project Consortium (2007). Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447, 799–816. Frazer, K.A., Elnitski, L., Church, D.M., Dubchak, I. and Hardison, R.C. (2003). Cross-species sequence comparisons: A review of methods and available resources. Genome Res 13, 1–12. Frazer, K.A., Pachter, L., Poliakov, A., Rubin, E.M. and Dubchak, I. (2004). VISTA: Computational tools for comparative genomics. Nucleic Acids Res 32,W273–W279. Goldstein, D.B., Tate, S.K. and Sisodiya, S.M. (2003). Pharmacogenetics goes genomic. Nat. Rev. Genet. 4, 937–947. Houlden, H., Girard, M., Cockerell, C., Ingram, D., Wood, N.W., Goossens, M., Walker, R.W. and Reilly, M.M. (2004). Connexin 32 promoter P2 mutations: A mechanism of peripheral nerve dysfunction. Ann Neurol 56, 730–734. Hull, J., Shackleton, S. and Harris, A. (1993). Abnormal mRNA splicing resulting from three different mutations in the CFTR gene. Hum Mol Genet 2, 689–692. International Chicken Genome Sequencing Consortium (2004). Sequence and comparative analysis of the chicken genome provide unique perspectives on vertebrate evolution. Nature 432, 695–716. International Human Genome Sequencing Consortium (2001). Initial sequencing and analysis of the human genome. Nature 409, 860–921.
■
129
International Human Genome Sequencing Consortium (2004). Finishing the euchromatic sequence of the human genome. Nature 431, 931–945. Ionasescu,V., Searby, C. and Ionasescu, R. (1994). Point mutations of the connexin32 (GJB1) gene in X-linked dominant Charcot-MarieTooth neuropathy. Hum Mol Genet 3, 355–358. Kamal, M., Xie, X. and Lander, E.S. (2006). A large family of ancient repeat elements in the human genome is under strong selection. Proc Natl Acad Sci U S A 103, 2740–2745. Kent, W.J. (2002). BLAT--the BLAST-like alignment tool. Genome Res 12, 656–664. Kent, W.J., Sugnet, C.W., Furey, T.S., Roskin, K.M., Pringle, T.H., Zahler, A.M. and Haussler, D. (2002). The human genome browser at UCSC. Genome Res 12, 996–1006. Kerem, B., Rommens, J.M., Buchanan, J.A., Markiewicz, D., Cox, T.K., Chakravarti, A., Buchwald, M. and Tsui, L.C. (1989). Identifica-tion of the cystic fibrosis gene: Genetic analysis. Science 245, 1073–1080. Lindblad-Toh, K., Wade, C.M., Mikkelsen, T.S., Karlsson, E.K., Jaffe, D.B., Kamal, M., Clamp, M., Chang, J.L., Kulbokas, E.J., Zody, M.C. et al. (2005). Genome sequence, comparative analysis and haplotype structure of the domestic dog. Nature 438, 803–819. Logan, J., Hiestand, D., Daram, P., Huang, Z., Muccio, D.D., Hartman, J., Haley, B., Cook, W.J. and Sorscher, E.J. (1994). Cystic fibrosis transmembrane conductance regulator mutations that disrupt nucleotide binding. J Clin Invest 94, 228–236. Loots, G.G., Locksley, R.M., Blankespoor, C.M., Wang, Z.E., Miller, W., Rubin, E.M. and Frazer, K.A. (2000). Identification of a coordinate regulator of interleukins 4, 13, and 5 by cross-species sequence comparisons. Science 288, 136–140. Margulies, E.H., Blanchette, M., Haussler, D. and Green, E.D. (2003). Identification and characterization of multi-species conserved sequences. Genome Res 13, 2507–2518. Matys, V., Kel-Margoulis, O.V., Fricke, E., Liebich, I., Land, S., Barre-Dirrie, A., Reuter, I., Chekmenev, D., Krull, M., Hornischer, K. et al. (2006). TRANSFAC and its module TRANSCompel: Transcriptional gene regulation in eukaryotes. Nucleic Acids Res 34, D108–D110. Mouse Genome Sequencing Consortium (2002). Initial sequencing and comparative analysis of the mouse genome. Nature 420, 520–562. Nadeau, J.H. (2001). Modifier genes in mice and humans. Nat Rev Genet 2, 165–174. Nishihara, H., Smit, A.F. and Okada, N. (2006). Functional noncoding sequences derived from SINEs in the mammalian genome. Genome Res 16, 864–874. Plessy, C., Dickmeis, T., Chalmel, F. and Strahle, U. (2005). Enhancer sequence conservation between vertebrates is favoured in developmental regulator genes. Trends Genet 21, 207–210. Pollard, K.S., Salama, S.R., Lambert, N., Lambot, M.A., Coppens, S., Pedersen, J.S., Katzman, S., King, B., Onodera, C., Siepel, A. et al. (2006). An RNA gene expressed during cortical development evolved rapidly in humans. Nature 443, 149–150. Prabhakar, S., Poulin, F., Shoukry, M.,Afzal,V., Rubin, E.M., Couronne, O. and Pennacchio, L.A. (2006a). Close sequence comparisons are sufficient to identify human cis-regulatory elements. Genome Res 16, 855–862. Prabhakar, S., Noonan, J.P., Paabo, S. and Rubin, E.M. (2006b). Accelerated evolution of conserved noncoding sequences in humans. Science 314, 786. Rat Genome Sequencing Project Consortium (2004). Genome sequence of the Brown Norway rat yields insights into mammalian evolution. Nature 428, 493–521.
130
CHAPTER 10
■
Comparative Sequence Analysis
Riordan, J.R., Rommens, J.M., Kerem, B., Alon, N., Rozmahel, R., Grzelczak, Z., Zielenski, J., Lok, S., Plavsic, N., Chou, J.L. et al. (1989). Identification of the cystic fibrosis gene: Cloning and characterization of complementary DNA. Science 245, 1066–1073. Rommens, J.M., Iannuzzi, M.C., Kerem, B., Drumm, M.L., Melmer, G., Dean, M., Rozmahel, R., Cole, J.L., Kennedy, D., Hidaka, N. et al. (1989). Identification of the cystic fibrosis gene: Chromosome walking and jumping. Science 245, 1059–1065. Schwartz, S., Zhang, Z., Frazer, K.A., Smit, A., Riemer, C., Bouck, J., Gibbs, R., Hardison, R. and Miller, W. (2000). PipMaker – a web server for aligning two genomic DNA sequences. Genome Res 10, 577–586. Schwartz, S., Elnitski, L., Li, M., Weirauch, M., Riemer, C., Smit, A., Green, E.D., Hardison, R.C. and Miller, W. (2003). MultiPipMaker and supporting tools: Alignments and analysis of multiple genomic DNA sequences. Nucleic Acids Res 31, 3518–3524. Scott, L.J., Mohlke, K.L., Bonnycastle, L.L., Willer, C.J., Li, Y., Duren, W.L., Erdos, M.R., Stringham, H.M., Chines, P.S., Jackson, A.U., Prokunina-Olsson, L. et al. (2007). A genome-wide association study of type 2 diabetes in Finns detects multiple susceptibility variants. Science 316, 1341–1345.
Shoba, T., Dheen, S.T. and Tay, S.S. (2002). Retinoic acid influences Phox2 expression of cardiac ganglionic cells in the developing rat heart. Neurosci Lett 321, 41–44. Stalker, J., Gibbins, B., Meidl, P., Smith, J., Spooner, W., Hotz, H.R. and Cox, A.V. (2004). The Ensembl Web site: Mechanics of a genome browser. Genome Res 14, 951–955. Thompson, J.D., Higgins, D.G. and Gibson, T.J. (1994). CLUSTAL W: Improving the sensitivity of progressive multiple sequence alignment through sequence weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids Res 22, 4673–4680. Tzetis, M., Efthymiadou, A., Doudounakis, S. and Kanavakis, E. (2001). Qualitative and quantitative analysis of mRNA associated with four putative splicing mutations (621+3A→G, 2751+2T→A, 296+1G→C, 1717-9T→C-D565G) and one nonsense mutation (E822X) in the CFTR gene. Hum Genet 109, 592–601. Zeggini, E., Weedon, M.N., Lindgren, C.M., Frayling, T.M., Elliott, K.S., Lango, H., Timpson, N.J., Perry, J.R., Rayner, N.W., Freathy, R.M. et al. (2007). Replication of genome-wide association signals in UK samples reveals risk loci for type 2 diabetes. Science 316, 1336–1341.
RECOMMENDED RESOURCES Websites http://www.intlgenome.org Provides an overview of major genomesequencing projects as well as links to species-specific databases and other useful web resources. http://www.genome.gov/ENCODE Describes the rationale and specific components of the ENCODE (ENCyclopedia Of DNA Elements) Project. Also provides direct links to the generated data sets.
Articles Dermitzakis, E.T., Reymond, A. and Antonarakis, S.E. (2005). Conserved non-genic sequences – an unexpected feature of mammalian genomes. Nat Rev Genet 6, 151–157. This review gives an introduction to conserved non-coding sequences, and discusses their potential roles in genome biology.
Kleinjan, D.A. and van Heyningen, V. (2005). Long-range control of gene expression: Emerging mechanisms and disruption in disease. Am J Hum Genet 76, 8–32. This review discusses long-range transcriptional regulatory elements, and describes examples where disruption of such elements cause human disease. Miller, W., Makova, K.D., Nekrutenko, A. and Hardison, R.C. (2004). Comparative genomics. Annu Rev Genomics Hum Genet 5, 15–56. Nardone, J., Lee, D.U., Ansel, K.M. and Rao A. (2004). Bioinformatics for the “bench biologist”: How to find regulatory regions in genomic DNA. Nat Immunol 5, 768–774. These two reviews provide an excellent overview of comparative sequence analysis, and discuss a number of issues to consider when pursuing such studies.
CHAPTER
11 DNA Methylation Analysis: Providing New Insight into Human Disease Susan Cottrell, Theo deVos, Juergen Distler, Carolina Haefliger, Ralf Lesche, Achim Plum and Matthias Schuster
INTRODUCTION The human genome contains four bases – guanine, adenine, thymine, and cytosine; but the cytosines can be either methylated or unmethylated at the fifth carbon position in the pyrimidine ring (Figure 11.1a). In general, cytosines can only be methylated when they are in the context of a CpG dinucleotide, or in other words, a cytosine immediately followed by a guanine. These CpG dinucleotides are under-represented in the genome, occurring at only about 20% of the frequency expected assuming random distribution. In some regions, called CpG islands, the frequency of CpG dinucleotides is only 65% of the expected frequency (Takai and Jones, 2002). These stretches of 500–1000 base pairs typically overlap the promoter or exon 1 region of genes, and there are estimated to be 30,000 CpG islands in the human genome (Lander et al., 2001). CpGs in islands are predominantly unmethylated, while CpGs outside of islands are typically methylated. The methylation status of a CpG island is normally correlated with the chromatin structure and expression levels of nearby genes (Stirzaker et al., 2004). CpG islands associated with actively transcribed genes are typically unmethylated. When a CpG island Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
is methylated, methyl-CpG-binding domain (MBD) proteins recognize the methylated CpG and recruit the necessary factors for chromatin condensation and gene inactivation. This DNA methylation state is maintained during cell division by a family of enzymes called DNA methyltransferases (DNMTs). DNMT1 is thought to be responsible for maintaining methylation patterns through cell divisions, and DNMT3b is the primary enzyme involved in de novo methylation (Lopatina et al., 2002). Mutations in one of the DNMT genes, DNMT3B, are responsible for the rare genetic disorder ICF syndrome, which is characterized by severe immunodeficiency and DNA hypomethylation (Hansen et al., 1999). Mice lacking any of the DNMTs die either in utero or soon after birth (Okano et al., 1999). Although most CpG islands in the human genome are typically unmethylated in healthy cells, methylation of some islands plays a role in several normal physiological processes. In females, CpG islands on one of the X chromosomes become hypermethylated in the process of chromosomal inactivation (Hellman and Chess, 2007; Mohandas et al., 1981). On autosomes, methylation is also important during development for inactivation of maternally- or paternally-derived alleles of certain genes in a process called genomic imprinting. For instance, a region 2 kb upstream Copyright © 2009, Elsevier Inc. All rights reserved. 131
(a)
■
DNA Methylation Analysis
NH2
—
—
NH2
C
C
N C — O—
N Cytosine
C
N
C
C — O—
CH3
(c)
C
CH3 —
—
CHAPTER 11
genomic DNA
5-TCGCTGG-3 3-AGCGACC-5 —
132
CH3
C bisulfite treatment
N
5-methyl-cytosine
(b)
bisulfiteconverted DNA Methylation-specific restriction digestion
5-TCGUTGG-3
bis I strand
3-AGCGAUU-5
bis II strand
PCR
bisulfite-converted PCR amplified DNA Amplification
5-TCGTTGG-3 3-AGCAACC-5
“G-rich” strand “C-rich” strand
3-TCGCTAA-5 5-AGCGATT-3
Figure 11.1 Detection of methylated cytosines. (a) Cytosine and 5-methyl-cytosine. (b) Cleavage by some restriction enzymes is blocked by methylation of cytosines in the recognition site. After digestion, various methods can be used to detect the intact methylated DNA, including locus-specific PCR amplification. (c) Genomic DNA is shown at the top with methylated (red) and unmethylated (green) cytosines. The bisulfite treatment converts unmethylated cytosines into uracils. The resulting strands are no longer complementary, and each strand can be amplified by conversion-specific PCR primers.
of the H19 gene for a non-coding RNA is paternally methylated, leading to repressed IGF2 expression (Bartolomei et al., 1993). During the process of carcinogenesis, two important changes in the methylation pattern of the genome occur. First, sporadic CpGs (outside of CpG islands) become hypomethylated (Ehrlich et al., 1982). This global hypomethylation has been linked to genomic instability, possibly due to mobilization of repetitive elements (Chen et al., 1998; Feinberg, 2004). Meanwhile, some CpG islands, particularly near tumor-suppressor genes, can become hypermethylated during carcinogenesis (Herman and Baylin, 2003; Jones and Baylin, 2002). The hypermethylation is functionally similar to a deletion or mutation and can serve as one method of gene inactivation in Knudson’s classic two-hit hypothesis (Jones and Laird, 1999). Different combinations of aberrantly methylated genes are found in nearly all types of cancer (Costello et al., 2000).
TECHNOLOGY TO ASSESS DNA METHYLATION Cytosine methylation has minimal impact on both thermodynamics and kinetics of common DNA polymerases and
can, therefore, not be directly detected by any amplification or hybridization-based methods. Instead, there are three basic principles that can be employed to distinguish differentially methylated DNA sequences: (i) methylation-specific restriction enzymes (Figure 11.1b), (ii) methylation-specific antibodies and proteins, and (iii) methylation-specific bisulfite modification (Figure 11.1c). All existing methods for discovering and evaluating DNA methylation markers are based on one or more of these principles. Methylation-sensitive restriction enzymes, which typically digest unmethylated but not methylated DNA, have been in use for many years (Bird and Southern, 1978). The recognition sites contain one or more CpG sites. Sometimes cleavage is blocked completely by methylation of the CpG site, but more often the rate of cleavage is affected. Therefore, these methods must be optimized for complete and specific digestion. More recently, restriction enzymes have been introduced that specifically digest methylated DNA (Krueger et al., 1995). Methods based on restriction enzymes are limited to CpGs that are present within a restriction site. The second major category of DNA methylation methodology relies on naturally occurring proteins whose activity requires methylation detection. Methylation-specific antibodies
Technology to Assess DNA Methylation
have been generated that specifically bind to DNA containing methylated cytosines (Gebhard et al., 2006). In addition, methylbinding proteins, which play an integral part in bridging DNA methylation information with cellular function, have been successfully used to separate differentially methylated DNA fractions (Cross et al., 1994). Methylation analysis methods based on bisulfite conversion of sample DNA rely on the unique property of inorganic bisulfite to transform unmethylated cytosine to uracil (Figure 11.1) without affecting methylated cytosines (Frommer et al., 1992). Thus, the bisulfite reaction converts the epigenetic methylation information into sequence information, which can be analyzed by common molecular biology methods, most notably PCR. By converting most of the cytosines to uracils, bisulfite conversion of DNA has the effect of reducing the complexity of the genome, thereby making primer design and multiplexing more challenging. In addition, the chemical reaction can cause DNA damage if sub-optimal conditions are used (Grunau et al., 2001). Despite these caveats, bisulfite-based methods are extremely powerful for sensitive and quantitative analysis of DNA methylation at one or many CpG sites. Methods for Methylation Marker Discovery The human genome contains tens of thousands of CpG islands associated with individual genes and their activity. The number of individual methylation positions that could be aberrantly methylated in diseased states is several orders of magnitude higher. Several genome-wide discovery methods to identify differentially methylated regions associated with the specific diagnostic question have been developed. Many of these methods are based on digestion of sample DNA with a methylationsensitive restriction enzyme. The individual discovery techniques differ in their methods for detection of these restriction cuts, but they include techniques such as linker ligation combined with subtractive hybridization (MCA; Toyota et al., 1999), arbitrarily primed PCR (AP-PCR; Liang et al., 2002), array hybridization (Differential methylation hybridization, DMH; Huang et al., 1999), and two-dimensional gel electrophoresis (RLGS; Costello et al., 2002; Hatada et al., 1991). These methylation marker discovery methods use methylation-sensitive restriction enzymes to differentiate two classes of samples, such as a DNA from samples of responders to a drug and DNA from non-responders. For a thorough analysis, the entire marker discovery process often involves multiple comparisons. Furthermore, the throughput is low enough that only a few samples can reasonably be processed. Using pooled DNA from multiple patients reduces the likelihood of identifying polymorphic differences associated with individual patients, and it allows for the enrichment of methylation differences associated with disease, which is essential for identification of markers with high prevalence for screening applications. One approach that is becoming more commonly used is array-based discovery of aberrant methylation (Fukasawa et al., 2006; Hayashi et al., 2007;Taylor et al., 2007). In DMH, originally used for the identification of epigenetic alterations in breast and
■
133
ovarian cancer (Huang et al., 1999; Yan et al., 2001), genomic DNA is digested by a methylation-insensitive restriction enzyme and then linkers are ligated to the fragments. These products are then digested with a methylation-sensitive restriction enzyme and finally amplified by PCR via the linker sequence. Amplificates are only generated from methylated templates that are not digested by the methylation-sensitive restriction enzymes. The PCR products can be labeled with fluorescent dyes and hybridized to a CpG-island microarray. A recent paper describes the combination of a modified and optimized pre-chip DMH workflow and a custom-made oligonucleotide array representing approximately 50,000 regions of the human genome (Lewin et al., 2007). The advantages of DMH are its reproducibility, speed, and throughput (allowing the option of analyzing individual samples instead of pools). The readout of the microarray not only identifies the differentially methylated sequence, but also provides semi-quantitative methylation values. Recently, the Human Epigenome Project, a large-scale effort to identify, catalog, and interpret genome-wide DNA methylation patterns of all human genes in all major tissues, has been initiated by The Wellcome Trust Sanger Institute (Cambridge, UK), the Centre National de Génotypage (Paris, France), and Epigenomics AG (Berlin, Germany) (Table 11.1). Based on public versions of the sequence of the human genome established by the Human Genome Project, the Human Epigenome Project will systematically uncover the epigenetic information layer that is still hidden to a large extent. A second Human Epigenome Project was recently proposed by an international group of 40 scientists (Esteller, 2006; Jones and Martienssen, 2005) (Table 11.1). Methods for Analyzing Markers in Tissue Samples Methylation analysis in tissues is often not limited by the quantity of sample material available, but the quality of the material, particularly formalin-fixed and paraffin-embedded samples, can be an issue (Schuster, 2004). The isolated DNA is usually damaged to some extent depending on the fixation procedure. In addition, the proportion of the tissue type with aberrant methylation (i.e., the percent tumor) can vary from sample to sample. Laser capture microdissection (LCM), which is used to isolate the tissue type of interest, helps overcome this issue. There are very different levels at which DNA methylation can be assessed (Figure 11.2) (Laird, 2003), ranging from genomewide to a particular locus or CpG. The “methylation content” of a cell population corresponds to the overall proportion of methylated cytosines within the entire genome. In contrast, the “methylation level” designates the percentage of methylated DNA strands at one genomic locus, while the “methylation pattern” describes the individual methylation status of a specific set of CpG positions on one DNA strand within a region of interest. Tissue testing is best performed using marker information contained in the “methylation level”. In principle, the methylation level of a DNA sample can be measured in a straightforward manner after bisulfite treatment because the methylation ratio at the CpG site of interest is
134
CHAPTER 11
TABLE 11.1
■
DNA Methylation Analysis
Multicenter efforts for large-scale epigenetic analysis Goals
Human Epigenome Project
●
●
International Alliance for the Human Epigenome and Disease (AHEAD)
●
Encyclopedia of DNA Elements (ENCODE)
●
●
Approach
Institutions
References
“To identify, catalogue and interpret genomewide DNA methylation patterns of all human genes in all major tissues” Identify “Methylation Variable Positions”
Bisulfite DNA sequencing
Wellcome Trust Sanger Institute; Epigenomics, AG; The Centre National de Genotypage
Eckhardt et al., 2006 http://www.epigenome. org/
A comprehensive catalog of potential chromatin alterations Describe “reference epigenomes”
To be determined; likely to include a variety of methods to assess DNA methylation and histone modifications
To be determined
Jones and Martienssen 2005
Identify all functional elements in the human genome
Chip-on-chip assays for DNA methylation and chromatin modifications; comparative genomics; DNase I hypersensitivity site mapping
Welcome Trust Sanger Institute; Stanford University; University of Virginia; Affymetrix; Municipal Institute of Medical Research; Ludwig Institute for Cancer Research; Yale University; University of Washington; others
ENCODE project consortium, 2007 http://www.genome. gov/10005107
(a) Methylation content
(b) Methylation level
(c) Methylation pattern
Figure 11.2 Methods for assessment of DNA methylation. The “methylation content” of a cell population corresponds to the overall proportion of methylated cytosines within the entire genome (a), the “methylation level” designates the percentage of methylated DNA strands at a one genomic locus (b), and the “methylation pattern” describes the individual methylation status of a specific set of CpG positions on one DNA strand within a region of interest (c).Reproduced with permission of Macmillan Magazines Ltd from Nature Reviews Cancer 3(4), April 2003, 253–266. Copyright (2003) Macmillan Magazines Ltd.
translated into an allele ratio. However, in most cases not enough DNA will be available for direct analysis. The bisulfite-treated DNA can be amplified without disturbing this allele ratio using conventional PCR with the primers designed to avoid CpG
sites. After amplification, the product can be analyzed using any of a number of analysis methods typically used for the quantification of allele ratios. In MS-SNuPE (Methylation-Specific Single Nucleotide Primer Extension), extension primers
Technology to Assess DNA Methylation
designed to hybridize just upstream of a CpG site are extended by one nucleotide in a methylation-specific manner. Detection and quantification based on HPLC (El-Maarri et al., 2002) and incorporation of radioactivity have been described (Gonzalgo et al., 1997). Another example is pyrosequencing, in which the sequential incorporation of additional bases into the extension primer is translated into a light flash of the same intensity (Colella et al., 2003). Alternatively, hybridization of methylationspecific probes to amplified bisulfite-treated DNA has been combined with different readout systems, including microarrays (Adorjan et al., 2003) and real-time PCR (Quantitative MethyLight; Eads et al., 2000). All of the methods mentioned measure relative levels of methylation, but can be used to quantify DNA methylation levels when they are appropriately calibrated on mixtures of methylated and unmethylated DNA templates. The methods described in the previous paragraph use an unbiased amplification of bisulfite-treated DNA for measuring methylation levels. Methylation in tissues can also be estimated using various forms of methylation-specific amplification. The most commonly used method in methylation research is methylation-specific PCR (MSP) or the real-time version, MethyLight (Eads et al., 2000). These methods use primers designed for a certain methylation pattern of bisulfite-converted DNA. MethyLight technology is a combination of methylation-specific primers and a methylation-specific real-time PCR probe (Figure 11.3). Since these methods only amplify and detect the methylated portion of the DNA sample, they can only provide qualitative methylation pattern information. In combination with an estimation of total DNA amount, which can be measured for instance by a control real-time PCR assay, methylation-specific assays can be used to estimate relative methylation pattern levels (Ogino et al., 2006a). For tissue-based analyses, a quantitative methylation assay is likely to be important, because in many cases the required information will depend not on the presence or absence of methylation but on the level of methylation. Responders to a certain drug, for instance, might have fewer inactivated copies of a gene than non-responders. In some instances, the methylation difference could be quite subtle, and it is not clear which of the methods will provide the level of precision required. Methods for Analyzing Markers in Disease Screening Samples The samples of choice for early detection or screening are body fluids that can be obtained by non-invasive procedures. Serum and plasma are optimal targets for methylation marker analyses, but investigators have also analyzed methylation in urine (Hoque et al., 2005), sputum, bronchial lavage, ductal lavage, ejaculate, and other “remote samples” (reviewed in Laird, 2003). In healthy individuals, only small amounts of circulating DNA are observed in plasma and serum, whereas high amounts have been described in patients with various conditions. The main sources of circulating DNA are thought to be apoptotic or necrotic cell death (Holdenrieder et al., 2001). DNA is released from degrading cells after nuclease cleavage of the chromatin between the nucleosomes. Therefore, the tumor- or disease-derived DNA fragments
(a)
■
135
light F
R
+ ++ + + + + + +
(b)
light
(c) F
R
+++ + ++ +++ ++++ +
(d)
Figure 11.3 Sensitive real-time PCR assays for methylation detection MethyLight assays (a, b) have two methylationspecific primers that bind only to the methylated (a) but not the unmethylated (b) versions of their binding sites. If both primer binding sites are methylated, PCR products are formed and will be detected by methylation-specific real-time PCR probes (a). In contrast, HM assays (c, d) use primers that amplify all methylation states. The methylation specificity of the priming is provided by the blocker oligonucleotide, which prevents the primer from binding to unmethylated DNA (d). Therefore, only methylated DNA is amplified and will be detected by a methylation-specific fluorescent probe (c). Filled circles indicate methylated CpGs, and open circles indicate unmethylated CpGs.
can be as short as the smallest nucleosomal fragment, approximately 180 base pairs (Cottrell and Laird, 2003; Jahr et al., 2001). In contrast to tissue-based applications, methylation markers for screening must fulfill additional biological requirements. Even if the majority of free DNA in blood is derived from cancer cells, a significant amount of DNA from other cell types, particularly leukocytes, will be present. Therefore, the marker must be methylated in a high percentage of tumor cells and relatively unmethylated in leukocytes and other tissues contributing background DNA. Qualitative markers, which are methylated only in tumor-derived DNA, are well suited for screening applications. The clinical sensitivity (proportion of diseased people testing positive) and specificity (proportion of healthy individuals testing negative) of the marker is not only dependent on its biological properties, but also on the analytical performance of the assay. The challenge for the final diagnostic assay is a high analytical sensitivity (detecting small amounts of methylated tumor DNA) at a maximum specificity (no detection of unmethylated DNA). In practice, methylation assays suitable for screening must be capable of detecting a few copies of the methylated marker DNA in a background of 1–10 thousand-fold excess unmethylated DNA. In addition, the DNA extraction method must be optimized for high recovery yield, particularly of short DNA fragments found in plasma or serum. Bisulfite treatment is known to cause damage to the DNA, but milder treatment
136
CHAPTER 11
■
DNA Methylation Analysis
conditions can be used without risk of incomplete conversion of the DNA (Berlin et al., 2005). MethyLight may be an assay that is suitable for detection of disease in remote samples. In contrast to MSP, MethyLight minimize false-positive results caused by possible mis-priming of the methylation-specific primer, as the methylation status of the amplificates are analyzed on a second level by the methylation-specific fluorescent probe. MethyLight has been successfully used for various applications, but primers and probes must be carefully designed in order to ensure sensitive and specific detection. HeavyMethyl (HM) technology is another way to achieve methylation-specific priming, and, like MSP, can be combined with a probe for real-time detection (Cottrell et al., 2004). HM assays combine methylation-non-specific primers and a nonextendable 3-modified blocker oligonucleotide. The blocker binds to unmethylated DNA, preventing the primers from accessing the priming site (Figure 11.3). The blocker does not bind to the methylated version, allowing the primers to access the priming sites. A methylation-specific fluorescent probe is used to ensure specificity and allow for methylation quantification. HM assays typically have a limit of detection of 25–50 pg of methylated DNA and can detect this methylated DNA in a background of 50 ng of unmethylated DNA. Although HM design requires more effort compared to a MethyLight assay, the HM assay has advantages. Unmethylated template that is amplified unintentionally will be targeted by the blocker in the next cycle. Therefore, the analytical specificity of HM assays is very high.
CLINICAL IMPACT OF DNA METHYLATION ANALYSIS Cancer has been viewed as an accumulation of chromosomal aberrations and therefore been called a “genetic disease”. However, it has become clear over the last two decades that epigenetic changes play a crucial role in carcinogenesis. Research on DNA methylation in cancer has been expanding rapidly, and for this reason, in the following sections, we will rely on examples from oncology to demonstrate how methylation profiles can be applied for basic disease research, early detection, disease stratification, treatment response prediction, and drug target identification. While attention has focused on methylation in carcinogenesis, a similar groundswell of research is emerging on methylation in other diseases, especially autoimmune and cardiovascular conditions (see Table 11.2). For example, aberrantly hypomethylated DNA in circulation in systemic lupus erythematosus might induce an immune response due to its similarity to unmethylated microbial DNAs (Januchowski et al., 2004). Further contributions of defective methylation to lupus pathogenesis could be due to transcription of endogenous retroviruses (Okada et al., 2002) or the increase in the expression of certain genes related to autoreactivity (Richardson, 2002). In cardiovascular disease, DNA methylation is associated with atherosclerosis pathogenesis in two ways: First, the decrease in essential factors (e.g., folate and vitamin D)
for the synthesis of S-adenosyl methionine (the main methyl group donor in the methylation reactions) leads to global DNA hypomethylation and coexisting hyperhomocystinemia (Zaina et al., 2005). Second, Lund et al. (2004) provided evidence that methylation profiles are early markers of the disease in a mouse model. Taken together, these data reinforce the notion that methylation profiles can be highly valuable to study the pathogenesis of a variety of conditions (Table 11.2). Impact of Methylation Research on Basic Disease Understanding Researchers have spent many years studying the role of permanent alterations in DNA sequence, such as mutations, deletions, and insertions, in carcinogenesis. Cancer has long been thought of as a disease that arises after accumulation of mutational events in growth control genes. In recent years, RNA expression profiling has been used to examine the functional consequences of these sequence alterations, as outlined in other chapters in this book. There is now overwhelming evidence that epigenetic modification can also control RNA and protein levels. As the methylation analysis toolbox expands, we are gaining more insight into how alterations in these patterns of methylation influence disease. As an example, methylation analysis has provided a new molecular understanding of colon cancer. Colon cancer develops in a well-described pathway from normal epithelium, to aberrant crypts, to adenomas, then to adenocarcinoma, each with specific histological features and molecular alterations. The first alterations to be studied were sequences changes: for instance, APC mutations have been implicated in the formation of adenomas (Powell et al., 1992) and TP53 mutations have been implicated in the transition from adenoma to adenocarcinoma (Baker et al., 1990). More recently, the role of aberrant methylation events of genes in these same pathways has been elucidated. For example, a hereditary form of colon cancer is commonly caused by mutations in the MLH1 gene, but this same gene can induce sporadic forms of the disease when it is aberrantly methylated (Cunningham et al., 1998). Some colon cancers are characterized by microsatellite instability (MSI), and the majority of these tumors have methylated MLH1 alleles. Furthermore, research on methylation in colon cancer has established that tumors can not only be divided based on microsatellite stability status, but also on their CpG island methylator phenotype (CIMP) status (Toyota et al., 1999). Colon tumors, particularly those with MSI, have a bimodal distribution based on the methylation status of multiple genes. Those tumors with multiple methylated genes (CIMP+) are more likely to have BRAF mutations and wildtype KRAS (Ogino et al., 2006a, b). Clearly, a complete understanding of the pathogenesis of colon cancer requires knowledge of both genetic and epigenetic events. Methylation Markers for Early Detection and Diagnosis Early detection is an effective success factor in the fight against any type of cancer. Aberrant DNA methylation patterns show
TABLE 11.2
Examples of DNA methylation in human disease
Disease/Disorder
Phenotype
Mechanism
Genes involved
Reference
Cancer
Tumorigenesis; Increased cell division; metastasis
Hypermethylation
Tumor-suppressor genes
Hypomethylation Loss of imprinting
Global Insulin-like grow factor 2 (IGF2), GTP-binding RAS-like 3 (DIRAS3), mesoderm-specific transcript homolog (MEST) Mut L Homolog 1 (MLH1)
Jones and Baylin, 2002; Herman and Baylin, 2003 Ehrlich et al., 1982 Feinberg, 2007
Heritable germline hypermethylation
Hitchins et al., 2007
Embryonal tumors; omphalocele; macroglossia (large tongue); gigantism
Disruption of an imprinting locus
11p15 (IGF2, H19 ncRNA, cyclindependent kinase inhibitor 1C (CDKN1C), pleckstrin homologylike domain, family A, member 2 (PHLDA2), solute carrier family 22, member 1 (SLC22A1), potassium voltage-gated channel (KCNQ1))
Weksberg et al., 2003
Prader–Willi syndrome
Developmental delay; obesity; genital hypoplasia; distinct facial features
Disruption of an imprinting locus
15q11-13 (small nuclear ribonucleoprotein polypeptide N (SNRPN), ncRNAs)
Nicholls and Knepper, 2001
Angelman syndrome
Developmental delay; ataxia; microcephaly; distinct facial features
Disruption of an imprinting locus
15q11-13 (Ubiquitin E3 ligase (UBE3A))
Nicholls and Knepper, 2001
Pseudohypoparathyroidism
Hypoparathyroidism without sensitivity to parathyroid hormone
Disruption of an imprinting locus
20q13 (GNAS complex locus (GNAS1))
Bastepe et al., 2001
Immunodeficiency/centromeric instability/facial anomalies (ICF syndrome)
Immunodeficiency; chromosomal breakage; facial defects
Mutations
DNA Methyltransferase 3b (DNMT3b)
Hansen et al., 1999
Rett syndrome
Motor abnormalities; ataxia; handwringing; poor verbal skills
Mutations
Methyl-CpG-binding protein 2 (MeCP2)
Amir et al., 1999
Systemic Lupus Erythematosus
Autoimmune disorder; inflammation; tissue damage
Hypomethylation
global, perforin (PRF1), CD70, integrin, alpha L (ITGAL)
Januchowski et al., 2004
Atherosclerosis
Thickening and hardening of arterial walls
Hypermethylation
Estrogen receptor, Monocarboxylate Transporter 3 (SLC16A3)
Kim et al., 2007; Zhu et al., 2005
Schizophrenia
Hallucinations; delusions
Hypomethylation
Global
Shimabukuro et al., 2006
Hypermethylation
Reelin (RELN)
Grayson et al., 2005
Hypermethylation
Homeobox A10 (HOXA10)
Wu et al., 2005
Fragile X syndrome
Mental retardation; autism; large ears; macroorchidism
Hypermethylation of an expanded trinucleotide repeat
Fragile X mental retardation 1 (FMR1)
Oberle et al., 1991
137
Growth of endometrial cells outside of the uterus
■
Endometriosis
Clinical Impact of DNA Methylation Analysis
Beckwith–Wiedemann syndrome
138
CHAPTER 11
■
DNA Methylation Analysis
promise as biomarkers for early detection of cancer for several reasons: Tumor-specific methylation changes occur early in tumorigenesis, appear to be stable, provide an amplifiable signal, and can be assayed with high analytical sensitivity. DNA methylation patterns characteristic of tumor cells have been found for many types of cancer in several types of body fluids (Laird, 2003). DNA methylation changes occur frequently in colon cancer and can be present early in malignant transformation (Grady, 2005). Feasibility studies have shown that tumor-derived methylated DNA markers can be detected in serum, plasma, and stool (reviewed in Laird, 2003). These studies analyzed a limited number of patient samples, and larger clinical studies will be necessary to assess the quality of these markers for a screening approach. Prostate-specific antigen (PSA) based screening has resulted in detection of many cases of prostate cancer, but, due to the low specificity of this assay, its usefulness for prostate cancer screening is no longer certain. DNA methylation markers may improve screening and diagnosis of prostate cancer considerably in the future. GSTP1, encoding glutathione S-transferase P1, is the bestcharacterized methylation marker for prostate cancer. Methylation of GSTP1 in prostate cancer has been consistently reported with a frequency of about 70–90% (reviewed in Henrique and Jeronimo, 2004). Interestingly, GSTP1 methylation is also reported in some high-grade prostatic intraepithelial neoplasia (PIN) lesions, a precursor to prostate cancer, indicating that GSTP1 hypermethylation is an early event in prostate cancer development (Henrique et al., 2006). Several groups have demonstrated the feasibility of using GSTP1 methylation and other methylation markers to detect prostate cancer in bodily fluids. For example, Hoque et al. (2005) analyzed a panel of nine genes in urine sediments from 52 cancer patients and 91 controls using MethyLight assays. A panel containing GSTP1 and three other genes (CDKN2A, ARF, and MGMT) detected the prostate cancer cases with 87% sensitivity and 100% specificity. Recently, Lofton-Day et al. (2006) conducted a study to identify and validate DNA methylation-based markers for colorectal cancer using plasma samples. At a set specificity of 95%, the lead methylation marker (Septin 9) showed in several studies sensitivity values of 51, 65, and 50%, respectively. In total, approximately 2000 samples were tested, including over 600 plasma samples from colorectal cancer patients, 600 with related non-malignant diseases, and over 600 normal controls from an age-matched, colonoscopy-verified healthy population. Importantly, early-stage cancers were identified with the same sensitivity as later stage cancers (Lofton-Day et al., 2006). The test was able to detect colorectal cancers regardless of their location, addressing a critical medical need and shortfall of the existing fecal occult blood tests. Once a methylation-based screening test is developed, its use can potentially be extended to monitoring for disease recurrence after treatment, similar to the role of PSA for prostate cancer.
aggressiveness of tumors and the responsiveness to certain drug regimens, respectively. For example, prognostic markers for breast cancer are required to determine which patients can be spared from aggressive therapy due to very good prognosis. Recently, several reports have been published exploring the prognostic relevance of DNA methylation at the promoter of the PLAU gene, which codes for the matrix remodeling enzyme urokinase-type plasminogen activator (uPA) (Guo et al., 2002; Pakneshan et al., 2004; Xing and Rabbani, 1999). Interestingly, PLAU-promoter DNA methylation was negatively associated with tumor grade; grade 1 tumors were frequently methylated and grade 3 tumors were unmethylated. These findings are consistent with results from large studies that have demonstrated that protein expression in serum of uPA and its cell surface receptor uPAR are important determinants of distant spread of breast cancer (Look et al., 2002). Shinozaki et al. (2005) found that in a panel of 151 primary breast tumors, hypermethylation of the CDH1 gene was significantly associated with lymphovascular invasion, infiltrating ductal histology, and lack of estrogen receptor expression. On the other hand, RASSF1A and RARB hypermethylation were significantly more common in estrogen receptor-positive and human epidermal growth factor receptor 2-positive tumors, respectively. Other genes that have been associated with prognosis in breast cancer and which appear to be regulated via DNA methylation are CDH3 (Paredes et al., 2005), the tuberous sclerosis (TSC) genes TSC1 and TSC2 (Jiang et al., 2005), and LATS1/LATS2, tumor-suppressor genes that have been implicated in the regulation of the cell cycle (Takahashi et al., 2005). We have recently identified a methylation marker, PITX2, which predicts outcome in hormone receptor-positive, node-negative breast cancer patients treated with tamoxifen monotherapy (Maier et al., 2004, 2007). The marker was originally identified and validated in two independent cohorts of patients treated with adjuvant tamoxifen monotherapy. The findings were then validated in an independent study analyzing paraffin-embedded tumors of 422 node-negative patients treated with tamoxifen only after surgery. In the group with low PITX2 methylation, 94% of the patients were metastasis-free after 10 years, compared to only 84% in the group with high PITX2 methylation. In a multivariate model, PITX2 methylation added significant information to conventional factors such as tumor size, grade, and age (Harbeck et al., 2005). Analysis of PITX2 methylation in a large cohort of patients not receiving adjuvant therapy confirmed that PITX2 methylation is a prognostic marker associated with tumor aggressiveness (Martens et al., 2005). These results provide substantial evidence that PITX2 DNA methylation is suitable for routine clinical use in order to predict outcome in node-negative, tamoxifen-treated patients, and to identify low-risk patients who can be spared the burden of additional cytotoxic therapy.
Methylation Markers for Disease Prognosis Molecular classification holds great promise for assessment of optimal treatment options. DNA methylation markers have been reported as both prognostic and predictive markers, indicating the
Methylation Markers for Treatment Response Prediction A number of studies have provided evidence that specific methylation changes are associated with responses to a variety of cancer
References
therapeutic agents currently used in the clinics (reviewed in Maier et al., 2005). These alterations in methylation patterns could serve as predictive markers of drug response and be used by clinicians and patients to support treatment selection. Furthermore, methylation markers could be developed together with a novel drug during clinical trials in order to better select the patients who are likely to respond. The inclusion of predictive markers in clinical trials is becoming increasingly important as more targeted therapies are being tested. Many anti-cancer therapies disturb DNA integrity and thus interfere with DNA synthesis and successful cellular replication, ultimately inducing cell death. Tumor response to DNA damaging agents is closely linked to expression of several DNA-repair enzymes, many of which are regulated by methylation. Resistance can result from either a gain of methylation event or loss of methylation event. Methylation of MGMT, which codes for a DNArepair enzyme, is associated with response to alkylating agents. In this case, tumor cells rely on the excess MGMT product to specifically repair the damage that occurred during therapy with the alkylating agent (Ludlum, 1990). Esteller et al. (2000) investigated the methylation status of the MGMT gene in 47 glioma samples from patients treated with the chloroethylating agent carmustine. The MGMT promoter region was methylated in 19 (40%) of the samples, 12 (63%) of which were from patients who had a response to carmustine. Of the 28 tumors with unmethylated MGMT, only 1 patient (4%) responded to the drug (p 0.001, univariate analysis). Paz and coworkers (2004) found similar results in glioma patients. Balana and colleagues (2003) demonstrated that detection of these treatment response markers might be possible in serum for disease where affected tissue is unavailable. MLH1 is a mismatch repair enzyme that is activated in response to DNA damage. However, MLH1 expression is not only required to repair the damage, but also seems to be linked to apoptotic signaling: Its activation induces processes leading to programmed cell death. Therefore, in contrast to MGMT, methylation of the MLH1 promoter is associated with resistance to DNA damaging agents, such as temozolomide, dacarbazine, and
■
139
cisplatin. Expression of MLH1 increases the effect of the drugs, presumably by acting as a sensor detecting DNA damage caused by the drug, and by activating processes that eventually lead to apoptosis of the cell (Agarwal and Kaye, 2003; Brown et al., 1997; Karran and Bignami, 1994). Other drug classes for which methylation markers were shown to be associated with tumor response in patients or sensitivity of preclinical models are taxanes (CHFR, TRAG3, RASSF1, BRCA1), platinum compounds (ABCB1, FANCF, RASSF1), retinoids (RARB, RBP1), and anti-hormonal therapies (ESR1, ESR2, AR) (Maier et al., 2005).
CONCLUSIONS With methylation-specific restriction enzymes, methylationbinding proteins, and bisulfite conversion, researchers have many technical options for detecting and measuring methylation information in the human genome. While MSP and bisulfite sequencing have been the workhorses in basic research on methylation, new developments in real-time PCR and arraybased analysis are expanding the repertoire. Techniques such as MethyLight have the level of analytical performance required for clinical application, allowing basic research on aberrant methylation in disease to be translated into sensitive and specific diagnostic tests that will be extremely valuable to clinicians and patients. The growing knowledge of aberrant methylation has already been integrated into models of carcinogenesis, and a complete understanding of disease initiation and progression must incorporate epigenetic analysis as well as traditional genetic analysis. With the development and validation of biomarkers based on aberrant methylation events, the emergence of methylation markers for clinical oncology is anticipated in short order. The wealth of information provided by methylation analysis in oncology and the fundamental importance of methylation in gene expression are certain to drive the expansion of methylation analysis into autoimmune, cardiovascular, and other diseases as well.
REFERENCES Adorjan, P., Distler, J., Lipscher, E., Model, F., Muller, J., Pelet, C., Braun, A., Florl, A., Gutig, D., Grabs, G. et al. (2003). Tumor class prediction and discovery by microarray-based DNA methylation analysis. Nucleic Acids Res 30, e21. Agarwal, R. and Kaye, S.B. (2003). Ovarian cancer: Strategies for overcoming resistance to chemotherapy. Nat Rev Cancer 7, 502–516. Amir, R.E.,Van den Veyver, I.B., Wan, M., Tran, C.Q., Francke, U. and Zoqhbi, H.Y. (1999). Rett syndrome is caused by mutations in X-linked MECP2, encoding methyl-CpG-binding protein 2. Nat Genet 23, 185–188. Baker, S.J., Preisinger, A.C., Jessup, J.M., Paraskeva, C., Markowitz, S., Willson, J.K., Hamilton, S. and Vogelstein, B. (1990). p53 gene mutations occur in combination with 17p allelic deletions as late events in colorectal tumorigenesis. Cancer Res 50, 7717–7722.
Balana, C., Ramirez, J., Taron, M., Roussos, Y., Ariza, A., Ballester, R., Sarries, C., Mendez, P., Sanchez, J. and Rosell, R. (2003). O6-methyl-guanine-DNA methyltransferase methylation in serum and tumor DNA predicts response to 1,3-bis(2-chloroethyl)-1nitrosourea but not to temozolamide plus cisplatin in glioblastoma multiforme. Clin Cancer Res 9, 1461–1468. Bartolomei, M.S., Webber, A.L., Brunkow, M.E. and Tilghman, S.M. (1993). Epigenetic mechanisms underlying the imprinting of the mouse H19 gene. Genes Dev 7, 1663–1673. Bastepe, M., Lane, A.H. and Jueppner, H. (2001). Paternal uniparental isodisomy of chromosome 20q – and the resulting changes in GNAS1 methylation – as a plausible cause of pseudohypoparathyroidism. Am J Hum Genet 68, 1283–1289. Berlin, K., Cardon, K. and Ballhause, M. (2005). Improved bisulfite conversion of DNA. Patent WO 2005/038051.
140
CHAPTER 11
■
DNA Methylation Analysis
Bird, A.P. and Southern, E.M. (1978). Use of restriction enzymes to study eukaryotic DNA methlation: I, The methylation pattern in ribosomal DNA from Xenopus laevis. J Mol Biol 118, 27–47. Brown, R., Hirst, G., Gallagher, W., McIlwrath, A., Margison, G., van der Zee, A. and Anthoney, D. (1997). hMLH1 expression and cellular responses of ovarian tumour cells to treatment with cytotoxic anticancer agents. Oncogene 15, 45–52. Chen, R., Pettersson, U., Beard, C., Jackson-Grusby, L. and Jaenisch, R. (1998). DNA hypomethylation leads to elevated mutation rates. Nature 395, 89–93. Colella, S., Shen, L., Baggerly, K.A., Issa, J.-P.J. and Krahe, R. (2003). Sensitive and quantitative universal Pyrosequencing methylation analysis of CpG sites. BioTechniques 35, 146–150. Costello, J.F., Fruhwald, M.C., Smiraglia, D.J., Rush, L.J., Robertson, G.P., Gao, X.,Wright, F.A., Feramisco, J.D., Peltomaki, P., Lang, J.C. et al. (2000). Aberrant CpG-island methylation has nonrandom and tumour-type-specific patterns. Nat Genet 24, 132–138. Costello, J.F., Smiraglia, D.J. and Plass, C. (2002). Restriction landmark genome scanning. Methods 27, 144–149. Cottrell, S. and Laird, P. (2003). Sensitive detection of DNA methylation. Ann N Y Acad Sci 983, 120–130. Cottrell, S., Distler, J., Goodman, N., Mooney, S., Kluth, A., Olek, A., Schwope, I., Tetzner, R., Ziebarth, H. and Berlin, K. (2004). A realtime PCR assay for DNA-methylation using methylation-specific blockers. Nucleic Acids Res 32, e10. Cross, S.H., Charlton, J.A., Nan, X. and Bird, A.P. (1994). Purification of CpG islands using a methylated DNA binding column. Nat Genet 6, 236–244. Cunningham, J.M., Christensen, E.R.,Tester, D.J., Kim, C.Y., Roche, P.C., Burgart, L.J. and Thibodeau, S.N. (1998). Hypermethylation of the hMLH1 promoter in colon cancer with microsatellite instability. Cancer Res 58, 3455–3460. Eads, C., Danenberg, K., Kawakami, K., Saltz, L., Blake, C., Shibata, D., Danenberg, P. and Laird, P. (2000). MethyLight: A high-throughput assay to measure DNA methylation. Nucleic Acids Res 28, E32. Eckhardt, F., Lewin, J., Cortese, R., Rakyan,V.K., Attwood, J., Burger, M., Burton, J., Cox, T.V., Davies, R., Down, T.A. et al. (2006). DNA methylation profiling of human chromosomes 6, 20, and 22. Nat Genet 38, 1378–1385. Ehrlich, M., Gama-Sosa, M.A., Huang, L.H., Midgett, R.M., Kuo, K.C., McCune, R.A. and Gehrke, C. (1982). Amount and distribution of 5-methylcytosine in human DNA from different types of tissues of cells. Nucleic Acids Res 10, 2709–2721. El-Maarri, O., Herbiniaux, U., Walter, J. and Oldenburg, J. (2002). A rapid, quantitative, non-radioactive bisulfite-SNuPE- IP RP HPLC assay for methylation analysis at specific CpG sites. Nucleic Acids Res 30, e25. ENCODE Project Consortium (2007). Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447, 799–816. Esteller, M. (2006). The necessity of a human epigenome project. Carcinogenesis 27, 1121–1125. Esteller, M., Garcia-Foncillas, J., Andion, E., Goodman, S., Hidalgo, O., Vanaclocha,V., Baylin, S. and Herman, J. (2000). Inactivation of the DNA-repair gene MGMT and the clinical response of gliomas to alkylating agents. N Engl J Med 343, 1350–1354. Feinberg, A.P. (2004). The epigenetics of cancer etiology. Semin Cancer Biol 14, 427–432. Feinberg, A.P. (2007). Phenotypic plasticity and the epigenetics of human disease. Nature 447, 433–440.
Frommer, M., McDonald, L.E., Millar, D.S., Collis, C.M., Watt, F., Griff, G.W., Molloy, P.L. and Paul, C.L. (1992). A genomic sequencing protocol that yields a positive display of 5-methylcytosine residues in individual DNA strands. Proc Natl Acad Sci USA 89, 1827–1831. Fukasawa, M., Kimura, M., Morita, S., Matsubara, K., Yamanaka, S., Endo, C., Sakurada, A., Sato, M., Kondo, T., Horii, A. et al. (2006). Microarray analysis of promoter methylation in lung cancers. J Hum Genet 51, 368–374. Gebhard, C., Schwarzfischer, L., Pham, T.H., Schilling, E., Klug, M., Andreesen, R. and Rehli, M. (2006). Genome-wide profiling of CpG methylation identifies novel targets of aberrant hypermethylation in myeloid leukaemia. Cancer Res 66, 6118–6128. Gonzalgo, M.L. and Jones, P.A. (1997). Rapid quantitation of methylation differences at specific sites using methylation-sensitive single nucleotide primer extension (Ms-SNuPE). Nucleic Acids Res 25, 2529–2531. Grady, W.M. (2005). Epigenetic events in the colorectum and in colon cancer. Biochem Soc Trans 33, 684–688. Grayson, D.R., Jia, X., Chen,Y., Sharma, R.P., Mitchell, C.P., Guidotti, A. and Costa, E. (2005). Reelin promoter hypermethylation in schizophrenia. Proc Natl Acad Sci USA 102, 9341–9346. Grunau, C., Clark, S.J. and Rosenthal, A. (2001). Bisulfite genomic sequencing: Systematic investigation of critical experimental parameters. Nucleic Acids Res 29, e65–e65. Hansen, R.S., Wijmenga, C., Luo, P., Stanek, A.M., Canfield, T.K., Weemaes, C.M. and Gartler, S.M. (1999). The DNMT3B DNA methyltransferase gene is mutated in the ICF immunodeficiency syndrome. Proc Natl Acad Sci USA 96, 14412–14417. Harbeck, N., Bohlmann, I., Ross, J.S., Gruetzmann, R., Kristiansen, G., Margossian, A., Hartmann, A., Cufer, T., Paradiso, A. and Maier, S. EpiBreast Group (2005). Multicenter study validates PITX2 DNA methylation for risk prediction in tamoxifen-treated, node-negative breast cancer using paraffin-embedded tumor tissue. J Clin Oncol (ASCO Annual Meeting Proceedings) 23, 505. Hatada, I., Hayashizaki, Y., Hirotsune, S., Komatsubara, H. and Mukai, T. (1991). A genomic scanning method for higher organisms using restriction sites as landmarks. Proc Natl Acad Sci USA 88, 9523–9527. Hayashi, H., Nagae, G.,Tsutsumi, S., Kaneshiro, K., Kozaki,T., Kaneda, A., Sugisaki, H. and Aburatani, H. (2007). High-resolution mapping of DNA methylation in human genome using oligonucleotide tiling array. Hum Genet 120, 701–711. Hellman, A. and Chess, A. (2007). Gene body-specific methylation on the active X chromosome. Science 315, 1141–1143. Henrique, R. and Jeronimo, C. (2004). Molecular detection of prostate cancer: A role for GSTP1 hypermethylation. Eur Urol 46, 660–669. Henrique, R., Jeronimo, C.,Teixeira, M.R., Hoque, M.O., Carvalho, A.L., Pais, I., Ribeiro, F.R., Oliveira, J., Lopes, C. and Sidransky, D. (2006). Epigenetic heterogeneity of high-grade prostatic intraepithelial neoplasia: Clues for clonal progression in prostate carcinogenesis. Mol Cancer Res 4, 1–8. Herman, J.G. and Baylin, S.B. (2003). Gene silencing in cancer in association with promoter hypermethylation. N Engl J Med 349, 2042–2054. Herman, J.G., Graff, J.R., Myohanen, S., Nelkin, B.D. and Baylin, S.B. (1996). Methylation-specific PCR: A novel PCR assay for methylation status of CpG islands. Proc Natl Acad Sci USA 93, 9821–9826.
References
Hitchins, M.P., Wong, J.J., Suthers, G., Suter, C.M., Martin, D.I., Hawkins, N.J. and Ward, R.L. (2007). Inheritance of a cancerassociated MLH1 germ-line epimutation. N Engl J Med 356, 731–733. Holdenrieder, S., Stieber, P., Bodenmuller, H., Busch, M., Fertig, G., Furst, H., Schalhorn, A., Schmeller, N., Untch, M. and Seidel, D. (2001). Nucleosomes in serum of patients with benign and malignant diseases. Int J Cancer 95, 114–120. Hoque, M.O., Topaloglu, O., Begum, S., Henrique, R., Rosenbaum, E., Van Criekinge, W., Westra, W.H. and Sidransky, D. (2005). Quantitative methylation-specific polymerase chain reaction gene patterns in urine sediment distinguish prostate cancer patients from control subjects. J Clin Oncol 23, 6569–6575. Huang, T.H., Perry, M.R. and Laux, D.E. (1999). Methylation profiling of CpG islands in human breast cancer cells. Hum Mol Genet 8, 459–470. Jahr, S., Hentze, H., Englisch, S., Hardt, D., Fackelmayer, F.O., Hesch, R.D. and Knippers, R. (2001). DNA fragments in the blood plasma of cancer patients: Quantitations and evidence for their origin from apoptotic and necrotic cells. Cancer Res 61, 1659–1665. Januchowski, R., Prokop, J. and Jagodzinski, P.P. (2004). Role of epigenetic DNA alterations in the pathogenesis of systemic lupus erythematosus. J Appl Genet 45, 237–248. Jiang, W.G., Sampson, J., Martin, T.A., Lee-Jones, L., Watkins, G., Douglas-Jones, A., Mokbel, K. and Mansel, R.E. (2005). Tuberin and hamartin are aberrantly expressed and linked to clinical outcome in human breast cancer: The role of promoter methylation of TSC genes. Eur J Cancer 41, 1628–1636. Jones, P.A. and Laird, P.W. (1999). Cancer epigenetics comes of age. Nat Genet 21, 163–167. Jones, P.A. and Baylin, S.B. (2002). The fundamental role of epigenetic events in cancer. Nat Rev Genet 3, 415–428. Jones, P.A. and Martienssen, R. (2005). A blueprint for a Human Epigenome Project: The AACR Human Epigenome Workshop. Cancer Res 65, 11241–11246. Karran, P. and Bignami, M. (1994). DNA damage tolerance, mismatch repair and genome instability. Bioessays 16, 833–839. Kim, J., Kim, J.Y., Song, K.S., Lee, Y.H., Seo, J.S., Jelinek, J., Goldschmidt-Clermont, P.J. and Issa, J.P. (2007). Epigenetic changes in estrogen receptor beta gene in atherosclerotic cardiovascular tissues and in-vitro vascular senescence. Biochim Biophys Acta 1772, 72–80. Krueger, T., Wild, C. and Noyer-Weidner, M. (1995). McrB: A prokaryotic protein specifically recognizing DNA containing modified cytosine residues. EMBO J 14, 2661–2669. Laird, P.W. (2003). The power and promise of DNA methylation markers. Nat Rev Cancer 3, 253–266. Lander, E., Linton, L., Birren, B., Nusbaum, C., Zody, M., Baldwin, J., Devon, K., Dewar, K., Dewar, I., Doyle, M. et al. (2001). Initial sequencing and analysis of the human genome. Nature 409, 860–921. Lewin, J., Plum, A., Hildmann, T., Rujan, T., Eckhardt, F., Liebenberg,V., Lofton-Day, C. and Wasserkort, R. (2007). Comparative DNA methylation analysis in normal and tumour tissues and in cancer cell lines using differential methylation hybridisation. Int J Biochem Cell Biol 39, 1539–1550. Liang, G., Gonzalgo, M.L., Salem, C. and Jones, P.A. (2002). Identification of DNA methylation differences during tumorigenesis by methylation-sensitive arbitrarily primed polymerase chain reaction. Methods 27, 150–155.
■
141
Lofton-Day, C., Model, F.,Tetzner, R., deVos,T., Schuster, M., Lesche, R., Sledziewski, A. and Day, R. (2006). A real-time PCR test for Septin 9 gene methylation identifies early stage colorectal cancer in plasma. 97th Annual Meeting of the AACR. Abstract LB-224. Look, M.P., van Putten, W.L., Duffy, M.J., Harbeck, N., Christensen, I.J., Thomssen, C., Kates, R., Spyratos, F., Ferno, M. and EppenbergerCastori, S. (2002). Pooled analysis of prognostic impact of urokinase-type plasminogen activator and its inhibitor PAI-1 in 8377 breast cancer patients. J Natl Cancer Inst 94, 116–128. Lopatina, N., Haskell, J.F., Andrews, L.G., Poole, J.C., Saldanha, S. and Tollefsbol, T. (2002). Differential maintenance and de novo methylating activity by three DNA methyltransferases in aging and immortalized fibroblasts. J Cell Biochem 84, 324–334. Ludlum, D.B. (1990). DNA alkylation by the haloethylnitrosoureas: Nature of modifications produced and their enzymatic repair or removal. Mutat Res 233, 117–126. Lund, G., Andersson, L., Lauria, M., Lindholm, M., Fraga, M.F., VillarGarea, A., Ballestar, E., Esteller, M. and Zaina, S. (2004). DNA methylation polymorphisms precede any histological sign of atherosclerosis in mice lacking apolipoprotein E. J Biol Chem 279, 29147–29154. Maier, S., Nimmrich, I., Marx, A., Eppenberger-Castori, S., Jaenicke, F., Paradiso, A., Spyratos, F., Foekens, J., Schmitt, M. and Harbeck, N. (2004). DNA methylation profile predicts risk of recurrence in tamoxifen-treated, node-negative breast cancer patients. J Clin Oncol (ASCO Annual Meeting Proceedings) 22, 525. Maier, S., Dahlstroem, C., Haefliger, C., Plum, A. and Piepenbrock, C. (2005). Identifying DNA methylation biomarkers of cancer drug response. Am J Pharmacogenomics 5, 223–232. Maier, S., Nimmrich, I., Koenig,T., Eppenberger-Castori, S., Bohlmann, I., Paradiso, A., Spyratos, F., Thomssen, C., Mueller, V., Naehrig, J. et al. (2007). DNA-methylation of the homeodomain transcription factor PITX2 reliably predicts risk of distant disease recurrence in tamoxifen-treated, node-negative breast cancer patients – Technical and clinical validation in a multi-center setting in collaboration with the European Organisation for Research and Treatment of Cancer (EORTC) PathoBiology group. Eur J Cancer 43, 1679–1686. Martens, J.W., Maier, S., Schwope, I., Koenig,T., Hartmann, O., Kluth, A., Look, M., Harbeck, N., Bolt-de Vries, J., Sieuwerts, A. et al. (2005). DNA-hypermethylation of PITX2 and risk of cancer recurrence in lymph node-negative hormone receptor-positive breast cancer patients. 96th Annual Meeting of the AACR. Abstract 2762. Mohandas, T., Sparkes, R.S. and Shapiro, L.J. (1981). Reactivation of an inactive human X chromosome: Evidence for X inactivation by DNA methylation. Science 211, 393–396. Nicholls, R.D. and Knepper, J.L. (2001). Genome organization, function, and imprinting in Prader-Willi and Angelman syndromes. Annu Rev Genomics Hum Genet 2, 153–175. Oberle, I., Rousseau, F., Heitz, D., Kretz, C., Devys, D., Hanauer, A., Boue, J., Bertheas, M.F. and Mandel, J.L. (1991). Instability of a 550-base pair DNA segment and abnormal methylation in fragile X syndrome. Science 252, 1097–1102. Ogino, S., Kawasaki, T., Brahmandam, M., Cantor, M., Kirkner, G.J., Spiegelman, D., Makrigiorgos, G.M.,Weisenberger, D.J., Laird, P.W., Loda, M. et al. (2006a). Precision and performance characteristics of bisulfite conversion and real-time PCR (MethyLight) for quantitative DNA methylation analysis. J Mol Diagn 8, 209–217. Ogino, S., Cantor, M., Kawasaki, T., Brahmandam, M., Kirkner, G., Weisenberger, D.J., Campan, M., Laird, P.W., Loda, M. and
142
CHAPTER 11
■
DNA Methylation Analysis
Fuchs, C.S. (2006b). CpG island methylator phenotype (CIMP) of colorectal cancer is best characterized by quantitative DNA methylation analysis and prospective cohort studies. Gut 55, 1000–1006. Okada, M., Ogasawara, H., Kaneko, H., Hishikawa, T., Sekigawa, I., Hashimoto, H., Maruyama, N., Kaneko, Y. and Yamamoto, N. (2002). Role of DNA methylation in transcription of human endogenous retrovirus in the pathogenesis of systemic lupus erythematosus. J Rheumatol 29, 1978–1982. Okano, M., Bell, D.W., Haber, D.A. and Li, E. (1999). DNA methyltransferases Dnmt3a and Dnmt3b are essential for de novo methylation and mammalian development. Cell 99, 247–257. Paredes, J., Albergaria, A., Oliveira, J.T., Jeronimo, C., Milanezi, F. and Schmitt, F.C. (2005). P-cadherin overexpression is an indicator of clinical outcome in invasive breast carcinomas and is associated with CDH3 promoter hypomethylation. Clin Cancer Res 11, 5869–5877. Paz, M.F., Yaya-Tur, R., Rojas-Marcos, I., Reynes, G., Pollan, M., Aguirre-Cruz, L., Garcia-Lopez, J.L., Piquer, J., Safont, M.J. and Balana, C. (2004). CpG island hypermethylation of the DNA repair enzyme methyltransferase predicts response to temozolomide in primary gliomas. Clin Cancer Res 10, 4933–4938. Powell, S.M., Zilz, N., Beazer-Barclay, Y., Bryan, T.M., Hamilton, S.R., Thibodeau, S.N., Vogelstein, B. and Kinzler, K.W. (1992). APC mutations occur early during colorectal tumorigenesis. Nature 359, 235–237. Richardson, B.C. (2002). Role of DNA methylation in the regulation of cell function: Autoimmunity, aging and cancer. J Nutr 132, 2401S–2405S. Schuster, M. (2004). DNA methylation analysis: A powerful diagnostic tool. J Lab Med 28, 233–238. Shimabukuro, M., Sasaki, T., Imamura, A., Tsujita, T., Fuke, C., Umekage, T., Tochigi, M., Hiramatsu, K., Miyazaki, T., Oda, T. et al. (2006). Global hypomethylation of peripheral leukocyte DNA in male patients with schizophrenia: A potential link between epigenetics and schizophrenia. J Psychatr Res e-publication. Shinozaki, M., Hoon, D.S., Giuliano, A.E., Hansen, N.M., Wang, H.J., Turner, R. and Taback, B. (2005). Distinct hypermethylation profile of primary breast cancer is associated with sentinel lymph node metastasis. Clin Cancer Res 11, 2156–2162.
Stirzaker, C., Song, J.Z., Davidson, B. and Clark, S.J. (2004). Transcriptional gene silencing promotes DNA hypermethylation through a sequential change in chromatin modifications in cancer cells. Cancer Res 64, 3871–3877. Takahashi,Y., Miyoshi,Y., Takahata, C., Irahara, N., Taguchi, T., Tamaki,Y. and Noguchi, S. (2005). Down-regulation of LATS1 and LATS2 mRNA expression by promoter hypermethylation and its association with biologically aggressive phenotype in human breast cancers. Clin Cancer Res 11, 1380–1385. Takai, D. and Jones, P.A. (2002). Comprehensive analysis of CpG islands in human chromosomes 21 and 22. Proc Natl Acad Sci USA 99, 3740–3745. Taylor, K.H., Pena-Hernandez, K.E., Davis, J.W., Arthur, G.L., Duff , D.J., Shi, H., Rahmatpanash, F.B., Sjahputera, O. and Caldwell, C.W. (2007). Large-scale CpG methylation analysis identifies novel candidate genes and reveals methylation hotspots in acute lymphoblastic leukemia. Cancer Res 67, 2617–2625. Toyota, M., Ho, C., Ahuja, N., Jair, K.W., Li, Q., Ohe-Toyota, M., Baylin, S.B. and Issa, J.P. (1999). Identification of differentially methylated sequences in colorectal cancer by methylated CpG island amplification. Cancer Res 59, 2307–2312. Weksberg, R., Smith, A.C., Squire, J. and Sadowski, P. (2003). BeckwithWiedemann syndrome demonstrates a role for epigenetic control of normal development. Hum Mol Genet 12, R61–R68. Wu,Y., Halverson, G., Basir, Z., Strawn, E.,Yan, P. and Guo, S.W. (2005). Aberrant methylation at HOXA10 may be responsible for its aberrant expression in the endometrium of patients with endometriosis. Am J Obstet Gynecol 193, 371–380. Yan, P.S., Chen, C.M., Shi, H., Rahmatpanah, F.,Wei, S.H., Caldwell, C.W. and Huang, T.H. (2001). Dissecting complex epigenetic alterations in breast cancer using CpG island microarrays. Cancer Res 61, 8375–8380. Zaina, S., Lindholm, M.W. and Lund, G. (2005). Nutrition and aberrant DNA methylation patterns in atherosclerosis: More than just hyperhomocystinemia?. J Nutr 135, 5–8. Zhu, S., Goldschmidt-Clermont, P.J. and Dong, C. (2005). Inactivation of monocarboxylate transporter MCT3 by DNA methylation in atherosclerosis. Circulation 112, 1353–1361.
RECOMMENDED RESOURCES Websites
Journal
http://www.epigenome.org – The website for the Human Epigenome Project. http://www.microarrays.ca – This is the website for the microarray center at the University Health Network in Toronto. They describe a spotted array containing over 12,000 CpG island clones. http://epigenome-noe.net/ – The website for the Epigenome Network of Excellence, a consortium of 25 European research groups. http://www.dnamethsoc.com/ – Website for the Epigenetics Society.
Laird, P.W. (2003). The power and promise of DNA methylation markers. Nat Rev Cancer 3(4), 253–266. This paper provides a comprehensive review of methylation detection and measurement methods as well as a review of the literature applying these methods to clinical samples.
CHAPTER
12 Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine Michelle M. Kittleson, Rafael Irizarry, Bettina Heidecker and Joshua M. Hare
INTRODUCTION Transcriptomics is defined as the genome-wide study of mRNA expression levels. With this promising technology, it is now possible to assess the expression of tens of thousands of gene transcripts simultaneously, providing a resolution and precision of phenotypic characterization of the transcriptome not previously possible. Because the state of the transcriptome in a given diseased tissue may contain a highly accurate representation of key biological phenomena, patterns of gene expression have potential to provide insights into disease mechanisms and also to identify markers useful for diagnostic, prognostic, and therapeutic purposes. This chapter will provide an overview of this powerful new technology, with a discussion of the statistical methods and applications of transcriptomics as well as current issues in this field. We discuss microarrays that are based on genomic information as well as specialized microarrays designed to detect micro-RNAs.
GENE EXPRESSION TECHNOLOGY Types of Microarrays The study of the transcriptome can be accomplished with the use of microarrays. Microarrays rely on the principle of Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
complementary hybridization of nucleotide sequences and have taken advantage of robotic technology to create platforms as small as a few inches wide, containing tens of thousands of DNA sequences for analysis. There are two commonly used types of gene expression microarrays. The first type, traditionally known as cDNA microarrays, utilizes relatively long multimers of probe cDNAs (500–5000 bases). Such microarrays are made primarily in individual institutions and are often organ- or disease-specific, examples include the CardioChip from Brigham and Women’s Hospital (Barrans et al., 2001) and the LymphoChip from Stanford University (Alizadeh et al., 1999). Usually, one or a few long probes are used to assay the expression of a given gene. They often use two-colors to measure the identity and intensity of gene expression (i.e., red and green), so that on an individual microarray, the gene expression of two samples is compared or individual samples are compared to a reference sample. Thus, the two samples are tagged with distinct labels that fluoresce at a distinct and non-overlapping wavelength and the relative, as opposed to absolute, level of expression is obtained (Figure 12.1) (Albelda and Sheppard, 2000). The second type of microarray is known as an oligonucleotide microarray (Figure 12.2). These microarrays use smaller DNA probes (20–80 bases) and have a number of advantages: (i) they are commercially produced by companies such as Affymetrix, Copyright © 2009, Elsevier Inc. All rights reserved. 143
144
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
(a) RNA isolation Sample A
Sample B
(e) Imaging Sample A B Sample B A
Sample A B
(b) cDNA generation (c) Labeling of probe Reverse Transcriptase Fluorescent tags
(d) Hybridization to array
Figure 12.1 The use of two-color cDNA microarrays. First, the messenger RNA (mRNA) is isolated from two separate samples (a). The mRNA from each sample is treated with reverse transcriptase (b) and labeled with a distinct fluorescent tag (c). The two pools of labeled RNA are hybridized to the microarray, containing a full set of thousands or tens of thousands of DNA sequences, (d). The microarray is scanned and the color of each spot is determined (e). In this example, genes expressed only in Sample A would be red, genes expressed only in Sample B would be green and those genes expressed equally in both samples would be yellow. This allows genes that are specifically expressed in response to the specific treatment or disease to be determined. (Reproduced with permission from Albelda, S.M. and Sheppard, D. Functional genomics and expression profiling: Be there to be square. Am J Respir Cell Mol Biol 2000: 23, 265–269).
Agilent, Amersham, or AME Bioscience, and thus the quality control is standardized and the cost of production is minimized; (ii) they typically contain an unbiased genome-wide rather than organ system-specific set of probes; and (iii) they contain more gene transcripts (almost 50,000 on the newest Affymetrix U133 Plus 2.0 GeneChip Array, for example). These are one-color microarrays, so that individual chips are used to study individual samples. In addition, multiple short probes are used to assay the expression of a given gene. Thus, a given gene may be represented by multiple probes spanning the length of the gene, such that the absolute gene expression is determined by incorporating the level of hybridization intensity of all the probes that make up a given gene. Both cDNA and oligonucleotide microarrays consist of a solid support, usually a glass slide or a nylon membrane, onto which DNA sequences are attached. For a microarray experiment, RNA from a sample is extracted, labeled, and, after
hybridization, complementary sequences remain bound to the array. Expressed genes are identified by the position of the bound probes on the microarray, and their abundance is determined by the intensity of the measured signal. Thus, an experiment with a single microarray can provide information on the expression of thousands of genes simultaneously: the transcriptome of a given tissue sample can thus be determined. The two most popular microarrays can be divided into Affymetrix GeneChips, a one channel, multi-probe, high-density oligonucleotide microarray, and two-color, one-probe per gene cDNA microarrays.There is no company that dominates the market for the latter so we will refer to these here as two-color arrays. Quality Control After microarray hybridization, the next essential step in microarray analysis is assurance of adequate array quality. All arrays
Gene Expression Technology
■
145
Hybridized probe cell Single-stranded, labeled RNA target
GeneChip® Probe Array
Oligonucleotide probe 24 μm 1.28 cm Millions of copies of a specific oligonucleotide probe
20,000 different complementary probes
Image of hybridized probe array Total RNA
cDNA AAAA
Reverse Transcription
Biotin-labeled cRNA B In Vitro Transcription
AAAA AAAA
Fragmentation
GeneChip® Expression Array
Fragmented, Biotin-labeled cRNA
Hybridization
Wash and stain
Scan and quantitate
Figure 12.2 (a) Schematic relating the oligonucleotide microarray platform, probe sequences, and scanned, hybridized image. The microarray is a solid support upon which nucleic acid probes corresponding to known gene transcripts are attached to specific locations. (b) The many steps involved in microarray analysis. Total RNA is isolated from tissue samples and used to synthesize doublestranded cDNA, which is then used as a template to make biotin-labeled cRNA. Fragmented, biotin-labeled cRNA is hybridized to a microarray, containing nucleic acid probes attached to the solid support. After washing, the microarray is scanned. By monitoring the amount of label associated with each probe location, it is possible to infer the abundance of each mRNA species represented. (Reproduced with permission from www.affymetrix.com).
should be subject to quality control. Array manufacturers typically provide software for quality control assessments. Typically these include tools for visual inspection of the array images for the presence of artifacts, and assessment of background levels (Figure 12.3). The academic community has produced quality control methodology as well for the Affymetrix (Bolstad et al., 2005) and two-color platforms (Ritchie et al., 2006). Only
arrays in which all parameters are within acceptable levels should be subject to further analysis. Normalization The final step in microarray preparation is conversion of feature-level data into gene expression levels (see also Chapter 13). Because genes on a microarray are represented by multiple
146
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
(a)
TABLE 12.1 Two major applications of microarray studies – Gene discovery and molecular signature analysis Gene discovery
Molecular signature analysis
Goal of analysis
Elucidate novel genetic pathways and therapeutic targets
Refine diagnosis and treatment
What is identified
Differentially expressed genes between disease states
A pattern of genes that characterizes a clinical parameter: diagnosis, prognosis, or response to therapy
Goal of validation
Confirm levels of gene expression
Assess the predictive accuracy of the molecular signature
Validation strategy
Quantitative PCR, Northern blotting, RNase protection assays
Apply the molecular signature to independent samples
(b)
Figure 12.3 Scanned images of hybridized Affymetrix microarrays. (a) A proper scanned image, with uniform intensity and no image artifacts. (b) An image with artifacts. The four, irregular dark spots at the center of the scanned image suggest irregularities of hybridization caused by variation in technique. This could affect the evaluation of gene expression and a new sample should be processed.
features, the hybridization intensities must be summarized to determine the overall expression or relative expression levels. In the case of two-color platforms, red and green foregrounds and background measurements are summarized into one value that quantifies relative expression between the two samples. In the Affymetrix platforms, the various features representing a gene need to be summarized into one value that quantifies expression. Background correction and normalization are issues that are common to all platforms. Background and signal adjustment corrects for background noise, adjusts for cross hybridization, and ensures that all values fall on the proper scale. Normalization eliminates systematic differences between arrays, either using a reference array or the combined information from all the arrays. The background-corrected, normalized, and converted probe summaries of gene expression can then be used for gene-level microarray analysis.
Major Applications of Microarray Analysis The challenge in microarray experiments is in the experimental design and statistical analysis, where the number of variables assessed is orders of magnitude higher than the number of individuals studied. There is concern of the dangers of data-mining, where lists of thousands of genes are generated without an understanding of the technical and statistical pitfalls of microarray analysis. To avoid this, it is essential to understand the distinct applications of this technology. Because the state of the transcriptome in a given diseased tissue may contain a highly accurate representation of key biological phenomena, patterns of gene expression have potential to provide insights into disease mechanisms and also to identify markers useful for diagnostic, prognostic, and therapeutic purposes. Thus, the two distinct major applications of this powerful technology are gene discovery and molecular signature analysis, and these two applications have different goals, statistical methods, and validation strategies (Table 12.1).
GENE DISCOVERY Overview Gene discovery involves the identification of differentially expressed genes between two or more states, for example disease
Gene Discovery
and normal tissue. The goal of gene discovery is to elucidate novel genetic pathways or cause-specific therapies. Experimental Design Given that gene discovery involves the comparison of gene expression between two or more disease states, samples must be chosen that belong to different groups: for example, normal versus disease, with and without treatment, or before and after an intervention. However, three major issues in experimental design bear further consideration: the utility of randomization, replication, and pooling. 1. Randomization is not widely used in biomedical laboratory experiments, but offers significant advantages here. One of the most striking aspects of oligonucleotide microarray experiments is the extent to which the experimental conditions can affect the resulting expression data. While normalization methods can reduce the impact of some of these differences, randomization is the best way to deal with them. In this manner, the samples can be grouped so that any unique chance features are shared equally by the two groups and thus are not potentially confounded with group differences. For example, all normal samples should not be analyzed on one day and all diseased samples on the next day. Similar considerations can apply whenever differences will occur across other factors known to contribute extraneous variability, such as chip batch, operator, reagents, scanners, and so on. Arrange those varying factors that can be controlled in a manner similar to that just described, and randomize across the remainder (Bolstad et al., 2004). 2. Another major issue in experimental design is whether replication should be performed. Here, it is essential to make the distinction between technical replicates and biological replicates. Technical replicates involve multiple measurements from the same tissue sample, whereas biological replicates involve multiple measurements from the same disease state. Clearly biological replication leads to data that are better for reaching conclusions that might apply more generally to the disease state, and technical replication leads to data better for reaching conclusions about that particular sample. In most cases, biological replicates are more relevant to the aims of an experiment than technical replicates. Chips are expensive, and biological replication will allow better use of limited resources getting more data at the level exhibiting higher variation, across subjects, say, than at the level exhibiting lower variation, within subjects (Bolstad et al., 2004). 3. A third major experimental design issue in gene discovery analyses is whether or not to pool samples. One perspective is that pooling provides a form of “biological averaging” and should make it possible to get more precise results with fewer chips than hybridizing RNA from individual subjects to separate chips. While this may be true, there is no compelling evidence for this in the literature, and pooling could mislead without being apparent. Suppose that one subject contributing to the pool is very different from the
■
147
remainder. Depending on the tissue under study, pooling this subject’s RNA with that of the others gives it the ability to exert a large influence on the measured expression values for many genes, whereas running it on a separate chip leaves open the possibility of identifying this subject as an outlier. Furthermore, averaging can be performed at the statistical instead of biological level. This provides the same variance reduction property without losing the ability to detect outliers and assess variance. Thus, if pooling is envisaged in an experiment that could be carried out at the same cost without pooling, it is better not to pool (Kendziorski et al., 2005). Frequently, pooling is seen to be necessary to get sufficient mRNA from the tissue in question, in which case the possible drawbacks must inevitably be accepted or at least weighed against the possible drawbacks of the alternative, which is usually amplification (Bolstad et al., 2004). Statistical Methods Many statistical methods have been used to identify differentially expressed genes. The first attempts simply looked at fold-changes, and fold-changes 2 were initially considered meaningful. However, investigators soon realized that the significance of twofold change depends on both gene and sample. For this reason, it is now considered essential to compare expression between two or more groups by taking into account not only the magnitude of the difference between groups, but also the variability of expression between groups. The naïve solution to summarizing the relationship within and across-group variances is the well-known t-test/F-test and their respective p-values. However, there are two major problems with this approach. First, for small samples (fewer than 5 subjects), the t-and F-statistics tend to favor genes that, by chance, have small within-group variances. The second problem is that of multiple comparisons. When performing thousands of tests at once, the customary definition of a p-value no longer holds. For example, if we are measuring expression for 10,000 genes, all of which have no true across-group difference, we expect the 100 genes to reach p-values equal or smaller than 0.01. Various solutions have been given to both these problems. Empirical Bayes methods and Stein estimators have been proposed to improve estimates of within-group variation (Cui et al., 2005; Kendziorski et al., 2003; Lonnstedt and Speed, 2002; Smyth, 2004), and these methods greatly improve the results obtained with the t-test in small samples. Solutions to the multiple comparison problem have also been proposed (Dudoit et al., 2003; Storey, 2002). A popular piece of software that combines all these ideas is the Significance Analysis of Microarrays (SAM) (Tusher et al., 2001). SAM identifies genes with statistically significant changes in expression by identifying a set of gene-specific statistics (similar to the t-test, thus taking into account both magnitude of change and variability of expression) and a corresponding false discovery rate (similar to a p-value adjusted for multiple comparisons; see Figure 12.4). These ad hoc procedures provide similar results to the more rigorous procedures cited above.
148
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
(a)
d(i )
Observed relative difference d(i )
(b)
efficiency, as well as posttranslational protein modification and interactions, will all affect the measured protein abundance.
xI (i ) xU (i ) s(i ) S0
MOLECULAR SIGNATURE ANALYSIS
10
Overview In contrast to transcriptome analysis for gene discovery, the goal of molecular signature analysis is to identify a pattern of gene expression that is associated with a clinical parameter, such as etiology, prognosis, or response to therapy, thus potentially providing diagnostic or prognostic precision greater than that available from standard clinical information (Table 12.1).
5
0
5
10 10
5 5 0 Expected relative difference dE(i )
10
Figure 12.4 Depiction of Significance Analysis of Microarrays method of identifying differentially expressed genes. (a) SAM uses a modified t-test statistic to determine the observed and expected relative differences in gene expression. To calculate the observed relative difference in gene expression, d(i), the difference in average gene expression between the two groups is divided by the standard deviation of repeated measures of gene expression, s(i) and a constant, so to minimize the impact of genes with low levels of expression. The expected relative difference is the average of the relative difference of all possible permutations of the samples. (b) The observed relative difference in gene expression is then plotted against the expected relative difference, and this scatterplot is used to determine which genes are differentially expressed. Differentially expressed genes are those displaced from the line of unity by a distance greater than the chosen threshold, and the threshold is chosen to minimize the false discovery rate. (Reproduced with permission from Tusher, V.G., Tibshirani, R. and Chu,G. (2001). Significance analysis of microarrays applied to the ionizing radiation response. Proc Nat Acad Sci USA 98, 5116–5121).
Validation Once differentially expressed genes are identified, the transcript abundance of selected genes of interest are confirmed by a complementary method, such as quantitative PCR, Northern blotting, or RNase protection assays (Cook and Rosenzweig, 2002b). Less commonly, levels of the corresponding protein have also been measured, with less agreement between transcript and protein abundance. However, this is not surprising, since differences in mRNA localization, processing, stability, translation
Experimental Design Many features already discussed in the section on “Gene Discovery” apply as well to the design of experiments focused on molecular signature analysis, including the importance of randomization of experimental conditions and the utility of biological, rather than technical, replicates. While pooling samples is not ideal in microarray analyses focused on gene discovery, it is clearly impossible in studies focused on molecular signature analysis since the identity of individual samples is the goal. Thus, pooling of samples should never be performed in a study focused on molecular signature analysis. In molecular signature analysis, samples are first divided into groups based on a clinically relevant parameter, such as disease etiology, prognosis, or response to therapy. Then a molecular signature is created by choosing genes whose expression is solidly associated with the parameter in question, and by weighting genes based on their individual predictive strengths. Statistical Methods From a mathematical standpoint, molecular signature analysis is nothing other than what the computer science community denotes “machine learning” and the statistics community denotes “classification.” There are dozens, if not hundreds, of methods that have been used successfully in real-world applications such as voice recognition, zip code readers, and fraud detection. Relatively simple methods that are successful in other fields appear to work well in the context of microarrays. However, all these methods have been developed for cases where the number of predictors is smaller than the number of outcomes. With microarrays, this is rarely the case, given that there are more than 10,000 predictors. Nevertheless, if the number of genes included as predictors can be reduced to, say, less than 50 and chosen according to evidence in favor of differential expression, then existing methods work rather well (Dudoit et al., 2002). In fact, the simplest methods, including K-nearest methods and diagonalized linear discriminant analysis (DLDA), work better than more complicated ones. Prediction Analysis of Microarrays (PAM) is one approach, similar to DLDA, that is designed specifically for molecular signature analysis (Tibshirani et al., 2002).
Molecular Signature Analysis
EWS
NB
149
RMS
0
500
1000
Gene
1500
2000
BL
■
0.5
0.0
0.5
0.5
0.0
0.5
0.5
0.0
0.5
0.5
0.0
0.5
Average expression
Figure 12.5 Depiction of Prediction Analysis of Microarray’s method of nearest shrunken centroids. Samples are first divided into classes based on a pre-defined parameter. In this case, there are four groups from a dataset containing samples of four different types of small round blue cell tumors of childhood: Burkitt lymphoma (BL), Ewing sarcoma (EWS), neuroblastoma (NB), and rhabdomyosarcoma (RMS). The gray bars represent the standardized class centroid, the average expression of each gene in a given class divided by the within-class standard deviation (thus genes with stable expression have a greater contribution to the class centroid). The red bars represent the shrunken centroid, “de-noised” versions of centroids after subtraction of a given threshold determined by crossvalidation. After shrinkage, a small number of genes remain and these act as prototypes for each class. Independent samples are classified based on their squared distance from each prototypic class shrunken centroid. (Reproduced with permission from Tibshirani et al. 2002. Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Nat Acad Sci USA 99, 6567–6572).
PAM uses the method of nearest shrunken centroids to identify and validate the smallest set of genes whose expression is associated with a predefined class (Figure 12.5). A standardized class centroid is a composition of mean values of expression of all individual genes in a given class divided by the within-class standard deviation (SD). Standardization allows greater weight to be given to genes whose expression is stable within a class.
PAM creates shrunken centroids by shrinking the class centroids towards the overall centroid by a threshold amount. This overall centroid is calculated by dividing the average expression of each gene in all classes by the pooled within-class SD. Thus, the shrunken centroids are “de-noised” versions of centroids that act as prototypes for each class. The threshold is chosen by 10-fold cross-validation. In this process, gene expression prediction profiles
150
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
Overall CV plot 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
22283 17755 13602 9778 6777 4590 2918 1917 1224 803 545 357 236 161 116 78 52 38 28 20 15 11 7 4 3 2 1 1 1 0
Misclassification error
(a)
0
1
2
3
4
5
Threshold (b) Actual, predicted classes and predicted posterior probabilities LC1003.CEL LC1008.CEL LC1022.CEL Sample label 1 1 1 Actual class 1 1 1 Predicted class Probability: class 1 0.94986124 0.983986836 0.894571 Probability: class 2 0.05013876 0.016013164 0.105429
Actual class Predicted class Probability: class 1 Probability: class 2
H1a.CEL 1 1 0.92565835 0.07434165
H1b.CEL 2 2 0.386107146 0.613892854
H2a.CEL 1 1 0.935907 0.064093
LC1024.CEL 1 1 0.9658909 0.0341091
LC1025.CEL 1 1 0.980457 0.019543
LC1026.CEL 1 1 0.970757 0.029243
LT1001.CEL 2 2 0.042903 0.957097
LT1002.CEL 2 2 0.163733 0.836267
H2b.CEL 2 1 0.9809665 0.0190335
H3a.CEL 1 2 0.181081 0.818919
H3b.CEL 2 1 0.998226 0.001774
H4a.CEL 1 1 0.936414 0.063586
H4b.CEL 2 1 0.932558 0.067442
Figure 12.6 Sample output from Prediction Analysis of Microarrays (PAM) software. (a) Cross-validation plot listing the error rates in classification from 10-fold cross-validation of the training set. The x-axis is the threshold and number of genes in the molecular signatures, and the y-axis is the error rate. Here, the error rate in the training set is minimized by between 52 and 545 genes. Thus, a profile with 52 genes will be chosen as the molecular signature in order to minimize both the error rate and the number of genes in the profile. (b) The output using this 52-gene signature on the test set to determine if samples belong to class 1 or class 2. PAM provides the predicted class as well as the probability of belonging to class 1 or class 2. The sample is assigned to the class with the greater probability.
with different numbers of genes (i.e., from different thresholds) are fit on the basis of 90% of the samples and then tested on the remaining 10%. This process is executed multiple times, and the output is the error rate using different thresholds. The prediction profile is chosen as the smallest list of genes that can be used to predict the classification of an individual sample with the minimal error rate (Figure 12.6). To then classify independent samples, PAM computes each test sample’s squared distance from each of the class centroids.The predicted class is the one whose centroid is closest to the expression profile of the test sample. In contrast to SAM and other methods of identifying differentially expressed genes, PAM focuses mainly on the stability of gene expression and on the smallest number of genes required to create a molecular signature. Validation While the goal of gene discovery is to identify differentially expressed genes that offer insight into novel genetic pathways or cause-specific therapies, the goal of molecular signature analysis is to identify a pattern of genes that differentiates between clinical entities with a precision not possible based on standard clinical information. Thus, the identity of the mRNA transcripts in the signature or whether they are translated into protein may or may not have immediately discernable bearing on the utility of
the pattern. Therefore, the validation strategy is also unique: testing the accuracy of the identified molecular signature in samples distinct from those used to create the signature: an independent test set of samples (Table 12.1). However, because most laboratories have access to only a limited number of samples, withholding a substantial proportion of the samples from the training set for the sake of creating a validation set may considerably reduce the performance of the molecular signature. Cross-validation procedures use the data more efficiently. A small number of specimens are withheld, and most of the specimens are used to build a signature. The signature is used to predict class membership for the withheld specimens. This process is iterated, leaving out a new set of specimens at each step, until all specimens have been classified. In leave-one-out cross-validation, for example, each specimen is excluded from the training set one at a time and then classified on the basis of the predictor built from the data for all of the other specimens. The leave-one-out crossvalidation procedure provides a nearly unbiased estimate of the true error rate of the classification procedure (Simon et al., 2003a). However, when cross-validation (instead of an independent dataset) is used, all three steps of creating a molecular signature must undergo cross-validation, including: (i) selection of informative genes, (ii) computation of weights for selected informative
Current Issues in Gene Expression Analysis
genes, and (iii) creation of a prediction rule. Failure to crossvalidate all steps may lead to substantial bias in the estimated error rate and is a major pitfall in studies focused on molecular signature analysis (Simon et al., 2003b). Applications in Cancer Research In neoplastic disease, molecular signature analysis can determine prognosis and response to therapy. In breast cancer, a molecular signature was identified that predicted disease outcome in young patients with breast cancer better than standard clinical and histological criteria: a poor prognosis signature was associated with a fivefold increased risk of distant metastases in 5 years, a difference that would justify early intensive adjuvant chemotherapy (van de Vijver et al., 2002). In large B-cell lymphoma, a molecular signature predicted survival better than standard clinical methods (Rosenwald et al., 2002). Similar results have been obtained for acute myeloid leukemia (Bullinger et al., 2004; Valk et al., 2004), chronic lymphocytic leukemia (Calin et al., 2005), prostate cancer (Dhanasekaran et al., 2001; Lapointe et al., 2004), CNS tumors (Pomeroy et al., 2002), and non-small cell lung cancer (Chen et al., 2007). After its successful application in cancer, molecular signature analysis has been expanded to other fields in medicine, for example, congenital diseases (Eshaque and Dixon, 2006) enabling even preimplantation diagnosis (Salvado et al., 2004), and cardiovascular medicine. Applications in Cardiovascular Disease In cardiovascular disease, molecular signature analysis is in its earliest stages, but shows promise. In a proof-of-concept study, a diagnostic TBB was developed, which is able to distinguish between the two major forms of cardiomyopathy, ischemic and nonischemic (Kittleson et al., 2004). Several studies, investigating transcriptomic changes during left ventricular assist device (LVAD) placement followed, suggesting possible TBBs of recovery, which might provide mechanistic insight and suggest potential therapeutic targets (Hall et al., 2004; Margulies et al., 2005c). However, those studies also raised an important issue in microarray analysis, namely the difficulty to distinguish transcriptomic changes that were causative for a certain event from those that could result from a bystander effect (Depre et al., 1998; Margulies et al., 2005b). A way to reduce such confounding variables, is to obtain tissue at very early disease-stages, for example, by obtaining biopsies or blood samples (Heidecker and Hare, 2007b). Clinical TBBs derived from peripheral blood mononuclear cells (PBMC) are commercially available as diagnostic microarray chips to detect early rejection in cardiac transplant recipients (Deng et al., 2006b). Further, the use of TBBs was suggested as a screening tool for cardiovascular patients, who might require cardiac catheterization and stent implantation (Seo et al., 2004). Function of MicroRNAs in Cardiology Recent transcriptomic studies have discovered important regulatory functions of microRNAs (miRNAs) in cardiac remodeling
■
151
and heart failure (van et al., 2007b). MiRNAs are non-translated small single-stranded RNA molecules of 21–23 nucleotides in length. These small RNA species (Jackson and Standart, 2007b) either degrade bound mRNA or directly inhibit translation of mRNA by pairing with their target mRNAs in a sequence specific manner. Thus, miRNAs conduct negative regulatory function on gene expression (Jackson and Standart, 2007a). MiRNA of an intron of the MHC gene has been shown to be responsible for stress-induced cardiomyocyte growth and gene expression (van et al., 2007a). These findings point out the significant impact of gene–environment interactions on the phenotype of a cell or tissue and emphasize the value of comprehensive highthroughput technologies such as transcriptome profiling (Birney et al., 2007).
GENE DISCOVERY VERSUS MOLECULAR SIGNATURE ANALYSIS One important distinction in molecular signature analysis is the need first to identify a set of genes whose expression characterizes a pre-defined group of patients (with a clinically relevant distinction such as etiology, prognosis, or response to therapy), and then to test the predictive accuracy of this profile prospectively in an independent set of patients with varying phenotypes (Cook and Rosenzweig, 2002a). Another important distinction is that molecular signature analysis is based upon a pattern of gene expression rather than the identity of specific genes (Simon et al., 2003c) (Table 12.1). The prediction algorithm is able to compare an unknown sample and determine how closely it resembles one pattern versus the other; the absolute expression of an individual gene carries relatively small weight compared to the overall signature. Thus, validation cannot solely involve the confirmation of gene expression or gene product levels via a complementary technique such as quantitative PCR or immunofluorescence. Nevertheless, such validation can prove useful to address a different issue: whether the molecular signature offers utility independent of the microarray platform used to create it. This is important if diseasespecific platforms are developed.
CURRENT ISSUES IN GENE EXPRESSION ANALYSIS There are a number of issues in transcriptomic analysis that are being addressed by workers in the field (Table 12.2). Information Management and Reproducibility in Microarray Experiments Given the rapid growth of microarray research and the vast amount of information that can be gleaned from a single experiment, there are many challenges in designing studies and interpreting the results. To this end, the Microarray Gene Expression Data society (MGED) has developed the Minimum Information
152
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
TABLE 12.2 Current unresolved issues in transcriptomic analysis Current issues
Solutions
Study design, interpretation of results, reproducibility ●
●
Guidelines for experimental design, sample preparation, hybridization, normalization of array data, array design
MIAME
Data sharing
Standards Public Databases: National center for biotechnology information (NCBI); Gene expression omnibus (GEO); European bioinformatics institute (EMBLEBI) arrayexpress repository; Center for information biology gene expression (CIBEX); Cardiogenomics
Sample source ●
Relevance of investigated sample for later clinical application
Sample collection at early stage of disease, preferred tissue source: biopsies
●
Possible surrogates for diseased tissue or organ
PBMC
Power and sample size ●
Conventional power analysis not applicable (magnitude of change and error standard deviation not available)
Large sample size (in particular in clinical research), integrate biological knowledge with statistical analysis
About a Microarray Experiment (MIAME) standards that are needed to enable the interpretation of the results of the experiment unambiguously and potentially to reproduce the experiment (Brazma et al., 2001a). MIAME includes details of the experimental design, sample preparation, hybridization procedures, normalization algorithms, and array design. The ultimate goal is to establish a standard for recording and reporting microarray-based gene expression data, which will in turn facilitate the establishment of databases and public repositories and enable the development of data analysis tools. Such public repositories exist, including the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO); the European Bioinformatics Institute (EMBL-EBI) ArrayExpress repository; and the Center for Information Biology Gene Expression CIBEX) database. There is also a cardiology-specific repository, the Cardiogenomics Web site (www.cardiogenomics.com), a National Heart Lung and Blood Institute-sponsored Program for Genomic Applications. Many
journals currently require that submissions comply with the MIAME standard, but not all require that a complete dataset be submitted to one of three databases prior to manuscript submission. Furthermore, the public repositories do not require data being submitted in raw as well as normalized fashion, which would be essential for future studies. We believe that mandatory submission of raw microarray data files to public repositories, by promoting collaboration among scientists, could be essential for maximizing the utility of microarray research. The Source of Tissue for Analysis To date, microarray analyses in oncology have used mainly tumor tissue from biopsies. However, other pathological conditions cannot rely on such a ready source of tissue access. For example, microarray studies in cardiovascular disease have mainly utilized discarded myocardial tissue obtained at the time of cardiac transplantation, and there are limitations to this approach. The tissue is obtained from patients late in the disease course, and the conclusions, therefore, may not be applicable to patients at an earlier stage of disease. Thus, in the future, microarray analyses in cardiomyopathy will ideally focus on endomyocardial biopsy tissue obtained from patients at earlier stages of disease. Although endomyocardial biopsy is a safe procedure, future research will focus on PBMC as possible surrogate of diseased tissue, given easy accessibility by venipuncture (Heidecker and Hare, 2007a; Liew, 2005). In the cancer literature, molecular signatures derived from peripheral blood leukocytes offer comparable predictive accuracy to those from solid tumor samples in classifying subjects by cancer type and type of therapy (DePrimo et al., 2003; Twine et al., 2003). This may also be feasible in cardiovascular disease, as peripheral blood molecular signatures correlated with biopsyproven allograft rejection in cardiac transplant recipients (Deng et al., 2006a; Horwitz et al., 2004), cardiac genes in circulating blood were differentially expressed in patients with coronary artery disease relative to controls (Ma and Liew, 2003), and TBBs for familial combined hyperlipidemia were discovered in lymphocytes (Morello et al., 2004). Power and Sample Size in Microarray Experiments Clearly, molecular signature analysis is still in its earliest stages in cardiomyopathy, and there are still many unanswered questions. For example, there is limited knowledge on the sample sizes required in microarray experiments. The largest microarray study in cardiomyopathy to date has involved 199 patient samples (Margulies et al., 2005a), and in the oncology literature, each study has employed fewer than 300 patients. However, this should be considered a strength of these analyses: a succession of smaller studies, performed quickly and with the use of improving technology, will outperform larger studies locked into outdated approaches (Liu and Karuturi, 2004). The greater question is, however, whether traditional power calculations have any role in microarray experiments. A conventional power analysis is of limited value because the necessary variables are usually all unknown: we will only rarely be able
Conclusions
to nominate in advance the gene, the magnitude of change () of interest, or the magnitude of the measurement error SD. We also need to know the SD for all genes, and these are not the same. Nevertheless, it might still be argued that we should be able to determine the power we have with a given sample to detect all genes whose log fold change is some value or greater, and whose measurement SD is no more than a given (conservative) value, even if we will not know in advance which genes these are. However, there is a much greater obstacle to making use of conventional power analyses. In microarray studies, or other similar “-omic” studies, we are not measuring just one gene’s expression; in a typical experiment, we measure gene expression in 20,000 genes. Not only do we have to specify the number of genes we expect to be differentially expressed, but we need to decide on loss functions for false positives and negatives. In the typical situation where power calculations are used, we either get it right, commit one Type I (false positive) error, or one type II (false negative) error. Keep in mind that power is the probability of not committing a type II error when the null hypothesis is false. This definition does not apply to microarray applications where we report lists of genes hoping that a small percentage of the list are false positives and the list is long enough so that the number of false negatives is small. There is no easy solution to these problems, but current experiments rely on a few principles. First, one should aim to get as many case and control samples as possible, bearing in mind the important requirement of homogeneity. Second, the quality of the statistical analysis should improve, since the sample size cannot be increased beyond a certain point. A “smarter analysis” is needed to overcome the limitations of modest sample size. In addition, we need to depart from the conventional “context-free” search for differentially expressed genes that get embodied in power calculations and their multiple testing analogs, and more fully integrate biological knowledge with statistical analysis. However, even with these approaches, multiple testing issues will remain, so we have to extend that theory to apply to our stronger analysis. In brief, we are still in the early days of the statistical analysis of microarray data. Much traditional thinking must be extended and strengthened (Bolstad et al., 2004).
ALTERNATIVE TECHNOLOGIES FOR ANALYSIS OF THE TRANSCRIPTOME While, as presented here, microarray analysis is the most widely used approach for evaluating the transcriptome, there are several additional technologies that have been developed and have been used in particular settings. Serial Analysis of Gene Expression (SAGE) SAGE is a method of quantification of “tags”, rather than direct measurement of gene expression (Lash et al., 2000a). After isolation of mRNA, a small part of sequence is extracted from
■
153
each transcript at a defined position. All extracted pieces get linked together for subsequent cloning in a bacterial vector. Tags contain a nucleotide sequence of 9–11 bases in length and are representative for a unique gene transcript (Lash et al., 2000b). Tags are directly 3-adjacent to the 3-most restriction site for a particular restriction enzyme. Whereas the most widely used restriction enzyme is NlaIII, others such as Sau3A have also been used (Lash et al., 2000c). Several pitfalls with this technique warrant consideration. Because of the short length of tags, genes might share common base sequences, resulting in ambiguous tag-to-gene assignment. On the other hand, one gene can have more than one tag, because of polymorphism in a population or alternate termination in an individual (Lash et al., 2000d). Finally, by providing a digital output, fidelity of this method is reduced (Lash et al., 2000e). However, results from SAGE analysis have been shown to correlate highly with microarray data (Ishii et al., 2000), and results can be shared in public databases (http://www.ncbi.nlm.nih.gov/projects/SAGE). An important advantage of SAGE compared to microarrays is that the mRNA sequence does not have to be known in advance, so new polymorphisms can be discovered. Polony Multiplex Analysis of Gene Expression (PMAGE) PMAGE has been recently introduced as an extremely sensitive technique, which detects mRNA even in low abundance as small as one transcript per three cells (Kim et al., 2007a). Each mRNA is converted to cDNA, which gets cleaved and amplified onto beads carrying primers (Kim et al., 2007b). Each polony (for “PCR colony”) bead gets cross-linked to an aminosylated glass slide, from which fluorescent signal intensity can be measured via microscopy-based imaging (Kim et al., 2007c). Massively Parallel Signature Signaling (MPSS) MPSS was developed primarily to reduce time and cost of DNA sequencing as compared to conventional techniques (i.e., Sanger Sequencing (Sanger et al., 1977) and fluorescence-based electrophoresis technologies (Prober et al., 1987)). MPSS enables sequencing of as much as 25 million bases in a four-hour period, a rate that is about 100-times faster than current stateof-the art conventional sequencing. DNA is fragmented and bound to beads that are encased in droplets containing all reactants necessary to amplify the DNA. After amplification to about 10 million copies, DNA carrying beads are loaded into picolitre reactor wells and sequenced via luciferin reaction (Rogers and Venter, 2005). The same principles have been used for quantification of gene transcripts (Jongeneel et al., 2003).
CONCLUSIONS In conclusion, transcriptomics with microarray or other technologies offers a powerful means of phenotypic resolution not possible with standard clinical or histological criteria. The
154
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
promising applications of gene expression analysis include gene discovery and molecular signature analysis.These applications could elucidate novel genetic pathways, provide insight into causespecific therapies, and provide information on prognosis and response to therapy to individualize patient management in a variety of diseases. Although there are unresolved issues in this field, including power and sample size requirements, there is still
much promise, and we can expect further advances in refining the design and statistical methods of this technology in the future.
ACKNOWLEDGEMENTS This work was supported by NIH grant HL-65455 to JHM.
REFERENCES Albelda, S.M. and Sheppard, D. (2000). Functional genomics and expression profiling: Be there or be square. Am J Respir Cell Mol Biol 23, 265–269. Alizadeh, A., Eisen, M., Davis, R.E., Ma, C., Sabet, H.,Tran,T., Powell, J.I., Yang, L., Marti, G.E., Moore, D.T. et al. (1999). The lymphochip: A specialized cDNA microarray for the genomic-scale analysis of gene expression in normal and malignant lymphocytes. Cold Spring Harb Symp Quant Biol 64, 71–78. Barrans, J.D., Stamatiou, D. and Liew, C.C. (2001). Construction of a human cardiovascular cDNA microarray: Portrait of the failing heart. Biochem Biophys Res Comm 280, 964–969. Birney, E., Stamatoyannopoulos, J.A., Dutta, A., Guigo, R., Gingeras, T.R., Margulies, E.H., Weng, Z.P., Snyder, M., Dermitzakis, E.T., Stamatoyannopoulos, J.A. et al. (2007). Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447, 799–816. Bolstad, B.M., Collin, F., Simpson, K.M., Irizarry, R.A. and Speed, T.P. (2004). Experimental design and low-level analysis of microarray data. Int Rev Neurobiol 60, 25–58. Bolstad, B.M., Irizarry, R.A., Gautier, L. and Wu, Z. (2005). Preprocessing High-density Oligonucleotide Arrays. In Bioinformatics and Computational Biology Solutions Using R and Bioconductor (R. Gentleman,V.J. Carey,W. Huber, R.A. Irizarry and S. Dudoit, eds), Springer, New York, pp. 13–32. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeckert, C., Aach, J., Ansorge, W., Ball, C.A., Causton, H.C. et al. (2001a). Minimum information about a microarray experiment (MIAME)-toward standards for microarray data. Nat Genet 29, 365–371. Bullinger, L., Dohner, K., Bair, E., Frohling, S., Schlenk, R.F., Tibshirani, R., Dohner, H. and Pollack, J.R. (2004). Use of geneexpression profiling to identify prognostic subclasses in adult acute myeloid leukemia. N Engl J Med 350, 1605–1616. Butte, A. (2002). The use and analysis of microarray data. Nat Rev Drug Discov 1, 951–960. Calin, G.A., Ferracin, M., Cimmino, A., Di Leva, G., Shimizu, M., Wojcik, S.E.,Iorio, M.V.,Visone, R.,Sever, N.I.,Fabbri, M. et al. (2005). A MicroRNA signature associated with prognosis and progression in chronic lymphocytic leukemia. N Engl J Med 353, 1793–1801. Chen, H.Y., Yu, S.L., Chen, C.H., Chang, G.C., Chen, C.Y., Yuan, A., Cheng, C.L., Wang, C.H., Terng, H.J., Kao, S.F. et al. (2007). A fivegene signature and clinical outcome in non-small-cell lung cancer. N Engl J Med 356, 11–20. Cook, S.A. and Rosenzweig, A. (2002b). DNA microarrays: Implications for cardiovascular medicine. Circ Res 91, 559–564. Cui, X., Hwang, J.T., Qiu, J., Blades, N.J. and Churchill, G.A. (2005). Improved statistical tests for differential gene expression
by shrinking variance components estimates. Biostatistics 6, 59–75. Deng, M.C., Eisen, H.J., Mehra, M.R., Billingham, M., Marboe, C.C., Berry, G., Kobashigawa, J., Johnson, F.L., Starling, R.C., Murali, S. et al. (2006a). Noninvasive discrimination of rejection in cardiac allograft recipients using gene expression profiling. Am J Transplant 6, 150–160. Depre, C., Shipley, G.L., Chen, W., Han, Q., Doenst, T., Moore, M.L., Stepkowski, S., Davies, P.J. and Taegtmeyer, H. (1998). Unloaded heart in vivo replicates fetal gene expression of cardiac hypertrophy. Nat Med 4, 1269–1275. DePrimo, S.E.,Wong, L.M., Khatry, D.B., Nicholas, S.L., Manning,W.C., Smolich, B.D., O’Farrell, A.M. and Cherrington, J.M. (2003). Expression profiling of blood samples from an SU5416 Phase III metastatic colorectal cancer clinical trial: A novel strategy for biomarker identification. BMC Cancer 3, 3. Dhanasekaran, S.M., Barrette, T.R., Ghosh, D., Shah, R., Varambally, S., Kurachi, K., Pienta, K.J., Rubin, M.A. and Chinnaiyan, A.M. (2001). Delineation of prognostic biomarkers in prostate cancer. Nature 412, 822–826. Dudoit, S., Fridlyand, J. and Speed, T.P. (2002). Comparison of discrimination methods for the classification of tumors using gene expression data. J. Am. Stat Assoc 97, 77–87. Dudoit, S., Schaffer, J.P. and Boldrick, J.C. (2003). Multiple hypothesis testing in microarray experiments. Stat Sci 18, 71–103. Eshaque, B. and Dixon, B. (2006). Technology platforms for molecular diagnosis of cystic fibrosis. Biotechnol Adv 24, 86–93. Golub, T.R., Slonim, D.K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J.P., Coller, H., Loh, M.L., Downing, J.R., Caligiuri, M.A. et al. (1999). Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 286, 531–537. Hall, J.L., Grindle, S., Han, X., Fermin, D., Park, S., Chen,Y., Bache, R.J., Mariash, A., Guan, Z. and Ormaza, S. (2004). Genomic profiling of the human heart before and after mechanical support with a ventricular assist device reveals alterations in vascular signaling networks. Physiol Genomics 17, 283–291. Heidecker, B. and Hare, J.M. (2007a). The use of transcriptomic biomarkers for personalized medicine. Heart Fail Rev 12, 1–11. Horwitz, P.A., Tsai, E.J., Putt, M.E., Gilmore, J.M., Lepore, J.J., Parmacek, M.S., Kao, A.C., Desai, S.S., Goldberg, L.R., Brozena, S.C. et al. (2004). Detection of cardiac allograft rejection and response to immunosuppressive therapy with peripheral blood gene expression. Circulation 110, 3815–3821. Ishii, M., Hashimoto, S.,Tsutsumi, S.,Wada,Y., Matsushima, K., Kodama,T. and Aburatani, H. (2000). Direct comparison of GeneChip and SAGE on the quantitative accuracy in transcript profiling analysis. Genomics 68, 136–143.
References
Jackson, R.J. and Standart, N. (2007). . How do microRNAs regulate gene expression? Sci STKE. re1 Jongeneel, C.V., Iseli, C., Stevenson, B.J., Riggins, G.J., Lal, A., Mackay, A., Harris, R.A., O’Hare, M.J., Neville, A.M., Simpson, A.J. et al. (2003). Comprehensive sampling of gene expression in human cell lines with massively parallel signature sequencing. Proc Natl Acad Sci USA 100, 4702–4705. Kendziorski, C., Irizarry, R.A., Chen, K.S., Haag, J.D. and Gould, M.N. (2005). On the utility of pooling biological samples in microarray experiments. Proc Natl Acad Sci USA 102, 4252–4257. Kendziorski, C.M., Newton, M.A., Lan, H. and Gould, M.N. (2003). On parametric empirical Bayes methods for comparing multiple groups using replicated gene expression profiles. Stat Med 22, 3899–3914. Kim, J.B., Porreca, G.J., Song, L., Greenway, S.C., Gorham, J.M., Church, G.M., Seidman, C.E. and Seidman, J.G. (2007a). Polony multiplex analysis of gene expression (PMAGE) in mouse hypertrophic cardiomyopathy. Science 316, 1481–1484. Kittleson, M.M. and Hare, J.M. (2005). Molecular signature analysis: The potential of gene expression analysis in cardiomyopathy. Future Cardiol. Kittleson, M.M., Ye, S.Q., Irizarry, R.A., Minhas, K.M., Edness, G., Conte, J.V., Parmigiani, G., Miller, L.W., Chen, Y., Hall, J.L. et al. (2004). Identification of a gene expression profile that differentiates between ischemic and nonischemic cardiomyopathy. Circulation 110, 3444–3451. Lapointe, J., Li, C., Higgins, J.P., van de, R.M., Bair, E., Montgomery, K., Ferrari, M., Egevad, L., Rayford, W., Bergerheim, U. et al. (2004). Gene expression profiling identifies clinically relevant subtypes of prostate cancer. Proc Natl Acad Sci USA 101, 811–816. Lash, A.E., Tolstoshev, C.M., Wagner, L., Schuler, G.D., Strausberg, R.L., Riggins, G.J. and Altschul, S.F. (2000a). SAGEmap: A public gene expression resource. Genome Res 10, 1051–1060. Liew, C.C. (2005). Expressed genome molecular signatures of heart failure. Clin Chem Lab Med 43, 462–469. Liu, E.T. and Karuturi, K.R. (2004). Microarrays and clinical investigations. N Engl J Med 350, 1595–1597. Lonnstedt, I. and Speed, T.P. (2002). Replicated microarray data. Statistical Sinica 12, 31–46. Lossos, I.S., Czerwinski, D.K., Alizadeh, A.A., Wechser, M.A., Tibshirani, R., Botstein, D. and Levy, R. (2004). Prediction of survival in diffuse large-B-cell lymphoma based on the expression of six genes. N Engl J Med 350, 1828–1837. Ma, J. and Liew, C.C. (2003). Gene profiling identifies secreted protein transcripts from peripheral blood cells in coronary artery disease. J Mol Cell Cardiol 35, 993–998. Margulies, K.B., Matiwala, S., Cornejo, C., Olsen, H., Craven, W.A. and Bednarik, D. (2005a). Mixed messages. Transcription patterns in failing and recovering human myocardium. Circ Res, 1–8. Morello, F., de Bruin, T.W., Rotter, J.I., Pratt, R.E., van der Kallen, C.J., Hladik, G.A., Dzau, V.J., Liew, C.C. and Chen, Y.D. (2004). Differential gene expression of blood-derived cell lines in familial combined hyperlipidemia. Arterioscler Thromb Vasc Biol 24, 2149–2154. Pomeroy, S.L., Tamayo, P., Gaasenbeek, M., Sturla, L.M., Angelo, M., McLaughlin, M.E., Kim, J.Y., Goumnerova, L.C., Black, P.M., Lau, C. et al. (2002). Prediction of central nervous system embryonal tumour outcome based on gene expression. Nature 415, 436–442.
■
155
Prober, J.M., Trainor, G.L., Dam, R.J., Hobbs, F.W., Robertson, C.W., Zagursky, R.J., Cocuzza, A.J., Jensen, M.A. and Baumeister, K. (1987). A system for rapid DNA sequencing with fluorescent chain-terminating dideoxynucleotides. Science 238, 336–341. Ritchie, M.E., Diyagama, D., Neilson, J., van Laar, R., Dobrovic, A., Holloway, A. and Smyth, G.K. (2006). Empirical array quality weights in the analysis of microarray data. BMC Bioinformatics 7, 261. Rogers, Y.H. and Venter, J.C. (2005). Genomics: Massively parallel sequencing. Nature 437, 326–327. Rosenwald, A., Wright, G., Chan, W.C., Connors, J.M., Campo, E., Fisher, R.I., Gascoyne, R.D., Muller-Hermelink, H.K., Smeland, E.B., Giltnane, J.M. et al. (2002). The use of molecular profiling to predict survival after chemotherapy for diffuse largeB-cell lymphoma. N Engl J Med 346, 1937–1947. Salvado, C.S., Trounson, A.O. and Cram, D.S. (2004). Towards preimplantation diagnosis of cystic fibrosis using microarrays. Reprod Biomed Online 8, 107–114. Sanger, F., Nicklen, S. and Coulson, A.R. (1977). DNA sequencing with chain-terminating inhibitors. Proc Natl Acad Sci USA 74, 5463–5467. Seo, D., Wang, T., Dressman, H., Herderick, E.E., Iversen, E.S., Dong, C., Vata, K., Milano, C.A., Rigat, F., Pittman, J. et al. (2004). Gene expression phenotypes of atherosclerosis. Arterioscler Thromb Vasc Biol 24, 1922–1927. Simon, R., Radmacher, M.D., Dobbin, K. and McShane, L.M. (2003b). Pitfalls in the use of DNA microarray data for diagnostic and prognostic classification. J Natl Cancer Inst 95, 14–18. Smyth, G.K. (2004). Linear models and empirical Bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Mol Biol 3, 1–27. Storey, J.D. (2002). A direct approach to false discovery rates. J Roy Stat Soc B, 479–498. Tan, P.K., Downey, T.J., Spitznagel, E.L., Jr., Xu, P., Fu, D., Dimitrov, D.S., Lempicki, R.A., Raaka, B.M. and Cam, M.C. (2003). Evaluation of gene expression measurements from commercial microarray platforms. Nucleic Acids Res 31, 5676–5684. Tibshirani, R., Hastie, T., Narasimhan, B. and Chu, G. (2002). Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci USA 99, 6567–6572. Tusher, V.G., Tibshirani, R. and Chu, G. (2001). Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 98, 5116–5121. Twine, N.C., Stover, J.A., Marshall, B., Dukart, G., Hidalgo, M., Stadler, W., Logan, T., Dutcher, J., Hudes, G., Dorner, A.J. et al. (2003). Disease-associated expression profiles in peripheral blood mononuclear cells from patients with advanced renal cell carcinoma. Cancer Res 63, 6069–6075. Valk, P.J., Verhaak, R.G., Beijen, M.A., Erpelinck, C.A., Barjesteh van Waalwijk van Doorn-Khosrovani, , Boer, J.M., Beverloo, H.B., Moorhouse, M.J., van der Spek, P.J., Lowenberg, B. and Delwel, R. (2004). Prognostically useful gene-expression profiles in acute myeloid leukemia. N Engl J Med 350, 1617–1628. van de Vijver, M.J., He, Y.D., van’t Veer, L.J., Dai, H., Hart, A.A., Voskuil, D.W., Schreiber, G.J., Peterse, J.L., Roberts, C., Marton, M.J. et al. (2002). A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347, 1999–2009. van, Rooij E., Sutherland, L.B., Qi, X., Richardson, J.A., Hill, J. and Olson, E.N. (2007b). Control of stress-dependent cardiac growth and gene expression by a microRNA. Science 316, 575–579.
156
CHAPTER 12
■
Transcriptomics: Translation of Global Expression Analysis to Genomic Medicine
RECOMMENDED RESOURCES Books Pevsner J. (2003). Bioinformatics and Functional Genomics. John Wiley & Sons, Inc. Gentleman, R., Carey, V.J., Huber, W., Irizarry, R.A., Dudoit, S. (2005). Bioinformatics and Computational Biology Solutions Using R and Bioconductor. Springer ScienceBusiness Media, Inc.
Public Online Databases Gene Expression Omnibus: http://www.ncbi.nlm.nih.gov/geo/ MGED society: http://www.mged.org/Workgroups/MIAME/miame.html European Bioinformatics Institute: http://www.ebi.ac.uk/ Swiss Institute of Bioinformatics: http://www.isb-sib.ch/ DBGET Database: http://www.genome.jp/dbget/dbget.links.html Weizmann Institute of Science, Gene Cards: http://www.genecards.org/ index.shtml
Kyoto Encyclopedia of Genes and Genomes: http://www.genome. jp/kegg/
Download Sites for Free Software Bioconductor: open source software: http://www.bioconductor.org/ Significance Analysis of Microarrays: http://www-stat.stanford.edu/tibs/ SAM/ Prediction Analysis of Microarrays: http://www-stat.stanford.edu/tibs/ PAM/ The R Project for Statistical Computing: http://www.r-project.org/ DNA chip Analyzer (dChip): http://biosun1.harvard.edu/complab/dchip/ install.htm Gene Pattern: http://www.broad.mit.edu/cancer/software/genepattern/
CHAPTER
13 DNA Microarrays in Biological Discovery and Patient Care Andrew J. Yee and Sridhar Ramaswamy
INTRODUCTION While “expression profiling” is now synonymous with DNA microarray technology, expression profiling dates back to the 1970s with techniques such as northern blotting for measuring RNA expression (Alwine et al., 1977). However, only one gene at a time could be queried with these modalities. Later developments enabled analysis of multiple genes (Liang and Pardee, 2003), including differential hybridization (Sargent, 1987), subtractive hybridization (Zimmermann et al., 1980), differential display (Liang and Pardee, 1992), and serial analysis of gene expression (SAGE) (Velculescu et al., 1995). However, these approaches required a significant amount of starting RNA and investment in labor and technical expertise, and were difficult to scale up. DNA microarrays revolutionized the approach to gene expression profiling. Compared to prior methods, DNA microarrays were both dramatically high throughput and less cumbersome. It should be noted that the concept behind the microarray format was not new, as microarrays were originally developed as a technique for large-scale DNA mapping and sequencing (Hoheisel, 2006). However, changing the support surface from a porous membrane to a solid surface, such as glass, afforded significant improvements (Schena et al., 1998) by increasing reaction kinetics (Southern et al., 1999) and reducing background noise (Cheung et al., 1999).
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
As an illustration of approaches to define and develop applications of the transcriptome measured in biological samples (as reviewed in Chapter 12), here we present specific issues related to the use of DNA microarrays, using applications especially in the context of cancer to illustrate the uses and challenges of gene expression profiling in genomic and personalized medicine.
MICROARRAY TECHNOLOGY Microarray Platforms Although the technology for DNA microarrays continues to evolve, the commonly used arrays can be divided into two main categories, as described in detail in Chapter 12: (i) oligonucleotide and (ii) spotted or complementary DNA (cDNA) microarrays (see Figure 13.1). Over the years, commercial oligonucleotide platforms have gained in popularity over “homemade” spotted cDNA microarrays as costs have fallen considerably, to the point where commercial microarrays are the de facto platform. At the time of this writing, the dominant platforms are produced by Affymetrix and Agilent. For additional details, the reader is referred to a review by Hardiman (2004) and also to reviews collected in a series of Nature Genetics supplements (Bowtell, 1999; Cheung
Copyright © 2009, Elsevier Inc. All rights reserved. 157
158
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
(a) M1 X X X X
M2 Shine light
M1 X X
X X Couple
X X A A
Repeat
Microarray (b)
Touch surface
Move pins
Repeat
Microarray
Figure 13.1 Comparison of microarray technologies. (a) Photolithography (as used by Affymetrix): a glass wafer modified with photolabile protecting groups (X) is selectively activated for DNA synthesis by shining light through a photomask (M1). The wafer is then flooded with a photoprotected DNA base (A–X), resulting in spatially defined coupling on the chip surface. A second photomask (M2) is used to deprotect defined regions of the wafer. Repeated deprotection and coupling cycles result in a highly dense microarray. (b) Spotted arrays: the gene of interest is loaded into a spotting pin by capillary action, and a small volume is transferred to a solid surface by physical contact between the pin and the solid substrate. After the first spotting cycle, the pin is washed and a second sample is loaded and deposited to an adjacent location. Robotic control systems and multiplexed printheads allow automated microarray fabrication. Adapted from Schena et al. (1998). (Reprinted with permission from Elsevier).
et al., 1999; Churchill, 2002; Duggan et al., 1999; Holloway et al., 2002; Lipshutz et al., 1999; Southern et al., 1999). Oligonucleotide Arrays Oligonucleotide microarrays are created either by in situ synthesis or deposition of presynthesized oligonucleotides ranging in size from 25- to 60-mers. Oligonucleotides can be synthesized directly in situ using photolithography techniques adapted from the microelectronics industry. The Affymetrix GeneChip platform is the primary example of this method, and because of its popularity, merits further discussion. Emerging from research carried out in the early 1990s (Fodor et al., 1991), this technology combines combinatorial, solid phase DNA synthetic chemistry with the benefits of photolithography. Successive photolithographic masks are used to select regions of the chip surface for exposure to light. Exposure to light deprotects photolabile groups on the nascent oligonucleotides undergoing synthesis. The light-mediated deprotection then allows these specified regions to couple with activated nucleoside monomers using standard DNA phosphoramidite synthetic chemistry. Each cycle thus extends the oligonucleotide by one base (see Figure 13.1). Over the past 15 years, the feature size of each oligonucleotide probe or feature has been reduced 10-fold, resulting in a 1.6 cm2 chip with 1.3 million unique features (multiple features are used to measure the expression of a particular gene). The detection sensitivity of these microarrays has been estimated at
1 in 300,000 RNAs (Lockhart et al., 1996). A practical limitation is the efficiency of each synthesis step – about 90–95%. This places a ceiling on in situ synthesis of an oligonucleotide to 25-mers. To control for non-specific binding, the probes on the Affymetrix chip are designed in pairs. One sequence is the exact complement of the target transcript, and the other paired sequence differs from the complement typically by one base pair near the middle of the probe. Difference in signal between the two probes can be used to control for non-specific binding and background contributions (Pease et al., 1994); 22 probes are routinely used for each expression measurement. Compared to cDNA arrays (see below), oligonucleotide arrays offer greater specificity and can distinguish single nucleotide polymorphisms (SNPs) and splice variants (Guo et al., 1994); the same technology is used in arrays for measuring DNA copy number and SNP profiling. The current generation of Affymetrix microarrays can measure the expression of over 47,000 transcripts. Hybridization is detected by using a confocal laser scanning microscope to image the intensity of one color, fluorescently labeled samples. First, experimental mRNA is enzymatically amplified (see below for details of RNA amplification) and then labeled with biotin through the partial substitution of UTP and CTP with biotin-11-UTP and biotin-11-CTP (see Chapter 12). The labeled mRNA is then hybridized to the microarray and detected by the binding of a fluorescent compound (streptavidin– phycoerythrin) to the biotin-labeled samples. Because of the
Microarray Technology
robustness and consistency of the manufacturing process, the single-color readout provides an absolute quantitation of mRNA populations (Ishii et al., 2000). In addition, experiments can be compared with samples hybridized to different Affymetrix arrays. Oligonucleotide microarrays may also be fabricated by using ink-jet printing technology. Nucleotide monomers are printed onto the chip and coupled using phosphoramidite chemistry (Hughes et al., 2001). Agilent has adopted this technology to create a 60-mer oligonucleotide array. Compared to the 25-mer photolithographic chip, a 60-mer array might offer increased sensitivity per probe. In contrast to the Affymetrix platform, there is only one probe per gene in this platform. The longer probes are also more tolerant of mismatches, improving analysis of polymorphic regions. Spotted, Complementary DNA Microarrays The first DNA microarrays were developed at Stanford University and based on cDNA samples spotted by a robotic arrayer at defined locations onto glass slides (Schena et al., 1995). These cDNA samples were generally PCR products generated from cDNA libraries or clone collections. The major benefit of spotted arrays is that they can be constructed in-house and tailored to the user’s needs. The features are generally 100–300 m in size and spaced at about the same distance. About 30,000 cDNAs can be spotted onto the surface of a microscope slide (Schulze and Downward, 2001). In the initial report, the sensitivity was estimated at 1 in 50,000 RNAs starting with 5 g of total RNA (Schena et al., 1995). While spotted arrays offer customizability, they are better suited for smaller-scale experiments, since from a practical perspective managing large clone libraries and creating high-quality arrays consistently can be arduous. In the past, an advantage of spotted arrays was that a priori knowledge of the sequence being spotted was not necessary. However, this is less relevant now that complete sequence information for many experimental organisms and humans is readily accessible. Gene expression using spotted microarrays is determined as a ratio of two samples rather than as an absolute measurement, owing to technical limitations and variability in spot size and probe concentration. Generally, cDNAs derived from experimental and reference RNA samples are labeled with deoxyribonucleotides coupled to different fluorophores (e.g., with Cy3 or Cy5) during first-strand cDNA synthesis. The two different populations are then competitively hybridized to the same array and imaged. While this method accurately compares expression levels between samples, comparing data from different platforms can be difficult, often requiring complex normalization algorithms (Dudley et al., 2002). Sample Preparation Tissue Sampling For biological applications, tissue sampling is an important variable. For example, tumors are heterogeneous mixtures of different cell types, including malignant cells with varying degrees of
■
159
differentiation, stromal elements, blood vessels, and inflammatory cells. Profiling of this whole sample may dilute out the “signal” profile of cancer cells with extraneous expression “noise” from surrounding tissues. Newer techniques such as laser capture microdissection (LCM) address the challenges of tissue heterogeneity by offering an approach for isolating homogeneous, individual cell populations (Emmert-Buck et al., 1996). While there are other approaches, such as fluorescence-activated cell sorting (FACS) and manual microdissection using micromanipulators (Emmert-Buck et al., 1994), LCM is faster, more versatile, and can be applied to solid tumors directly without the need for disaggregation into a cell suspension. Sgroi et al. (1999) first showed that it was feasible to use LCM to isolate subpopulations of breast cancer cells for expression profiling. Coupled with LCM has been the evolution of methodologies for purifying adequate amounts of high-quality RNA for expression profiling (see below). However, a theoretical limitation of focusing only on individual tissue components such as cancer cells relates to the growing appreciation that interactions with the tumor microenvironment (e.g., with stroma, endothelial, and immune cells) play a critical role in tumor progression. Sampling of “non” cancer cell components thus may be necessary for a profile to adequately reflect in vivo biology. An additional consideration is that the actual surgical manipulation itself may affect the expression profiling. Lin et al. (2006) compared expression profiles of prostate cancer from biopsies obtained in situ to biopsies obtained from prostates that were surgically removed. They found that 1.5% of measurable genes were altered as a result of the surgery, including transcripts for acute phase response proteins and regulators of cell proliferation. RNA Isolation Traditionally, microarray experiments have required roughly 50–200 μg of total RNA (Duggan et al., 1999). However, these quantities are not always easily obtainable, especially with techniques such as LCM. Protocols therefore have been developed for amplifying RNA from much smaller quantities, perhaps even down to the level of a single cell. For example, a single cell contains an estimated 0.1 pg of mRNA or 10 pg of total RNA, requiring an expansion of 108–109 fold in order to yield adequate amounts for profiling. Two commonly used amplification techniques include in vivo transcription and PCR amplification. In vivo transcription is a linear amplification approach, first described by Van Gelder et al. (1990). Messenger RNA is converted to double stranded cDNA using an oligo(dT) primer attached to a T7 promoter. T7 polymerase is then used to transcribe complementary copies of the cDNA population. One round of amplification can amplify the starting amount of mRNA by 1000-fold. A second or a third round of amplification can further increase the yield (Baugh et al., 2001). A second approach takes advantage of PCR amplification. PCRbased approaches have several advantages over linear amplification, including speed, cost, and theoretically unlimited degree of amplification. While it has been assumed that exponential amplification may lead to a biased representation of the starting
160
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
sample, results from one group suggest that this may not be as problematic as previously suggested (Iscove et al., 2002). Standardization of Data Critical to the success of data analysis is standardization of microarray data and annotation of the genes that comprise the microarrays. In 1999, the Microarray Gene Expression Database (MGED) group was created to establish guidelines for the format of microarray data. This led to the development of the Minimal Information About a Microarray Experiment (MIAME) standard (Brazma et al., 2001; Stoeckert et al., 2002). Most major scientific publications require compliance with MIAME standards and submission of data to public repositories. An additional key factor is the functional annotation of the genes being studied. Biological “knowledge” is intricate, hierarchical, and interrelated. Ontologies are a formalized, standardized means of coding biological knowledge in which concepts are systematically described by both their meaning and their relationship to each other (Bard and Rhee, 2004). Adequate ontologies can thus serve as a common anchor point for systematic integration of data from various microarray sources. An example of commonly used ontology is one devised by the Gene Ontology Consortium (Harris et al., 2004).
DATA ANALYSIS Gene expression studies pose many challenges for data organization, storage, and analysis (see Chapter 17). In concert with the development of microarray technology has been the evolution of sophisticated statistical methods for analyzing and interpreting highly dense microarray information (Ermolaeva et al., 1998; Quackenbush, 2001). These data can be one’s own experimental data or existing data from other investigators’ experiments, made available through public databases. Microarray data may be so rich in information that important findings may not be recognized initially, and only better appreciated in retrospect when re-analyzed or combined with newer data. Data analysis methodology is discussed further in the section “Informatic and Computational Platforms for Genomic Medicine” in Part I of this book and has been reviewed also by Allison et al. (2006). Unsupervised Learning The computational analysis of gene expression data has largely centered on two approaches: unsupervised and supervised learning (see Figure 13.2). Unsupervised methods include hierarchical and k-means clustering (Gaasterland and Bekiranov, 2000). The goal of clustering is to organize genes and samples clusters based on relative similarity (D’Haeseleer, 2005; Eisen et al., 1998). A hierarchy of clusters can be created, which may be visualized as tree diagrams or dendrograms. Unsupervised approaches have the advantage of being unbiased and allow for the identification of structure in a complex dataset without making any a priori assumptions. In fact, many of the initial microarray studies on cancer samples utilized unsupervised classification, for example,
lymphoma (Alizadeh et al., 2000) and breast cancer (Perou et al., 2000). However, because many different relationships are possible in a complex dataset, the predominant structure uncovered by clustering may not be clearly related to the most clinically or biologically interesting aspect of a dataset. Supervised Learning In contrast, supervised learning starts off with a known class distinction and uses this information upfront to select genes whose expression is best associated with this class distinction in a “training dataset” (Golub et al., 1999). This model is then applied to an independent “test” dataset to validate the accuracy of selected gene expression features in classification. Numerous supervised learning algorithms have been applied to gene expression datasets, such as k-nearest neighbors, neural networks, or support vector machines (Brown et al., 2000). The accuracy of supervised learning approaches heavily depends on the quality of the initial training data. Mining Several tools have been developed for distilling higher-level information from microarray data in order to better appreciate the “forest” among the innumerable “trees” of expression data. Two notable examples include modular analysis and Cancer Outlier Profile Analysis (COPA). Modular analysis examines the coordinated behavior of a set of genes in order to detect significant changes in modules of genes, rather than the individual genes themselves. This offers the ability to extract more “signal” from a given microarray experiment. Indeed, groups of gene expression may be more biologically interpretable and statistically robust than deciphering lists of individual genes. Mootha et al. (2003) used this approach to study changes in gene expression in the muscle of patients with type 2 diabetes. Their group introduced the gene set enrichment analysis (GSEA) and used the Kolmogorov– Smirnov test statistic, a type of non-parametric test statistic that does not make assumptions about population distribution. While the decrease in expression for individual genes was modest, around 20%, it was consistently decreased across a set of genes involved in oxidative phosphorylation, accounting for 89% of 106 genes. Thus, significant changes could be seen when looking at a group of genes, even when the expression of individual genes was not significantly different. GSEA has emerged as a useful tool for identifying biological processes from microarray data and has been refined into a robust method (Subramanian et al., 2005). Similarly, Segal et al. (2004) created a cancer-focused compendium of expression profiles from 26 studies. They were able to extract 456 modules and found that some modules were unique to a particular type of tumor whereas other modules were shared across a range of conditions, raising the possibility that these modules reflect common tumor progression mechanisms. For example, in acute lymphoblastic leukemia (ALL), a growthinhibitor module was specifically repressed. On the other hand, a bone osteoblastic module was found across a range of tumor types, suggesting a common mechanism for bone metastasis.
Applications
■
161
Supervised learning
Unsupervised learning Dataset Known classes Class A Class B
Unknown classes
Training set
Train model Cluster samples
Independent test set (“Unknowns”)
Assign class labels Apply model
Class A
Class B
Class A
Class discovery
Class B
Class prediction
Figure 13.2 Unsupervised versus supervised learning. Unsupervised learning: multiple tumor samples are clustered into groups based on overall similarity of their gene expression profiles. This approach is useful for discovering previously unappreciated relationships. Supervised learning: multiple tumor samples from different known classes are used to train a model capable of classifying unknown samples. This model is then applied to a test set for class label assignment. Adapted from Ramaswamy and Golub (2002). (Reprinted with permission from the American Society of Clinical Oncology).
The COPA method focuses on genes characterized by marked overexpression (Tomlins et al., 2005). This strategy is used in order to efficiently process large amounts of data. COPA was applied to the Oncomine database, a compilation of 132 gene expression datasets representing 10,486 microarray experiments (Rhodes et al., 2004). This approach was instrumental for discovering two transcription factors with outlier profiles, ERG and ETV1, and subsequently shed light on recurrent gene fusions of the 5 untranslated region of TMPRS22 to ERG or ETV1 in prostate cancer tissues. This study was notable for being the first study to identify recurrent gene re-arrangements in an epithelial cancer and suggested a novel mechanism for prostate cancer pathogenesis.
APPLICATIONS DNA microarrays provide a panoramic, quantitative overview of a sample’s expression output. The power of this is obvious, as biological processes generally result from the coordinated interaction of multiple genes.
Research with microarrays in humans can be divided into four categories: 1. 2. 3. 4.
normal tissue taxonomy; disease diagnosis and classification; disease prognostication; and dissection of biological mechanisms.
Since the lion’s share of research to date has been in cancer biology, in part due to the ready availability of tissue samples procured during routine diagnostic procedures, the following examples will draw largely from this field’s literature. Normal Tissue Taxonomy Microarrays provide a new way of approaching the same fundamental question posed by Linnaeus in the 18th century: how is A different from and similar to B? An interesting example of previously unknown molecular heterogeneity was illustrated with human fibroblasts (Chang et al., 2002). Here, the authors demonstrated that the major factor responsible for differences in gene expression was the site of origin of the fibroblast. This
162
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
suggests that a detailed positional memory is encoded in a fibroblast’s expression profile, and that different fibroblast subsets might be functionally distinct. Prior to the advent of microarray technology, this geographic variation was not appreciated. In addition to topographic location, the effect of aging on fibroblasts can also be measured and classified by expression profiling (Ly et al., 2000). In this study, using an Affymetrix oligonucleotide microarray, the expression profiles of fibroblasts isolated from young, middle-age, and old-age individuals and individuals with progeria were compared. The authors found consistent differences in the expression of genes between these groups in two functional classes: genes involved in cell cycle progression and genes involved in maintenance and remodeling of the extracellular matrix. Disease Diagnosis and Classification There have been a multitude of studies using expression profiling for disease diagnosis and classification, most notably in cancer. One seminal study used oligonucleotide microarrays to examine the expression of 6817 human genes in 72 patients with acute leukemia (Golub et al., 1999). Using unsupervised learning, gene expression was able to independently cluster leukemia samples into the known subsets of acute myelogenous leukemia (AML) and ALL (see Figure 13.3). Then, using supervised learning, gene sets that were differentially expressed in AML and ALL were used to correctly classify a group of unknown samples into the correct categories. Significantly, many markers that were both known, such as myeloperoxidase and terminal transferase, and unknown were useful for making this distinction. Although the difference between AML and ALL is often not difficult using modern histopathology, this study provided the first evidence that tumor expression profiles can be used for cancer classification. Tumor cell lines can also be classified molecularly through expression profiling (Ross et al., 2000). The National Cancer Institute’s Developmental Therapeutics Program bank of 60 cancer cell lines (NCI60) was profiled using a spotted cDNA array of 9703 cDNAs. The gene expression patterns observed in these cancer cell lines corresponded to the tissue of origin, which is remarkable since these cell lines have been selected for survival in the host and later in tissue culture. This illustrates a theme that the tissue of origin plays a major role in determining the expression profile. Similarly, a molecular taxonomy of tumor samples from 14 different tumor classes was created using a supervised learning algorithm (Ramaswamy et al., 2001). Some of these samples were from metastatic sites, and interestingly, the majority of them could be categorized from the classification information derived from the primary site, again demonstrating the importance of the tissue of origin in disease. Breast cancer has been intensively profiled. Perou et al. (2000) reported a molecular classification system based on the expression patterns of 65 breast adenocarcinoma specimens from 42 individuals using a spotted cDNA array. Using an unsupervised method, these breast tumors were categorized into four different subtypes based on their patterns of gene expression.
One clinically important subtype was already known (Erb-B2 or Her2/neu overexpressing cancers), and three others were previously unknown: estrogen receptor-positive/luminal-like cancers, basal-like cancers, and normal breast. In addition to this classification system, the authors were able to show the importance of cellular lineage in determining the expression profile. This was illustrated by examining the variation in expression between primary tumors that were biopsied before and after a course of chemotherapy. They also looked at the profiles of two primary lymph node metastasis pairs. The authors found that the paired samples in the same patient were significantly more related to each other than to tumors from other patients, despite intervening chemotherapy or metastatic evolution. These initial observations were subsequently validated in a larger dataset (Sorlie et al., 2001, 2003). Importantly, these molecularly defined subtypes were clinically relevant, as patients with basal-like tumors had a significantly worse prognosis compared to patients with other breast cancer subtypes. Outside of the realm of cancer, cDNA microarray technology has proved useful for exploring a broad range of diseases such as inflammatory diseases like rheumatoid arthritis and Crohn’s disease (Heller et al., 1997), schizophrenia (Mirnics et al., 2000), multiple sclerosis (Lock et al., 2002), and differentiating between ischemic and non-ischemic cardiomyopathy (Kittleson et al., 2004). Microarrays have also been helpful for re-examining basic questions such as aging, as discussed previously (Ly et al., 2000). Disease Prognosis A challenge in medicine and especially in oncology is prediction– forecasting which patients will and will not benefit from therapy, particularly when the therapy, such as chemotherapy, has significant risks. DNA microarrays offer the possibility of radically transforming medicine from a traditional one-size-fits-all approach to individualized care. Intuitively, more information, when rich in content and appropriately analyzed, should lead to better prediction models. However, this has not always been borne out with various histopathologic tumor markers (e.g., breast and colon cancer), despite much initial excitement when these markers first became available (Bast et al., 2001). In contrast, there have been several microarray experiments involving different types of cancers that have found that the expression signature can serve as a robust biomarker, better than currently used criteria, for predicting prognosis. A common approach is to identify an expression profile that can subclassify a tumor type and then find correlations between this subclassification and prognosis. The first such application was described by Alizadeh et al. (2000) in diffuse large B cell lymphoma (DLBCL). These investigators constructed the Lymphochip, a specialized spotted cDNA microarray enriched for genes preferentially expressed in lymphoid cells and for genes with known roles in immunology or cancer. Using this microarray, hierarchical clustering revealed two molecularly distinct forms of DLBCL: germinal center B-like DLBCL and activated B-like DLBCL. These subtypes were not
Applications
ALL
■
163
AML C-myb (U22376) Proteasome iota (X59417) MB-1 (U05259) Cyclin D3 (M92287) Myosin light chain (M31211) RbAp48 (X74262) SNF2 (D26156) HkrT-1 (S50223) E2A (M31523) Inducible protein (L47738) Dynein light chain (U32944) Topoisomerase II β (Z15115) IRF2 (X15949) TFIIEβ (X63469) Acyl-Coenzyme A dehydrogenase (M91432) SNF2 (U29175) (Ca2)-ATPase (Z69881) SRP9 (U20998) MCM3 (D38073) Deoxyhypusine synthase (U26266) Op 18 (M31303) Rabaptin-5 (Y08612) Heterochromatin protein p25 (U35451) IL-7 receptor (M29696) Adenosine deaminase (M13792) Fumarylacetoacetate (M55150) Zyxin (X95735) LTC4 synthase (U50136) LYN (M16038) HoxA9 (U82759) CD33 (M23197) Adipsin (M84526) Leptin receptor (Y12670) Cystatin C (M27891) Proteoglycan 1 (X17042) IL-8 precursor (Y00787) Azurocidin (M96326) p62 (U46751) CyP3 (M80254) MCL1 (L08246) ATPase (M62762) IL-8 (M28130) Cathepsin D (M63138) Lectin (M57710) MAD-3 (M69043) CD11c (M81695) Ebp72 (X85116) Lysozyme (M19045) Properdin (M83652) Catalase (X04085)
3 Low
2.5
2
1.5
1
0.5
0
0.5
Normalized expression
1
1.5
2
2.5
3 High
Figure 13.3 Gene expression profiling in acute leukemia. The 50 genes most highly correlated with the distinction between ALL and AML are shown. Each row corresponds to a gene, and each column corresponds to expression levels in different samples. Expression levels greater than the mean are shaded in red, and those below the mean are shaded in blue. The scale indicates SDs above or below the mean. The top panel shows genes highly expressed in ALL; the bottom panel shows genes more highly expressed in AML. Adapted from Golub et al. (1999). (Reprinted with permission from AAAS).
164
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
(a)
Gene-expression profiling
(b)
1.0
1.0 Good signature
0.8
0.6
0.4
Poor signature p 0.001
0.2
Probability of remaining metastasis-free
Probability of remaining metastasis-free
St Gallen criteria
0.8 Low risk 0.6 High risk 0.4
p 0.05
0.2
0.0
0.0 0
2
4
6 8 Years
10
0
12
2
4
22
8 6 Years
10
12
No. at risk
No. at risk Good signature
60 57 54 45 31 22 12
Low risk
22
Poor signature
91 72 55 41 26 17 9
High risk
129 107 88 69 48 34 19
21 17
9
5
2
Figure 13.4 Kaplan–Meier analysis of the probability that patients would remain free of distant metastases among 151 patients with lymph node-negative breast cancer by gene expression profiling (a) and by St Gallen clinical criteria (Goldhirsch et al., 2001) (b). Adapted from Van de Vijver et al. (2002). (Copyright © 2002, Massachusetts Medical Society. All rights reserved).
appreciated using traditional histopathology and immunohistochemical techniques, yet nevertheless have clinical significance as patients with germinal center B-like DLBCL had significantly better overall survival when treated with standard chemotherapy. Indeed, as observed with other initial studies in cancer classification, cell lineage was an important determinant of the molecular phenotype. Ultimately, it remains to be seen whether or not molecular phenotyping will supplant traditional diagnostic tools, and this remains an ongoing area of research. Similar predictive power has been exemplified in microarray studies of breast cancer. Traditional factors for predicting a patient’s outcome, such as tumor size, lymph node status, and estrogen receptor status, are imperfect. Van’t Veer et al. (2002) used an oligonucleotide microarray to analyze primary tumors from 117 patients with early-stage breast cancer and no lymph node involvement, who received minimal treatment after surgery, to identify a 70-gene expression signature that strongly predicted metastasis. In a subsequent validation study on 295 patients (which included the original cohort of 117 patients), this signature outperformed all currently used clinical predictors and more accurately identified patients at risk for distant metastasis (van de Vijver et al., 2002) (see Figure 13.4). For example, the profile could assign more patients into a low-risk group for metastasis (40%) compared to traditional clinical criteria (15%). These initial findings were subsequently confirmed in an independent group of patients from five European centers (Buyse et al., 2006). Taken together, these results suggested that primary tumor gene expression profiles can be used to molecularly stratify cancer patients according to risk of progression and treatment outcome.
Molecular Biology Expression profiling has also been a new approach for discovering novel roles for both individual genes and biological processes. In addition, expression profiling also reveals the power of performing experiments in silico and mining data from previous experiments to integrate existing information with new insights. For example, Clark et al. (2000) used expression profiling to identify RhoC, a small GTPase, as a potential mediator of melanoma metastasis. Using a mouse model to select for metastatic variants of the human melanoma cell line A375, the investigators were able to compare the expression profiles of the parental cell line and the metastasis. RhoC was one of several genes, which was overexpressed in the metastatic sublines. Functional studies with RhoC confirmed its role in this model system. Lamb et al. (2003) also used integrative methods to describe a novel mechanism for the action of cyclin D1, a cell cycle regulator implicated in cancer progression. After first developing an expression signature for cyclin D1 action through retroviral transduction and gene expression profiling, they then used this profile to mine a compendium of human tumor gene expression data to demonstrate the in vivo relevance of this signature, subsequently leading to the identification of C/EBPb as novel downstream effector of cyclin D1. Cellular Biology Expression profiling has also been able to uncover intriguing similarities between cancer and certain physiological processes such as wound repair and the response to hypoxia. Leveraging on the similarities between wound healing and cancer (Dvorak,
Future Directions
1986), Chang et al. (2004) derived an expression signature of the fibroblast response to serum in tissue culture, noting that serum is encountered in vivo at sites of tissue injury. Using this signature, they identified corresponding expression signatures in a variety of cancers that were similar to active wounds. Furthermore, the molecular features of wound healing also predicted increased risk of metastasis and death in several cancers such as breast, lung, and stomach cancer. The authors validated their findings with microarray data from the Dutch cohort of breast cancer patients (Chang et al., 2005; van de Vijver et al., 2002). Overall survival and distant metastasis-free survival were markedly diminished in patients whose tumors expressed the wound-response signature compared to tumors that did not express this signature. In addition to the serum response, using a similar strategy, the gene expression response to hypoxia in cell culture was also identified in several human cancers, particularly breast and ovarian cancer (Chi et al., 2006). Interestingly, this phenomenon was independent of the wound-response signature. Signatures identified in vitro, that are associated with activation of specific oncogenic pathways, have also been used to identify similar patterns of pathway dysregulation in human tumors (Bild et al., 2006). Interestingly, these signatures might be clinically relevant, as they appear to correlate with survival and treatment outcome in patients with solid tumors. Expression profiling has also yielded provocative insights into the biology of cancer metastasis (Ramaswamy et al., 2003). By comparing expression profiles of various primary and metastatic tumors from a wide variety of sites, the authors discovered a 17-gene-expression signature that identified primary cancers with a greater likelihood of metastasis. Interestingly, many of these “metastasis signature” genes were also genes commonly expressed by cancer-associated fibroblasts and other stromal cells. These findings suggested for the first time that generic molecular determinants of metastasis might exist, and that these determinants might arise very early in tumorigenesis. This observation suggests a model for the biology of metastasis that contrasts with traditional dogma that molecular changes leading to metastasis arise only later in the natural history of a tumor.
LIMITATIONS AND CHALLENGES While microarray technology continues to mature, as with any technology, expression profiles must be interpreted in the context of broader biological knowledge. For example, one assumption often made is that gene expression correlates well with protein quantity and therefore protein activity. This view does not take into account additional variables such as mRNA stability, protein degradation, or post-translational modifications that ultimately determine protein activity. Another supposition is that mRNA expression levels correlate well with biological activity. In contrast, small differences in expression may have a dramatic biological effect. For example, the expression level of transcription factors is often quite low even in tissues where these factors play active roles. Finally, expression profiles are not fixed; they
■
165
may vary dramatically in different physiological contexts (King and Sinha, 2001). Validation with Non-microarray Technologies Gene expression data generated from microarray experiments can be credentialed using other methods of expression measurement such as northern blotting (Taniguchi et al., 2001), real-time PCR (Dallas et al., 2005), or other transcriptomic approaches described in Chapter 12. These studies and others demonstrate that genes found to be expressed with current generation microarrays can be consistently verified. Microarray data can also be verified at the protein expression level. A particularly useful high-throughput technology for validation using immunohistochemistry involves tissue microarrays (Hans et al., 2004). Reproducibility of Microarray Data An ongoing concern since the inception of the DNA microarray has been its technical reproducibility (Marshall, 2004). For a general review of these initial concerns, the interested reader is referred to a review by Chuaqui et al. (2002). While initial studies raised concerns about the reproducibility of microarray data (e.g., Tan et al. [2003]), subsequent studies have found that reproducibility is better than initially described (Bammler et al., 2005; Irizarry et al., 2005; Larkin et al., 2005). Recently, the MicroArray Quality Control (MAQC) project involved 137 participants from 51 academic, government, and commercial institutions to assess the performance of seven microarray platforms of two commercially available RNA samples (Shi et al., 2006). Reassuringly, the MAQC project showed high intra-platform consistency across test sites as well as a high level of interplatform concordance in terms of identifying differentially expressed genes in a common set of biological samples.
FUTURE DIRECTIONS Newer Platform Technologies Alternatives to conventional microarray platforms are emerging. One notable example is a bead-based platform developed by Illumina, the BeadChip. This uses randomly assembled arrays of 50-mer coated 3-m beads in order to create an extremely dense platform that is denser than spotted arrays or photolithographically made arrays (Gunderson et al., 2004). Whereas in conventional arrays, the sequence and location of each probe is known beforehand, the identity (and the corresponding probe sequence) of the beads is decoded after deposition. A DNA-based decoding algorithm is used to identify each bead. Compared to conventional platforms, the main advantage of this platform is lower per-sample cost. A different type of platform departing from conventional technology is the CombiMatrix platform. This platform synthesizes oligonucleotides in situ on a semiconductor chip using routine phosphoramidite chemistry under electrochemical control
166
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
(Liu et al., 2006). Each feature on the chip is digitally addressed and controlled. When activated, an electrochemical reaction generating acid occurs, ultimately leading to oligonucleotide synthesis in situ. The main advantage of this technology is flexibility. Custom microarrays can be created on demand with less labor, whereas technologies with “fixed” content such as the Affymetrix chip require a significant investment of resources to create a new chip (e.g., creation of a new photolithographic mask). Formalin-Fixed, Paraffin-Embedded Samples Traditionally, obtaining RNA from human tissues for expression profiling has required that specimens be snap-frozen in liquid nitrogen within a half hour of resection and stored at 80°C or colder to minimize RNA degradation. Samples need to be processed quickly because changes in some mRNA species have been noted even a few minutes after manipulation and devascularization of a tissue (Huang et al., 2001). However, clinical specimens are generally fixed in formalin and embedded in paraffin for routine histopathologic analysis, which causes significant RNA degradation. Only a fraction of the RNA from fixed specimens, roughly 3%, is intact enough for cDNA synthesis, posing a major barrier to obtaining expression profiles from clinical specimens (Godfrey et al., 2000; Masuda et al., 1999). A method to maneuver around this degradation obstacle is the cDNA-mediated, annealing, selection, extension, and ligation (DASL) assay. This assay is a technology that combines the advantages of microarray-based gene expression profiling with multiplexed quantitative PCR (Fan et al., 2004). An upstream and downstream pair of oligonucleotides is designed to amplify a cDNA of interest. Each pair of oligonucleotides incorporates a universal primer sequence for PCR as well as a specific address sequence that hybridizes to a capture sequence on an array. With this methodology, cancer-specific expression profiles were generated with as little as 50 ng of total RNA from formalin-fixed tissues stored over a decade (Bibikova et al., 2004). Approaches such as the DASL assay thus have the potential to capitalize on the vast supply of archived, paraffin-embedded samples from large clinical trials or population-based cohort studies to advance expression profiling efforts. Functional Genomics A goal of functional genomics is to assign biological meaning to genes on a genome-wide level (Steinmetz and Davis, 2004) (see Chapter 16). As illustrated by the examples described earlier, expression profiling has proved useful for assigning functional groups to genes in model organisms such as yeast (Hughes et al., 2000) and for uncovering a role for metastasis with RhoC (Clark et al., 2000). Expression profiling has also been used to decipher regulatory networks of transcription by discovering common sequence motifs in upstream regions of genes that have similar expression profiles (Pilpel et al., 2001). These studies, though, have been mainly been done in yeast and await application in more complex systems such as humans. A new opportunity in functional genomics is to integrate expression profiling with RNA interference (RNAi) technologies
now that genome-wide RNA interference libraries are coming to fruition (Moffat et al., 2006). One can imagine using RNAi libraries to systematically knock down individual genes and the comparing the resulting expression phenotypes. One can then look for similarities and differences between expression profiles as a starting point for inferring the function of the genes that have been ablated by RNAi. Profiling the Transcriptome Expression profiling has also been used to explore the transcriptome (see also Chapter 12). The transcriptome represents the complete set of all the transcribed elements of the genome in contrast to the collection of transcripts that are only translated into proteins.There is a burgeoning interest in the function of the noncoding part of the transcriptome as the number of actual genes transcribed into proteins has been substantially revised downward and increasing numbers of non-coding transcripts have been discovered. Key to their detection has been tiling microarrays, specifically those developed by Affymetrix with photolithographic technology. Rather than only using probes for known exons, tiling arrays probe across the entire genome in an unbiased fashion. Cheng et al. (2005) used a 25-mer oligonucleotide-based array spaced every 5 bp to examine ten human chromosomes. They discovered that interestingly, of all the transcribed sequences, 19% were polyadenylated, 44% were not polyadenylated, and 37% were both polyadenylated and not polyadenylated. Additionally, about 10% of the genome was transcribed as polyadenylated transcripts, compared to 1–2% of the genome corresponding to exons. These findings suggest that the expressed genome may be orders of magnitude larger than previously appreciated. Part of the transcriptome also includes small (roughly 22 nucleotide) non-coding microRNAs (miRNAs). These miRNAs play an important role in regulating gene expression by either degrading mRNA or inhibiting its translation (Bartel, 2004). In cancer, for example, Lu et al. (2005) used a new bead-based profiling system to quantify miRNA expression because of the short size of the miRNAs.This used oligonucleotide capture probes that were coupled to beads, which in turn were differentially impregnated with variable degrees of a fluorescent mixture. Flow cytometry was then used to quantitate the beads and therefore the associated miRNA. Expression profiling of the miRNA population of various malignant and normal tissues revealed an unusually large degree of diversity of miRNA expression across cancers. Furthermore, miRNA signatures could be used to classify poorly differentiated cancers, which have traditionally been difficult to classify using standard mRNA-based profiling strategies. The interested reader is referred to a recent review by Calin and Croce (2006).
Clinical Translation Expression Profiling as a Biomarker The ability of expression signatures of tumor samples to predict prognosis has immediate potential clinical translation. A key lesson from these early clinical microarray studies is that appropriate sample size and careful experimental design are necessary
Future Directions
for identifying signatures that can serve as robust biomarkers (Michiels et al., 2005; Ntzani and Ioannidis, 2003; Simon et al., 2003). However, reassuringly, a recent study of prognostic breast cancer signatures showed that four out of five reported gene signatures had significant concordance in predicting breast cancer outcome, indicating that these signatures were tracking similar biological behaviors (Fan et al., 2006), pointing to the evolving sophistication of clinically oriented profiling efforts. The clinical application of expression profiles is most mature in breast cancer, owing to the extensive microarray data already available, and has recently been reviewed (Sotiriou and Piccart, 2007). The 70-gene-expression signature originally formulated by Van’t Veer et al. (2002) has been commercialized into the MammaPrint assay for clinical use. This assay serves as the basis of the MINDACT (Microarray In Node negative Disease may Avoid ChemoTherapy) clinical trial, which aims to determine whether the expression signature can be used for making clinical decisions (Novak, 2006). It will test the question of whether or not patients with tumors that have high risk features by clinical criteria – individuals who are routinely recommended adjuvant chemotherapy – but with a low-risk molecular prognosis as determined by MammaPrint, can defer chemotherapy without affecting metastasis-free survival. In other words, it will test if the molecular prognosis generated by MammaPrint can outperform clinical criteria. An alternative to using existing signatures is to adapt the expression profile into a RT-PCR-based assay. Compared to DNA microarray technology, RT-PCR has the potential to be less technically challenging, more cost-effective, and more easily integrated into routine clinical practice. Genomic Health International has developed the Oncotype assay which is currently available for clinical use (Paik et al., 2004). The Oncotype assay is an RTPCR assay based on a 21-gene signature (including five control genes) derived from 250 candidate genes selected from the published microarray literature. Rather than using fresh frozen tissues, the researchers were able to exploit archived, paraffinembedded tissues using RT-PCR as their profiling method (Cronin et al., 2004). They systematically quantified the expression of these 250 candidate genes across three historical groups of breast cancer patients. From this data, they derived a gene expression-based recurrence score for predicting breast cancer recurrence that was more powerful than standard histopathologic criteria. The reproducibility of the score was robust and was subsequently validated in a large, population-based study (Habel et al., 2006). Furthermore, Oncotype had the potential to identify patients who may be able to defer adjuvant chemotherapy treatment. Patients with low and intermediate recurrence score tumors did not appear to derive a significant benefit from chemotherapy (Paik et al., 2006). An advantage of using this RT-PCR-based approach is that it can be applied to tumor samples that have been collected and processed in routine fashion; no special handling (e.g., freezing tissues in liquid nitrogen) is required. Other examples of adapting expression profiling to an RT-PCR assay include a two-gene breast cancer signature (Ma et al., 2003) and a six-gene signature predicting survival in DLBCL (Lossos et al., 2004).
■
167
In the future, one of the challenges for translational research will be the integration of expression profiling with routine histopathology in clinical practice. As can be seen with the 21gene breast cancer signature, expression profiling in this application offers the advantage of reproducibility and minimizes human error and variability with pathological interpretation. However, cost continues to be a major limitation to widespread adoption of profiling technologies. An ongoing prospective study, similar in concept to MINDACT, is the TAILORx (Trial Assigning Individualized Options for Treatment) trial in North America, which will help answer the question of how to integrate microarray technology (i.e., Oncotype DX) into clinical practice for patients with breast cancer. TAILORx is examining whether chemotherapy is required for the intermediate-risk group defined by Oncotype’s recurrence score (Paik, 2007). Drug Development DNA microarrays have also facilitated drug target discovery and prediction of drug response. For general reviews, the reader is referred to a recent one by Stoughton and Friend (2005). Using the NCI60 cancer cell line panel, Scherf et al. (2000) constructed a database that correlated gene expression and drug activity patterns. They were able to find relationships between gene expression and drugs such as 5-fluorouracil and l-asparaginase. Gunther et al. (2003) employed a similar strategy with primary human neuronal precursor cells and various psychotropic drugs. Potti et al. (2006) have expanded on this approach by using the expression profiles of sensitive and resistant cancer cell lines for six different chemotherapy drugs to predict response to neoadjuvant chemotherapy for breast cancer. Stegmaier et al. (2004) have developed a high-throughput approach to drug discovery using expression profiling: gene expression-based high-throughput screening (GE-HTS). These authors were interested in compounds that could terminally differentiate AML cells. To discover these compounds, they used 5-gene-expression signature of the differentiated neutrophil and leukemic states as a surrogate. With an RT-PCR-based readout of these five genes, they then screened a small molecule library to identify drugs that could induce differentiation of leukemic blasts and discovered an epidermal growth factor receptor kinase inhibitor with this property (Stegmaier et al., 2005). Remarkably, GE-HTS discovered this potential therapeutic without prior detailed biological knowledge. A related approach to screening small molecules has been implemented through The Connectivity Map, an in silico tool recently developed at the Broad Institute that uses expression profiling to systematically discover functional connections among diseases, genetic perturbations, and drug actions (Lamb et al., 2006). Lamb et al. first created a reference collection of gene expression profiles from cultured human cancer cells treated with small molecules. They then developed pattern-matching software to mine these data. The Connectivity Map has been used to perform in silico discoveries of rapamycin as a new approach for treating glucocorticoid-resistant ALL (Wei et al., 2006) and the inhibition of HSP90 as a novel mechanism for inhibition of androgen receptor signaling in prostate cancer (Hieronymus et al., 2006).
168
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
CONCLUSIONS Over the past decade, expression profiling has matured from a revolutionary new technology to a routine, fundamental component of scientific inquiry. Its ability to globally quantitate the expression of thousands of genes at once has provided a new lens for classifying diseases, predicting outcomes, and discovering new drug targets. Commensurate with the technical achievements
have been advances in statistical analysis and interpretation of the microarray data being generated. In the future, one can imagine additional insights through integration of expression profiling with other “–omic” approaches. Ultimately, the real benefits of DNA microarrays will be realized only through translation into better treatment strategies and therapeutics, and the results of these efforts are eagerly anticipated.
REFERENCES Alizadeh, A.A., Eisen, M.B., Davis, R.E., Ma, C., Lossos, I.S., Rosenwald, A., Boldrick, J.C., Sabet, H., Tran, T., Yu, X. et al. (2000). Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503–511. Allison, D.B., Cui, X., Page, G.P. and Sabripour, M. (2006). Microarray data analysis: From disarray to consolidation and consensus. Nat Rev Genet 7, 55–65. Alwine, J.C., Kemp, D.J. and Stark, G.R. (1977). Method for detection of specific RNAs in agarose gels by transfer to diazobenzyloxymethyl-paper and hybridization with DNA probes. Proc Natl Acad Sci USA 74, 5350–5354. Bammler, T., Beyer, R.P., Bhattacharya, S., Boorman, G.A., Boyles, A., Bradford, B.U., Bumgarner, R.E., Bushel, P.R., Chaturvedi, K., Choi, D. et al. (2005). Standardizing global gene expression analysis between laboratories and across platforms. Nat Methods 2, 351–356. Bard, J.B. and Rhee, S.Y. (2004). Ontologies in biology: Design, applications and future challenges. Nat Rev Genet 5, 213–222. Bartel, D.P. (2004). MicroRNAs: Genomics, biogenesis, mechanism, and function. Cell 116, 281–297. Bast, R.C., Jr., Ravdin, P., Hayes, D.F., Bates, S., Fritsche, H., Jr., Jessup, J.M., Kemeny, N., Locker, G.Y., Mennel, R.G. and Somerfield, M.R. (2001). 2000 update of recommendations for the use of tumor markers in breast and colorectal cancer: Clinical practice guidelines of the American Society of Clinical Oncology. J Clin Oncol 19, 1865–1878. Baugh, L.R., Hill, A.A., Brown, E.L. and Hunter, C.P. (2001). Quantitative analysis of mRNA amplification by in vitro transcription. Nucleic Acids Res 29, E29. Bibikova, M.,Talantov, D., Chudin, E.,Yeakley, J.M., Chen, J., Doucet, D., Wickham, E., Atkins, D., Barker, D., Chee, M. et al. (2004). Quantitative gene expression profiling in formalin-fixed, paraffin-embedded tissues using universal bead arrays. Am J Pathol 165, 1799–1807. Bild, A.H., Yao, G., Chang, J.T., Wang, Q., Potti, A., Chasse, D., Joshi, M.B., Harpole, D., Lancaster, J.M., Berchuck, A. et al. (2006). Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439, 353–357. Bowtell, D.D. (1999). Options available–from start to finish–for obtaining expression data by microarray. Nat Genet 21, 25–32. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeckert, C., Aach, J., Ansorge, W., Ball, C.A., Causton, H.C. et al. (2001). Minimum information about a microarray experiment (MIAME)-toward standards for microarray data. Nat Genet 29, 365–371. Brown, M.P., Grundy, W.N., Lin, D., Cristianini, N., Sugnet, C.W., Furey, T.S., Ares, M., Jr. and Haussler, D. (2000). Knowledge-based
analysis of microarray gene expression data by using support vector machines. Proc Natl Acad Sci USA 97, 262–267. Buyse, M., Loi, S., van’t Veer, L., Viale, G., Delorenzi, M., Glas, A.M., d’Assignies, M.S., Bergh, J., Lidereau, R., Ellis, P. et al. (2006). Validation and clinical utility of a 70-gene prognostic signature for women with node-negative breast cancer. J Natl Cancer Inst 98, 1183–1192. Calin, G.A. and Croce, C.M. (2006). MicroRNA signatures in human cancers. Nat Rev Cancer 6, 857–866. Chang, H.Y., Chi, J.T., Dudoit, S., Bondre, C.,van de Rijn, M., Botstein, D. and Brown, P.O. (2002). Diversity, topographic differentiation, and positional memory in human fibroblasts. Proc Natl Acad Sci USA 99, 12877–12882. Chang, H.Y., Nuyten, D.S., Sneddon, J.B., Hastie, T., Tibshirani, R., Sorlie, T., Dai, H., He, Y.D., van’t Veer, L.J., Bartelink, H. et al. (2005). Robustness, scalability, and integration of a woundresponse gene expression signature in predicting breast cancer survival. Proc Natl Acad Sci USA 102, 3738–3743. Chang, H.Y., Sneddon, J.B., Alizadeh, A.A., Sood, R., West, R.B., Montgomery, K., Chi, J.T., van de Rijn, M., Botstein, D. and Brown, P.O. (2004). Gene expression signature of fibroblast serum response predicts human cancer progression: Similarities between tumors and wounds. PLoS Biol 2, E7. Cheng, J., Kapranov, P., Drenkow, J., Dike, S., Brubaker, S., Patel, S., Long, J., Stern, D., Tammana, H., Helt, G. et al. (2005). Transcriptional maps of 10 human chromosomes at 5-nucleotide resolution. Science 308, 1149–1154. Cheung, V.G., Morley, M., Aguilar, F., Massimi, A., Kucherlapati, R. and Childs, G. (1999). Making and reading microarrays. Nat Genet 21, 15–19. Chi, J.T., Wang, Z., Nuyten, D.S., Rodriguez, E.H., Schaner, M.E., Salim,A.,Wang,Y., Kristensen, G.B., Helland,A., Borresen-Dale,A.L. et al. (2006). Gene expression programs in response to hypoxia: Cell type specificity and prognostic significance in human cancers. PLoS Med 3, e47. Chuaqui, R.F., Bonner, R.F., Best, C.J., Gillespie, J.W., Flaig, M.J., Hewitt, S.M., Phillips, J.L., Krizman, D.B., Tangrea, M.A., Ahram, M. et al. (2002). Post-analysis follow-up and validation of microarray experiments. Nat Genet 32(Suppl), 509–514. Churchill, G.A. (2002). Fundamentals of experimental design for cDNA microarrays. Nat Genet 32(Suppl), 490–495. Clark, E.A., Golub, T.R., Lander, E.S. and Hynes, R.O. (2000). Genomic analysis of metastasis reveals an essential role for RhoC. Nature 406, 532–535. Cronin, M., Pho, M., Dutta, D., Stephans, J.C., Shak, S., Kiefer, M.C., Esteban, J.M. and Baker, J.B. (2004). Measurement of gene expression in archival paraffin-embedded tissues: Development and
References
performance of a 92-gene reverse transcriptase-polymerase chain reaction assay. Am J Pathol 164, 35–42. D’Haeseleer, P. (2005). How does gene expression clustering work?. Nat Biotechnol 23, 1499–1501. Dallas, P.B., Gottardo, N.G., Firth, M.J., Beesley, A.H., Hoffmann, K., Terry, P.A., Freitas, J.R., Boag, J.M., Cummings, A.J. and Kees, U.R. (2005). Gene expression levels assessed by oligonucleotide microarray analysis and quantitative real-time RT-PCR – how well do they correlate?. BMC Genomics 6, 59. Dudley, A.M., Aach, J., Steffen, M.A. and Church, G.M. (2002). Measuring absolute expression with microarrays with a calibrated reference sample and an extended signal intensity range. Proc Natl Acad Sci USA 99, 7554–7559. Duggan, D.J., Bittner, M., Chen, Y., Meltzer, P. and Trent, J.M. (1999). Expression profiling using cDNA microarrays. Nat Genet 21, 10–14. Dvorak, H.F. (1986). Tumors: Wounds that do not heal. Similarities between tumor stroma generation and wound healing. N Engl J Med 315, 1650–1659. Ebert, B.L. and Golub, T.R. (2004). Genomic approaches to hematologic malignancies. Blood 104, 923–932. Eisen, M.B., Spellman, P.T., Brown, P.O. and Botstein, D. (1998). Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA 95, 14863–14868. Emmert-Buck, M.R., Bonner, R.F., Smith, P.D., Chuaqui, R.F., Zhuang, Z., Goldstein, S.R., Weiss, R.A. and Liotta, L.A. (1996). Laser capture microdissection. Science 274, 998–1001. Emmert-Buck, M.R., Roth, M.J., Zhuang, Z., Campo, E., Rozhin, J., Sloane, B.F., Liotta, L.A. and Stetler-Stevenson, W.G. (1994). Increased gelatinase A (MMP-2) and cathepsin B activity in invasive tumor regions of human colon cancer samples. Am J Pathol 145, 1285–1290. Ermolaeva, O., Rastogi, M., Pruitt, K.D., Schuler, G.D., Bittner, M.L., Chen, Y., Simon, R., Meltzer, P., Trent, J.M. and Boguski, M.S. (1998). Data management and analysis for gene expression arrays. Nat Genet 20, 19–23. Fan, C., Oh, D.S., Wessels, L., Weigelt, B., Nuyten, D.S., Nobel, A.B., van’t Veer, L.J. and Perou, C.M. (2006). Concordance among geneexpression-based predictors for breast cancer. N Engl J Med 355, 560–569. Fan, J.B.,Yeakley, J.M., Bibikova, M., Chudin, E., Wickham, E., Chen, J., Doucet, D., Rigault, P., Zhang, B., Shen, R. et al. (2004). A versatile assay for high-throughput gene expression profiling on universal array matrices. Genome Res 14, 878–885. Fodor, S.P., Read, J.L., Pirrung, M.C., Stryer, L., Lu, A.T. and Solas, D. (1991). Light-directed, spatially addressable parallel chemical synthesis. Science 251, 767–773. Gaasterland, T. and Bekiranov, S. (2000). Making the most of microarray data. Nat Genet 24, 204–206. Godfrey, T.E., Kim, S.H., Chavira, M., Ruff, D.W., Warren, R.S., Gray, J.W. and Jensen, R.H. (2000). Quantitative mRNA expression analysis from formalin-fixed, paraffin-embedded tissues using 5 nuclease quantitative reverse transcription-polymerase chain reaction. J Mol Diagn 2, 84–91. Goldhirsch, A., Glick, J.H., Gelber, R.D., Coates, A.S. and Senn, H.J. (2001). Meeting highlights: International Consensus Panel on the Treatment of Primary Breast Cancer. Seventh International Conference on Adjuvant Therapy of Primary Breast Cancer. J Clin Oncol 19, 3817–3827.
■
169
Golub, T.R., Slonim, D.K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J.P., Coller, H., Loh, M.L., Downing, J.R., Caligiuri, M.A. et al. (1999). Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring. Science 286, 531–537. Gunderson, K.L., Kruglyak, S., Graige, M.S., Garcia, F., Kermani, B.G., Zhao, C., Che, D., Dickinson, T., Wickham, E., Bierle, J. et al. (2004). Decoding randomly ordered DNA arrays. Genome Res 14, 870–877. Gunther, E.C., Stone, D.J., Gerwien, R.W., Bento, P. and Heyes, M.P. (2003). Prediction of clinical drug efficacy by classification of drug-induced genomic expression profiles in vitro. Proc Natl Acad Sci USA 100, 9608–9613. Guo, Z., Guilfoyle, R.A., Thiel, A.J., Wang, R. and Smith, L.M. (1994). Direct fluorescence analysis of genetic polymorphisms by hybridization with oligonucleotide arrays on glass supports. Nucleic Acids Res 22, 5456–5465. Habel, L.A., Shak, S., Jacobs, M.K., Capra, A., Alexander, C., Pho, M., Baker, J., Walker, M., Watson, D., Hackett, J. et al. (2006). A population-based study of tumor gene expression and risk of breast cancer death among lymph node-negative patients. Breast Cancer Res 8, R25. Hans, C.P., Weisenburger, D.D., Greiner, T.C., Gascoyne, R.D., Delabie, J., Ott, G., Muller-Hermelink, H.K., Campo, E., Braziel, R.M., Jaffe, E.S. et al. (2004). Confirmation of the molecular classification of diffuse large B-cell lymphoma by immunohistochemistry using a tissue microarray. Blood 103, 275–282. Hardiman, G. (2004). Microarray platforms–comparisons and contrasts. Pharmacogenomics 5, 487–502. Harris, M.A., Clark, J., Ireland, A., Lomax, J., Ashburner, M., Foulger, R., Eilbeck, K., Lewis, S., Marshall, B., Mungall, C. et al. (2004). The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res 32, D258–D261. Heller, R.A., Schena, M., Chai, A., Shalon, D., Bedilion, T., Gilmore, J., Woolley, D.E. and Davis, R.W. (1997). Discovery and analysis of inflammatory disease-related genes using cDNA microarrays. Proc Natl Acad Sci USA 94, 2150–2155. Hieronymus, H., Lamb, J., Ross, K.N., Peng, X.P., Clement, C., Rodina, A., Nieto, M., Du, J., Stegmaier, K., Raj, S.M. et al. (2006). Gene expression signature-based chemical genomic prediction identifies a novel class of HSP90 pathway modulators. Cancer Cell 10, 321–330. Hoheisel, J.D. (2006). Microarray technology: Beyond transcript profiling and genotype analysis. Nat Rev Genet 7, 200–210. Holloway, A.J., van Laar, R.K., Tothill, R.W. and Bowtell, D.D. (2002). Options available–from start to finish–for obtaining data from DNA microarrays II. Nat Genet 32(Suppl), 481–489. Huang, J., Qi, R., Quackenbush, J., Dauway, E., Lazaridis, E. and Yeatman, T. (2001). Effects of ischemia on gene expression. J Surg Res 99, 222–227. Hughes, T.R., Mao, M., Jones, A.R., Burchard, J., Marton, M.J., Shannon, K.W., Lefkowitz, S.M., Ziman, M., Schelter, J.M., Meyer, M.R. et al. (2001). Expression profiling using microarrays fabricated by an ink-jet oligonucleotide synthesizer. Nat Biotechnol 19, 342–347. Hughes, T.R., Marton, M.J., Jones, A.R., Roberts, C.J., Stoughton, R., Armour, C.D., Bennett, H.A., Coffey, E., Dai, H., He, Y.D. et al. (2000). Functional discovery via a compendium of expression profiles. Cell 102, 109–126.
170
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
Irizarry, R.A., Warren, D., Spencer, F., Kim, I.F., Biswal, S., Frank, B.C., Gabrielson, E., Garcia, J.G., Geoghegan, J., Germino, G. et al. (2005). Multiple-laboratory comparison of microarray platforms. Nat Methods 2, 345–350. Iscove, N.N., Barbara, M., Gu, M., Gibson, M., Modi, C. and Winegarden, N. (2002). Representation is faithfully preserved in global cDNA amplified exponentially from sub-picogram quantities of mRNA. Nat Biotechnol 20, 940–943. Ishii, M., Hashimoto, S.,Tsutsumi, S.,Wada,Y., Matsushima, K., Kodama,T. and Aburatani, H. (2000). Direct comparison of GeneChip and SAGE on the quantitative accuracy in transcript profiling analysis. Genomics 68, 136–143. King, H.C. and Sinha, A.A. (2001). Gene expression profile analysis by DNA microarrays: Promise and pitfalls. JAMA 286, 2280–2288. Kittleson, M.M., Ye, S.Q., Irizarry, R.A., Minhas, K.M., Edness, G., Conte, J.V., Parmigiani, G., Miller, L.W., Chen, Y., Hall, J.L. et al. (2004). Identification of a gene expression profile that differentiates between ischemic and nonischemic cardiomyopathy. Circulation 110, 3444–3451. Lamb, J., Crawford, E.D., Peck, D., Modell, J.W., Blat, I.C., Wrobel, M.J., Lerner, J., Brunet, J.P., Subramanian, A., Ross, K.N. et al. (2006). The Connectivity Map: Using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929–1935. Lamb, J., Ramaswamy, S., Ford, H.L., Contreras, B., Martinez, R.V., Kittrell, F.S., Zahnow, C.A., Patterson, N., Golub, T.R. and Ewen, M.E. (2003).A mechanism of cyclin D1 action encoded in the patterns of gene expression in human cancer. Cell 114, 323–334. Larkin, J.E., Frank, B.C., Gavras, H., Sultana, R. and Quackenbush, J. (2005). Independence and reproducibility across microarray platforms. Nat Methods 2, 337–344. Liang, P. and Pardee, A.B. (1992). Differential display of eukaryotic messenger RNA by means of the polymerase chain reaction. Science 257, 967–971. Liang, P. and Pardee, A.B. (2003). Analysing differential gene expression in cancer. Nat Rev Cancer 3, 869–876. Lin, D.W., Coleman, I.M., Hawley, S., Dumpit, R., Gifford, D., Kezele, P., Hung, H., Knudsen, B.S., Kristal, A.R. and Nelson, P.S. (2006). Influence of surgical manipulation on prostate gene expression: Implications for molecular correlates of treatment effects and disease prognosis. J Clin Oncol 24, 3763–3770. Lipshutz, R.J., Fodor, S.P., Gingeras, T.R. and Lockhart, D.J. (1999). High density synthetic oligonucleotide arrays. Nat Genet 21, 20–24. Liu, R.H., Nguyen, T., Schwarzkopf , K., Fuji, H.S., Petrova, A., Siuda, T., Peyvan, K., Bizak, M., Danley, D. and McShea, A. (2006). Fully integrated miniature device for automated gene expression DNA microarray processing. Anal Chem 78, 1980–1986. Lock, C., Hermans, G., Pedotti, R., Brendolan, A., Schadt, E., Garren, H., Langer-Gould, A., Strober, S., Cannella, B., Allard, J. et al. (2002). Gene-microarray analysis of multiple sclerosis lesions yields new targets validated in autoimmune encephalomyelitis. Nat Med 8, 500–508. Lockhart, D.J., Dong, H., Byrne, M.C., Follettie, M.T., Gallo, M.V., Chee, M.S., Mittmann, M., Wang, C., Kobayashi, M., Horton, H. et al. (1996). Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat Biotechnol 14, 1675–1680. Lossos, I.S., Czerwinski, D.K., Alizadeh, A.A., Wechser, M.A., Tibshirani, R., Botstein, D. and Levy, R. (2004). Prediction of survival in diffuse large-B-cell lymphoma based on the expression of six genes. N Engl J Med 350, 1828–1837.
Lu, J., Getz, G., Miska, E.A., Alvarez-Saavedra, E., Lamb, J., Peck, D., Sweet-Cordero, A., Ebert, B.L., Mak, R.H., Ferrando, A.A. et al. (2005). MicroRNA expression profiles classify human cancers. Nature 435, 834–838. Ly, D.H., Lockhart, D.J., Lerner, R.A. and Schultz, P.G. (2000). Mitotic misregulation and human aging. Science 287, 2486–2492. Ma, X.J., Salunga, R., Tuggle, J.T., Gaudet, J., Enright, E., McQuary, P., Payette, T., Pistone, M., Stecker, K., Zhang, B.M. et al. (2003). Gene expression profiles of human breast cancer progression. Proc Natl Acad Sci USA 100, 5974–5979. Marshall, E. (2004). Getting the noise out of gene arrays. Science 306, 630–631. Masuda, N., Ohnishi, T., Kawamoto, S., Monden, M. and Okubo, K. (1999). Analysis of chemical modification of RNA from formalinfixed samples and optimization of molecular biology applications for such samples. Nucleic Acids Res 27, 4436–4443. Michiels, S., Koscielny, S. and Hill, C. (2005). Prediction of cancer outcome with microarrays: A multiple random validation strategy. Lancet 365, 488–492. Mirnics, K., Middleton, F.A., Marquez, A., Lewis, D.A. and Levitt, P. (2000). Molecular characterization of schizophrenia viewed by microarray analysis of gene expression in prefrontal cortex. Neuron 28, 53–67. Moffat, J., Grueneberg, D.A., Yang, X., Kim, S.Y., Kloepfer, A.M., Hinkle, G., Piqani, B., Eisenhaure, T.M., Luo, B., Grenier, J.K. et al. (2006). A lentiviral RNAi library for human and mouse genes applied to an arrayed viral high-content screen. Cell 124, 1283–1298. Mootha,V.K., Lindgren, C.M., Eriksson, K.F., Subramanian, A., Sihag, S., Lehar, J., Puigserver, P., Carlsson, E., Ridderstrale, M., Laurila, E. et al. (2003). PGC-1-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes. Nat Genet 34, 267–273. Novak, K. (2006). Where the chips fall. Nat Med 12, 158–159. Ntzani, E.E. and Ioannidis, J.P. (2003). Predictive ability of DNA microarrays for cancer outcomes and correlates: An empirical assessment. Lancet 362, 1439–1444. Paik, S. (2007). Development and clinical utility of a 21-gene recurrence score prognostic assay in patients with early breast cancer treated with tamoxifen. Oncologist 12, 631–635. Paik, S., Shak, S., Tang, G., Kim, C., Baker, J., Cronin, M., Baehner, F.L., Walker, M.G., Watson, D., Park, T. et al. (2004). A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med 351, 2817–2826. Paik, S., Tang, G., Shak, S., Kim, C., Baker, J., Kim, W., Cronin, M., Baehner, F.L., Watson, D., Bryant, J. et al. (2006). Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer. J Clin Oncol 24, 3726–3734. Pease, A.C., Solas, D., Sullivan, E.J., Cronin, M.T., Holmes, C.P. and Fodor, S.P. (1994). Light-generated oligonucleotide arrays for rapid DNA sequence analysis. Proc Natl Acad Sci USA 91, 5022–5026. Perou, C.M., Sorlie, T., Eisen, M.B., van de Rijn, M., Jeffrey, S.S., Rees, C.A., Pollack, J.R., Ross, D.T., Johnsen, H., Akslen, L.A. et al. (2000). Molecular portraits of human breast tumours. Nature 406, 747–752. Pilpel, Y., Sudarsanam, P. and Church, G.M. (2001). Identifying regulatory networks by combinatorial analysis of promoter elements. Nat Genet 29, 153–159. Potti, A., Dressman, H.K., Bild, A., Riedel, R.F., Chan, G., Sayer, R., Cragun, J., Cottrill, H., Kelley, M.J., Petersen, R. et al. (2006).
References
Genomic signatures to guide the use of chemotherapeutics. Nat Med 12, 1294–1300. Quackenbush, J. (2001). Computational analysis of microarray data. Nat Rev Genet 2, 418–427. Ramaswamy, S. and Golub, T.R. (2002). DNA microarrays in clinical oncology. J Clin Oncol 20, 1932–1941. Ramaswamy, S., Ross, K.N., Lander, E.S. and Golub, T.R. (2003). A molecular signature of metastasis in primary solid tumors. Nat Genet 33, 49–54. Ramaswamy, S., Tamayo, P., Rifkin, R., Mukherjee, S., Yeang, C.H., Angelo, M., Ladd, C., Reich, M., Latulippe, E., Mesirov, J.P. et al. (2001). Multiclass cancer diagnosis using tumor gene expression signatures. Proc Natl Acad Sci USA 98, 15149–15154. Rhodes, D.R., Yu, J., Shanker, K., Deshpande, N., Varambally, R., Ghosh, D., Barrette, T., Pandey, A. and Chinnaiyan, A.M. (2004). ONCOMINE: A cancer microarray database and integrated datamining platform. Neoplasia 6, 1–6. Ross, D.T., Scherf, U., Eisen, M.B., Perou, C.M., Rees, C., Spellman, P., Iyer, V., Jeffrey, S.S., Van de Rijn, M., Waltham, M. et al. (2000). Systematic variation in gene expression patterns in human cancer cell lines. Nat Genet 24, 227–235. Sargent, T.D. (1987). Isolation of differentially expressed genes. Methods Enzymol 152, 423–432. Schena, M., Heller, R.A., Theriault, T.P., Konrad, K., Lachenmeier, E. and Davis, R.W. (1998). Microarrays: Biotechnology’s discovery platform for functional genomics. Trends Biotechnol 16, 301–306. Schena, M., Shalon, D., Davis, R.W. and Brown, P.O. (1995). Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270, 467–470. Scherf , U., Ross, D.T., Waltham, M., Smith, L.H., Lee, J.K., Tanabe, L., Kohn, K.W., Reinhold, W.C., Myers, T.G., Andrews, D.T. et al. (2000). A gene expression database for the molecular pharmacology of cancer. Nat Genet 24, 236–244. Schulze, A. and Downward, J. (2001). Navigating gene expression using microarrays – A technology review. Nat Cell Biol 3, E190–E195. Segal, E., Friedman, N., Koller, D. and Regev, A. (2004). A module map showing conditional activity of expression modules in cancer. Nat Genet 36, 1090–1098. Sgroi, D.C., Teng, S., Robinson, G., LeVangie, R., Hudson, J.R., Jr and Elkahloun, A.G. (1999). In vivo gene expression profile analysis of human breast cancer progression. Cancer Res 59, 5656–5661. Shi, L., Reid, L.H., Jones, W.D., Shippy, R., Warrington, J.A., Baker, S.C., Collins, P.J., de Longueville, F., Kawasaki, E.S., Lee, K.Y. et al. (2006). The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nat Biotechnol 24, 1151–1161. Simon, R., Radmacher, M.D., Dobbin, K. and McShane, L.M. (2003). Pitfalls in the use of DNA microarray data for diagnostic and prognostic classification. J Natl Cancer Inst 95, 14–18. Sorlie, T., Perou, C.M., Tibshirani, R., Aas, T., Geisler, S., Johnsen, H., Hastie, T., Eisen, M.B., van de Rijn, M., Jeffrey, S.S. et al. (2001). Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA 98, 10869–10874. Sorlie, T., Tibshirani, R., Parker, J., Hastie, T., Marron, J.S., Nobel, A., Deng, S., Johnsen, H., Pesich, R., Geisler, S. et al. (2003). Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci USA 100, 8418–8423.
■
171
Sotiriou, C. and Piccart, M.J. (2007). Taking gene-expression profiling to the clinic: When will molecular signatures become relevant to patient care?. Nat Rev Cancer 7, 545–553. Southern, E., Mir, K. and Shchepinov, M. (1999). Molecular interactions on microarrays. Nat Genet 21, 5–9. Stegmaier, K., Corsello, S.M., Ross, K.N., Wong, J.S., Deangelo, D.J. and Golub, T.R. (2005). Gefitinib induces myeloid differentiation of acute myeloid leukemia. Blood 106, 2841–2848. Stegmaier, K., Ross, K.N., Colavito, S.A., O’Malley, S., Stockwell, B.R. and Golub, T.R. (2004). Gene expression-based high-throughput screening (GE-HTS) and application to leukemia differentiation. Nat Genet 36, 257–263. Steinmetz, L.M. and Davis, R.W. (2004). Maximizing the potential of functional genomics. Nat Rev Genet 5, 190–201. Stoeckert, C.J., Causton, H.C. and Ball, C.A. (2002). Microarray databases: Standards and ontologies. Nat Genet 32(Suppl), 469–473. Stoughton, R.B. and Friend, S.H. (2005). How molecular profiling could revolutionize drug discovery. Nat Rev Drug Discov 4, 345–350. Subramanian, A., Tamayo, P., Mootha, V.K., Mukherjee, S., Ebert, B.L., Gillette, M.A., Paulovich,A., Pomeroy, S.L., Golub,T.R., Lander, E.S. et al. (2005). Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA 102, 15545–15550. Tan, P.K., Downey,T.J., Spitznagel, E.L., Jr., Xu, P., Fu, D., Dimitrov, D.S., Lempicki, R.A., Raaka, B.M. and Cam, M.C. (2003). Evaluation of gene expression measurements from commercial microarray platforms. Nucleic Acids Res 31, 5676–5684. Taniguchi, M., Miura, K., Iwao, H. and Yamanaka, S. (2001). Quantitative assessment of DNA microarrays–comparison with Northern blot analyses. Genomics 71, 34–39. Tomlins, S.A., Rhodes, D.R., Perner, S., Dhanasekaran, S.M., Mehra, R., Sun, X.W., Varambally, S., Cao, X., Tchinda, J., Kuefer, R. et al. (2005). Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer. Science 310, 644–648. van’t Veer, L.J., Dai, H., van de Vijver, M.J., He,Y.D., Hart, A.A., Mao, M., Peterse, H.L., van der Kooy, K., Marton, M.J., Witteveen, A.T. et al. (2002). Gene expression profiling predicts clinical outcome of breast cancer. Nature 415, 530–536. van de Vijver, M.J., He, Y.D., van’t Veer, L.J., Dai, H., Hart, A.A., Voskuil, D.W., Schreiber, G.J., Peterse, J.L., Roberts, C., Marton, M.J. et al. (2002). A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347, 1999–2009. Van Gelder, R.N., von Zastrow, M.E., Yool, A., Dement, W.C., Barchas, J.D. and Eberwine, J.H. (1990). Amplified RNA synthesized from limited quantities of heterogeneous cDNA. Proc Natl Acad Sci USA 87, 1663–1667. Velculescu, V.E., Zhang, L., Vogelstein, B. and Kinzler, K.W. (1995). Serial analysis of gene expression. Science 270, 484–487. Wei, G., Twomey, D., Lamb, J., Schlis, K., Agarwal, J., Stam, R.W., Opferman, J.T., Sallan, S.E., den Boer, M.L., Pieters, R. et al. (2006). Gene expression-based chemical genomics identifies rapamycin as a modulator of MCL1 and glucocorticoid resistance. Cancer Cell 10, 331–342. Zimmermann, C.R., Orr, W.C., Leclerc, R.F., Barnard, E.C. and Timberlake, W.E. (1980). Molecular cloning and selection of genes regulated in Aspergillus development. Cell 21, 709–715.
172
CHAPTER 13
■
DNA Microarrays in Biological Discovery and Patient Care
RECOMMENDED RESOURCES Resources for expression profiling experiments Web site
URL
Resources
National Human Genome Research Institute
http://research.nhgri.nih.gov/ microarray/main.html
Protocols, web links
Broad Institute, Cancer Genomics Group
http://www.broad.mit.edu/cancer/
Gene expression analysis software, web links
Stanford Genomics
http://genome-www.stanford.edu/
Software, web links, cDNA microarray protocols
Jackson Laboratory, Statistical Genomics Group
http://www.jax. org/staff/Churchill/labsite/
Advice about the design of microarray experiments
Microarray Gene Expression Data Society (MGED)
http://www.mged.org/ Workgroups/MIAME/miame_ checklist.html
Minimal requirements for the publication of microarray data
Gene Expression Omnibus database
http://www.ncbi.nlm.nih.gov/geo/
Repository of gene expression data
Array Express database
http://www.ebi.ac.uk/arrayexpress/
Repository of gene expression data
Affymetrix
http://www.affymetrix.com
Commercial website with useful resources
This listing is adapted from Ebert and Golub (2004).
CHAPTER
14 Proteomics: The Deciphering of the Functional Genome Li-Rong Yu, Nicolas A. Stewart and Timothy D. Veenstra
INTRODUCTION The continuing development of novel analytical technologies and instrumentation has brought a dramatic new dimension to how biological research can be pursued. This new dimension has enabled discovery-driven research to be conducted at par with hypothesis-driven research. In a hypothesis-driven approach, the understanding of a particular biological entity – whether it be a cell, gene, or protein – is used to develop studies that are designed to answer a specific question about a single gene, transcript, or protein, for example. Hypothesis-driven studies utilize technologies that focus on one specific entity per experiment. Discoverydriven studies, however, are designed around broad questions and focus on global characteristics of a cell or organism. These studies utilize advanced technologies that are able to gather information on thousands of different genes, transcripts, proteins, and metabolites in a relatively rapid and comprehensive fashion. While these globally directed studies have seemingly been thrust upon the scientific community overnight, in reality it has taken decades of technological and instrumental development to bring this field of science to fruition. The first giant step in discovery-driven science is the completion of the Human Genome Project. This monumental effort could only have been accomplished with the many developments in cloning, amplification, and high-throughput gene sequencing (Yager et al., 1991). The analysis of thousands of gene transcripts using mRNA arrays has been made possible by the ability to synthesize mRNA probes
and coupling them to solid surfaces (Gilham, 1970). Shortly after the development of these genome – and transcriptome-scanning technologies, many scientists shifted their focus onto proteins. Proteomics – the analysis of the entire protein complement of a cell, tissue, or organism under a specific, defined set of conditions – in its present state has also been dependent on decades of technological and instrumental developments. These developments have included advances in mass spectrometry (MS) technology, protein fractionation techniques, and bioinformatics, to name a few. One of the difficult challenges in proteomics is the uncertainty of its complexity. Let us assume that as many as 15,000 genes are expressed within a given human cell at any single time. Let us then assume that these genes give rise, on average, to four different mRNA transcripts each, or 60,000 different transcripts in total. This estimate is very conservative given the presence of splice variants, single nucleotide polymorphisms, etc. that can arise from expressed genes. Now consider all of the events that can happen to a single transcript during and after translation into a mature protein. The protein can be processed from a preprotein into a mature protein.The most common event that happens to change a protein is post-translational modification. There are over 300 known modifications, however; some of the more common are (in no particular order) phosphorylation, glycosylation, methylation, and acetylation. A single protein that may have four potential phosphorylation sites will, in itself, give rise to 16 different isoforms ranging from all the sites phosphorylated to none of them being phosphorylated. If each of the 60,000
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 173
■
Proteomics: The Deciphering of the Functional Genome
possible protein products contained only four possible sites of modification, these would give rise to as many as 960,000 different proteins. Obviously, these numbers are extremely conservative and suggest that the human proteome is likely to contain well over 1,000,000 distinct species. This number is well beyond the current capacity of state-of-the-art technologies, resulting in a consistent undersampling of the proteome regardless of the analysis method chosen. It is this uncertainty in the overall measurement that results in the question that hovers around many global proteomic studies: “Was the protein absent in the sample or did we simply fail to detect it?”
Isoelectric point
Isoelectric point
(a)
(b) Molecular mass
CHAPTER 14
Molecular mass
174
In-gel enzymatic digestion
GEL-BASED AND SOLUTION-BASED PROTEOMICS Proteomics relies on three basic technological cornerstones: a method to fractionate complex protein or peptide mixtures, MS to acquire the data necessary to identify individual proteins, and bioinformatics to analyze and assemble the MS data. While the MS and bioinformatic components are somewhat similar in most applications, there are two very distinct methods to separate out complex protein samples in proteomics. The mere mention of the word “proteomics” conjures up an image of stained protein spots that have been fractionated by two-dimensional polyacrylamide gel electrophoresis (2D-PAGE). This separation technology, which was developed over 30 years ago, is still the most common separation technique used in proteomics today (O’Farrell, 1975). Advances have been made in this technology that enables thousands of proteins to be resolved in a single gel. While the primary role of 2D-PAGE is to separate complex mixtures of proteins, it also allows a means to compare the relative abundances of proteins from different proteomes. In a typical 2D-PAGE-based study, the proteomes from two distinct cell populations (e.g., control versus treated cells) are extracted and fractionated on separate gels (Figure 14.1) (Van den Bergh and Arckens, 2005). Since most proteins are colorless, the separated proteins need to be visualized (Westermeier and Marouga, 2005). This visualization is routinely accomplished using colorimetric stains, such as coomassie blue or silver stain. The resulting images are typically quite complex; therefore the spots observed on both gels are aligned so that the relative staining intensity of individual proteins can be compared between gels. Protein spots that are stained more intensely on one gel compared to the other are excised from the gel. The protein within the gel is then enzymatically digested by introducing trypsin into the gel piece. The resultant tryptic peptides are extracted from the gel and analyzed by MS or tandem MS (MS/MS) to acquire the raw data necessary to identify the protein(s) within the gel spot. This identification is accomplished using software designed to compare this type of data against databases containing large amounts of genomic or protein sequence information. While 2D-PAGE fractionation has been the subject of many different criticisms over the years, it still remains a cornerstone technology in proteomics (Van den Bergh and Arckens, 2005). Many of
Protein identification by peptide mapping or tandem MS m/z
Search database
Generate list of differentially abundant proteins
Figure 14.1 Quantitative proteomics using two-dimensional polyacrylamide gel electrophoresis (2D-PAGE). In this method, comparative proteome samples are separated on distinct 2DPAGE gels. After staining, protein spots that are more abundant on one gel compared to the other are excised from the gel. The protein(s) within the gel is then subjected to in-gel tryptic digestion and the resultant peptides are extracted and analyzed by mass spectrometry (MS). The MS data are then searched against the appropriate database to identify the protein(s) with the gel spot.
the criticisms, such as its inability to resolve membrane proteins and gel-to-gel irreproducibility have been addressed through the development of better reagents, equipment, and gel alignment software. For instance, over the past several of years, 2D-DIGE reagents have been developed that allow comparative proteomes to be labeled with different fluorophores (i.e., Cy2, Cy3, and Cy5) (Hoorn et al., 2006). This differential labeling allows for the relative abundance of proteins from different proteomes to be compared by their combination prior to their run on a single gel, obviating the separation irreproducibility that can be observed between gels. No single chromatographic or electrophoretic procedure, including 2D-PAGE, can completely resolve a mixture as complex as that of a proteome. As a complement to 2D-PAGE methods,
Mass Spectrometry
Proteome A (ICAT-13C0)
175
Proteome B (ICAT-13C9)
Pool proteins
Digest with trypsin
Enrich ICAT-peptides
Cleave biotin tags
light
heavy 9.03 Da
% Intensity
solution-based chromatographic techniques have been developed that permit direct on-line MS analysis of proteomes (Issaq et al., 2005). Analogous to 2D-PAGE, these solution-based methods typically use at least two different fractionation schemes to separate peptides/proteins prior to their entering the mass spectrometer. While many different combinations have been proposed, the most effective and popular has been the use of strong cation exchange (SCX) followed by reversed-phase high-performance liquid chromatography (RPLC). John Yates III first demonstrated the effectiveness of combining this multidimensional separation with MS (Washburn et al., 2001).This method, termed “MudPIT” (multidimensional protein identification technology), has shown the capability of identifying thousands of proteins per proteome study. In this scheme, the proteins extracted from a cell lysate are first digested into peptides. There are two major advantages in working with peptides instead of proteins. Peptides are more soluble and easier to separate than intact proteins, especially hydrophobic and membrane proteins, and peptides are more amenable to identification by MS than are proteins (as described in more detail later). The challenge with this strategy is that the number of species that have to be resolved has dramatically increased. The digestate is initially separated by SCX and fractions collected off this column are then loaded onto a RP column that is coupled directly on-line with the mass spectrometer. The mass spectrometer can then dynamically select and acquire the data necessary to identify peptides as they elute from the RP column. A limitation of MudPIT when compared to 2D-PAGE is in its inability to directly measure the relative abundance of proteins in different proteome samples. To meet this need, many different approaches have been developed, all of which use the data provided by the mass spectrometer for quantitation (Ong and Mann, 2005). Many of the methods rely on the use of differential stable-isotope labeling (Julka and Regnier, 2005). Analogous to DIGE, which uses differently fluorescent molecules with identical electrophoretic properties, stable-isotope labeling of proteomes uses chemically identical reagents that differ in their isotope content (i.e., mass). For example, the most popular labeling method termed isotope-coded affinity tags (ICAT) utilizes thiol-modifying reagents that differ in their carbon-13 (13C) content (Gygi et al., 1999). In this procedure, one proteome sample is labeled with the “light” ICAT reagent while a comparative sample is labeled with the “heavy” ICAT reagent that contains nine 13C atoms in place of nine 12C atoms found in the “light” version (Figure 14.2). The label is attached to the proteins after they have been separately extracted from the samples to be compared. Once the two samples are differentially modified, they can be combined. The proteins are digested into peptides and the modified peptides are extracted by virtue of a biotin tag that is present at the termini of the ICAT reagents. This enriched peptide mixture is finally analyzed using MudPIT and the relative abundance and identity of the peptides are determined from measurements provided by the mass spectrometer. While ICAT illustrates a post-extraction/chemical modification method, stable-isotope labeling can also be performed at the metabolic level by culturing cells in growth media that is highly enriched in a
■
NCSETQYESK ICAT
m/z MS quantitation
m/z MS/MS identification
Figure 14.2 Quantitative proteomics using isotope-coded affinity tags (ICAT). In this method, comparative proteome samples are labeled with chemically identical reagents that differ only by their carbon isotope content (i.e., 12C9 for the light reagent and 13C9 for the heavy reagent). After the proteins are modified, the proteome samples are combined and digested into tryptic peptides. The ICAT-modified peptides are extracted using avidin chromatography by virtue of the biotin moiety on the terminus of the ICAT reagents. After removal of the biotin portion, these ICAT-peptides are analyzed by reversed-phase liquid chromatography coupled directly on-line with a mass spectrometer. The mass spectrometer is operated in a data-dependent tandem mass spectrometry (MS/MS) mode, enabling the relative quantitation of the peptide in the two samples to be measured in the MS mode as well as its identity be discerned from the data acquired by MS/MS.
particular heavy isotope (e.g., 15N in place of 14N) or a “heavy” amino acid (e.g., 13C6-lysine) (Ong et al., 2003). A list of various sample preparation used in proteomics today, along with their advantages and disadvantages is provided in Table 14.1.
MASS SPECTROMETRY While high-resolution separations are critical for complex protein mixture analysis, MS is inarguably the technology that
176
CHAPTER 14
■
Proteomics: The Deciphering of the Functional Genome
T A B L E 1 4 . 1 List of advantages and disadvantages of popular sample preparation methods used in proteomics Method
Advantages
2D-PAGE
●
●
2D-DIGE
●
●
MudPIT
● ● ●
ICAT
●
Disadvantages
Staining provides relative quantitation Only sequence proteins of interest
●
Can compare multiple proteomes on single gel Reproducible separation
●
Highly sensitive Broad proteome coverage High throughput
●
Reduces complexity prior to MS analysis
●
● ●
●
●
● ●
iTRAQ
●
Allows 4 samples to be simultaneously compared
● ●
Laborious Limited sensitivity Irreproducibility a factor Laborious Requires fluorescent scanner
Limited individual protein coverage Unable to target specific proteins Lacks inherent ability for quantitative comparison Low throughput Limited to a binary comparison Unable to target specific proteins Low throughput Unable to target specific proteins
has driven the proteomics revolution. Although, it has been around for approximately a century, MS has gained increased prominence with the development of techniques to rapidly identify proteins. There are many different reasons that have propelled the MS to its prominent position within the field of proteomics. Among these are the sensitivity afforded using MS, allowing proteins and peptides present in the low femtomole (fmol, 1015 mol) to high attomole (amol, 1018 mol) range to be successfully identified. The mass measurement accuracy available using current MS technology, which can routinely be in the range of 1–5 ppm, also increases the confidence in the identification provided by the bioinformatic search of the raw data. While these and other technology specifications have been important, the development of MS/MS (McLafferty, 1981) and the coupling of on-line protein/peptide separations with MS (Henion, 1978) have been the key determinants that have enabled proteomics. Tandem MS allows tryptic peptides obtained from an enzymatic digestion of a complex proteome mixture, to be fragmented in such a way that sufficient sequence information can be obtained for its unambiguously identification. While proteins are typically identified through MS/MS identification of
their peptide surrogates, methods to characterize intact proteins by MS/MS are becoming more and more popular. The coupling of on-line separations, most commonly RPLC, has enabled complex mixtures to be fractionation prior to MS analysis so that thousands of peptides can be identified within a few hours, as shown in Figure 14.3 (Washburn et al., 2001). The throughput by which proteins in complex mixtures are identified by LC-MS is unparalleled using any other technology: a critical parameter for discovery science in which large amounts of data are collected in order to find the important factors. There are many different types of commercially available mass spectrometers, and it can be confusing as to which type is most suitable for a specific application. For instance, ion-trap mass spectrometers equipped with electrospray ionization (ESI) sources are the “work-horse” instruments for the characterization of complex proteomes when solution-based separations are used. Time of flight (TOF) instruments are popular for identifying proteins separated using 2D-PAGE, primarily because of their speed and mass accuracy. The choice of what instrument to be utilized needs to be weighed against the type of research that is being focused on. How is MS able to identify so many peptides in a complex mixture? A schematic of the process is shown in Figure 14.3. The mass spectrometer is operated in what is referred to as a “data-dependent” mode of operation. In this mode, the instrument takes a series of “snapshots” of peptides eluting from a LC column into the mass analyzer. This scan is referred to as the MS scan. In between these snapshots, a series of MS/MS scans of peptides that are isolated within the instrument for CID are acquired. The instrument selects peptides for MS/MS analysis based on their intensity in the preceding MS scan. A datadependent experiment can perform 10 MS/MS scans between every MS scan. The mass spectrometer continues this cycle of MS scan followed by ten MS/MS scans throughout the entire LC/MS/MS experiment. Depending on the length of the LC separation, upwards of 7000 MS/MS sequencing attempts are made on peptides eluting from the column into the mass spectrometer. Bioinformatic analysis will result in 10–25% (i.e., 700–1750) of these sequencing attempts resulting in successful peptide identifications. A list of popular mass analyzers used in proteomics and their performance characteristics is given in Table 14.2.
BIOINFORMATICS Probably no area of proteomics evolves faster than bioinformatics. Proteomics is critically dependent on bioinformatics to process the raw mass spectral data into protein data. While routinely used by every proteomic laboratory, the most critical software programs are those that take peptide mapping and/or tandem MS results and determine the protein or peptide sequence that most closely matches the experimental data. The two most popular software packages for matching experimental MS data to peptide/protein sequences are MASCOT
Bioinformatics
■
177
1789.8007 Automatically selected for tandem MS 7000/hr
1789.8007
100
100 1750
1821.8724
1790
Relative abundance
Relative abundance
80 1810
60 40 1404.6632 20
1203.6808
0 1000
1400
1912.8071 2442.0809 2829.2381 2328.9829 1800
2200
2600
60 40 20 0
3000
MS/MS spectrum
80
400
800
1200 m/z
1600
2000
m/z
Relative abundance
100 80 60 40 Peptide identifications
20 0 20
40
60
80 Time (min)
100
120
140
Protein identifications
Figure 14.3 Principles of data-dependent tandem mass spectrometry (MS/MS). In data-dependent MS/MS, the mass spectrometer conducts an MS scan to record peptide masses that can be detected during a reversed-phase liquid chromatography (RPLC)-MS analysis of a complex proteome sample. The mass spectrometer isolates the peptide producing the most intense signal (i.e., m/z 1789.8007) and subjects it to collisional induced dissociation (CID). The resultant fragment ions are detected by the mass spectrometer and are subsequently used to identify the peptide in a bioinformatic analysis of the data. The mass spectrometer then isolates the next most intense peptide and collects its MS/MS spectrum. Linear ion-trap mass spectrometers may sequentially select on the order of 10 peptides for CID prior to recording the next MS scan. In the following MS scan, the instrument looks for new peptides that may have eluted from the LC column and selects these for CID and possible identification.
TABLE 14.2 Performance characteristics of various mass analyzers commonly used in proteomic research Analyzer
Sensitivity
Resolution
Mass accuracy
Ion-trap
Good
Low
Low
Linear ion-trap
Excellent
Low
Low
Triplequadrupole
Good
Good
Good
TOF
Good
High
High
FTICR
Excellent
High
High
(Lubec et al., 2005) and SEQUEST (Yates et al., 1995). As with most software algorithms, they both do what can also be done manually but at a much faster speed. For instance, SEQUEST is capable of transforming thousands of raw tandem MS files into potential peptide identifications within minutes, whereas manually assigning a single raw file would take on the order of hours. These software algorithms are also capable of identifying modified peptides when the mass added to the peptide as a result of some covalent modification (e.g., phosphorylation) is known. While not in widespread distribution, software packages, such as Scoring ALgorithm for Spectral Analysis (SALSA) developed by Daniel Leibler, provides a means of identifying unanticipated modifications, by identifying spectra that display characteristic
178
CHAPTER 14
■
Proteomics: The Deciphering of the Functional Genome
product ions, neutral or charged losses of groups of signals that are indicative of a particular modification (Hansen et al., 2001). Taken together, these varieties of software packages have been critical to the enabling of data analysis on the scale needed for discovery-driven proteomic studies. Throughout this chapter, the theme has been to use proteomic technologies to measure the relative abundances of proteins in different samples. Software has been developed to do this both at the gel-staining level and the mass spectral data level (i.e., when stable-isotope labeling quantitation is employed). Where the major lack has been is how this relative abundance data is presented. It is still typically displayed as lists of proteins with their relative abundance between two samples being given as a fold-difference. Unfortunately, with the ability to generate abundance data for thousands of proteins, these lists become an “unsolvable puzzle” when trying to identify major effects on a cell’s proteome as a result of a specific treatment. Recently, software packages such as Ingenuity™ have been developed to distill abundance data into annotated protein pathways enabling the probability that any specific signaling pathway has been affected can be statistically measured (www.ingenuity.com/products/ pathways_analysis.html). Software packages such as these also allow transcriptomic and proteomic data to be directly compared at both the single biomolecule and pathway levels.
IMPACT OF PROTEOMICS ON UNDERSTANDING DISEASES It is impossible to find a common characteristic that threads every disease together. One prevalent characteristic, primarily in the area of cancer, is that the earlier it is detected the better the chance of survival. To detect cancer at earlier stages is going to require novel biomarkers that possess high sensitivity and specificity. This need has driven much of the proteomic research over the past 3–5 years. Much of the search for novel biomarkers falls within discovery-driven research, and therefore MS has played a major role. The aim, in most studies, is to leverage the ability of MS to identify proteins in biological samples from patients with specific cancers and compare that to lists of proteins identified in the same sample types obtained from healthy-matched controls. Proteins that are found to be more prevalent in samples from disease-afflicted patients are considered as potential biomarkers. While on the surface the premise sounds simple enough, in reality there are several daunting challenges, both technological and physiological. Even with the speed at which MS is capable of identifying proteins, it still undersamples any complex proteome mixture. When such entities as post-translational modifications and splice variants are taken into account, it is impossible to know exactly the number of distinct proteins within a cell or biofluid proteome. Therefore, when a protein is detected in a diseased sample but not in a matched control, there is always that uncertainty as to whether the protein is not really present or it just was not detected in the present experiment. As is
commonly quoted in science, “the absence of evidence is not the evidence of absence”. On the physiological side, the fact that the proteome of any cell or biofluid is dynamic results in a constantly changing background that occurs both between different patients and within the same patient. For instance, the serum proteome of an individual is different before and after lunch. Compound this with the genetic and lifestyle variability that is observed between different patients and it is easy to grasp how difficult it is to confidently identify a disease-specific change in a protein in this shifting matrix. Regardless, the promise of technologies such as MS that can scan proteomes at levels never before possible propels investigators to search for new diagnostic markers. While the success in discovering new markers has been limited, they do however exist. For instance, a biomarker for the bladder disorder interstitial cystitis (IC) was found using a combination of protein fractionation and MS techniques (Keay et al., 2004). Previous to this report, no useful diagnostic marker for this disorder existed and patients were typically diagnosed by the exclusion of other maladies. This study resulted in the discovery of a novel nineresidue sialoglycopeptide that was shown to be present only in the supernatant of cultured bladder epithelial cells obtained from patients with IC and not in healthy controls or patients with bacterial cystitis. One of the more intriguing MS-based proteomic techniques to identify signatures of diseases in blood involves the detection of peptides that signify exo- and endo-protease activities (Villanueva et al., 2006). In this study, low molecular weight peptides were extracted from serum and analyzed using matrixassisted laser desorption ionization (MALDI) TOF MS. The study showed that a subset of the serum peptides were able to provide a signature allowing patients with three types of solid tumors to be classified separately from controls without cancer. Sixty-one of the signature peptides were identified and shown to fall into several tight clusters (Villanueva et al., 2006). Most of the identified peptides were generated by cancertype specific exopeptidase activities that enabled highly accurate classification of a set of prostate cancer samples used to validate the markers. While this method does not identify a novel diagnostic or therapeutic protein marker, it does use MS to reveal a cancerspecific activity found within serum of cancer-affected patients.
CONCLUSIONS While proteomics technology has advanced greatly in the past decade and allows scientists to characterize proteomes to an extent never before possible or imagined, this capability in itself brings a whole new set of challenges. The initial challenge is how to glean important clinical information from massive datasets that are amounts of data “contaminated” with a vast background of other less important proteins. While MS provides the ability to conduct comparative analyses of biological samples for the discovery of novel diagnostic and therapeutic biomarkers, it is unable to indicate which of the differentially abundant proteins may be
Recommended Resources
disease-specific. Many times these experiments represent proverbial “needle-in-the-haystack” searches with the additional complication that no information about the “needle” is known. While many potential “needles” can be found, the next area of development is to determine those that should graduate to validation phase. Since validation of any prospective biomarkers is expensive both in terms of time and money, continuing developments in optimal methods to both acquire and analyze proteomics data will be critical.
■
179
ACKNOWLEDGEMENTS This project has been funded in whole or in part with federal funds from the National Cancer Institute, National Institutes of Health, under Contract N01-CO-12400. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organization imply endorsement by the United States Government.
REFERENCES Gilham, P.T. (1970). RNA sequence analysis. Annu Rev Biochem 39, 227–250. Gygi, S.P., Rist, B., Gerber, S.A.,Turecek, F., Gelb, M.H. and Aebersold, R. (1999). Quantitative analysis of complex protein mixtures using isotope-coded affinity tags. Nat Biotechnol 17, 994–999. Hansen, B.T., Jones, J.A., Mason, D.E. and Liebler, D.C. (2001). SALSA: A pattern recognition algorithm to detect electrophile-adducted peptides by automated evaluation of CID spectra in LC-MS-MS analyses. Anal Chem 73, 1676–1683. Henion, J.D. (1978). Drug analysis by continuously monitored liquid chromatography/mass spectrometry with a quadrupole mass spectrometer. Anal Chem 50, 1687–1693. Hoorn, E.J., Hoffert, J.D. and Knepper, M.A. (2006). The application of DIGE-based proteomics to renal physiology. Nephron Physiol 104, 61–72. Issaq, H.J., Chan, K.C., Janini, G.M., Conrads,T.P. and Veenstra,T.D. (2005). Multidimensional separation of peptides for effective proteomic analysis. J Chromatogr B Analyt Technol Biomed Life Sci 817, 35–47. Julka, S. and Regnier, F.E. (2005). Recent advancements in differential proteomics based on stable isotope coding. Brief Funct Genomic Proteomic 4, 158–177. Keay, S.K., Szekely, Z., Conrads, T.P., Veenstra, T.D., Barchi, J.J., Jr., Zhang, C.O., Koch, K.R. and Michejda, C.J. (2004). An antiproliferative factor from interstitial cystitis patients is a frizzled 8 proteinrelated sialoglycopeptide. Proc Natl Acad Sci USA 101, 11803–11808. Lubec, G., Afjehi-Sadat, L., Yang, J.W. and John, J.P. (2005). Searching for hypothetical proteins: Theory and practice based upon original data and literature. Prog Neurobiol 77, 90–127.
McLafferty, F.W. (1981). Tandem mass spectrometry. Science 214, 280–287. O’Farrell, P.H. (1975). High resolution two-dimensional electrophoresis of proteins. J Biol Chem 250, 4007–4021. Ong, S.E. and Mann, M. (2005). Mass spectrometry-based proteomics turns quantitative. Nat Chem Biol 1, 252–262. Ong, S.E., Foster, L.J. and Mann, M. (2003). Mass spectrometric-based approaches in quantitative proteomics. Methods 29, 124–130. Van den Bergh, G. and Arckens, L. (2005). Recent advances in 2D electrophoresis: An array of possibilities. Expet Rev Proteomics 2, 243–252. Villanueva, J., Shaffer, D.R., Philip, J., Chaparro, C.A., ErdjumentBromage, H., Olshen, A.B., Fleisher, M., Lilja, H., Brogi, E., Boyd, J. et al. (2006). Differential exoprotease activities confer tumorspecific serum peptidome patterns. J Clin Invest 116, 271–284. Washburn, M.P., Wolters, D. and Yates, J.R. III (2001). Large-scale analysis of the yeast proteome by multidimensional protein identification technology. Nat Biotechnol 19, 242–247. Westermeier, R. and Marouga, R. (2005). Protein detection methods in proteomics research. Biosci Rep 25, 19–32. Yager, T.D., Nickerson, D.A. and Hood, L.E. (1991). The Human Genome Project: Creating an infrastructure for biology and medicine. Trends Biochem Sci 16, 454–458. Yates, J.R., Eng, J.K., McCormack, A.L. and Schieltz, D. (1995). Method to correlate tandem mass spectra of modified peptides to amino acid sequences in the protein database. Anal Chem 67, 1426–1436.
RECOMMENDED RESOURCES Books Proteomics for Biological Discovery by Timothy D. Veenstra and John R. Yates Introduction to Proteomics:Tools for the New Biology by Daniel C. Liebler The Proteomics Protocols Handbook by John M. Walker Proteomics in Cancer Research by Daniel C. Liebler Proteins and Proteomics: A Laboratory Manual by Richard Simpson
Journals Molecular and Cellular Proteomics (mcponline.org) Journal of Proteome Research (jpr.org)
Proteomics (proteomics.org) Proteomics – Clinical Applications
Websites http://expasy.org http://www.hupo.org http://www.eupa.org http://www.asms.org
CHAPTER
15 Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms Christopher B. Newgard, Robert D. Stevens, Brett R. Wenner, Shawn C. Burgess, Olga Ilkayeva, Michael J. Muehlbauer, A. Dean Sherry and James R. Bain
INTRODUC TION Despite decades of research, the molecular etiology of chronic and indolent diseases and conditions such as diabetes, obesity, cancer, and cardiovascular disease remains obscure, in part because they represent slowly developing phenotypes caused by the interplay of a complex mixture of genetic and environmental effectors. For similar reasons, it has been very difficult to detect and diagnose these diseases at a stage that allows sufficient time for effective intervention. Comprehensive measurement of intermediary metabolites and changes in metabolic activity (metabolic flux) could lead to improved detection and understanding of these diseases. As will be discussed in this chapter, instrumentation for measurement of large groups of intermediary metabolites as well as computational methods for analyzing such data are emerging. This surge in development of “metabolomics” technologies confers a number of potential advantages to scientists interested in disease research. First, it is estimated that human beings contain approximately 5000 discrete small molecule metabolites, far
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 180
less than the estimate of 25,000 genes and 100,000 proteins. Thus, metabolomics may ultimately be the most tractable of the “omics” sciences. Second, metabolomics measures changes in metabolic or chemical milieu that are downstream of genomic and proteomic alterations, therefore potentially providing the most integrated picture of biological status. Third, identification of metabolic fingerprints for specific diseases may have particular practical utility for development of therapies, because metabolic changes immediately suggest enzymatic drug targets. And fourth, metabolomics is likely to be a powerful and precise tool for discerning mechanisms of action and possible toxicological effects of drug therapies. The purpose of this chapter is to describe current methodologies for metabolic profiling and flux analysis and to provide selected examples of insights into biological and disease mechanisms gained from these approaches. Studies involving integration of metabolomics with genomic and transcriptomic profiling methods are just beginning to emerge, and early examples of this exciting area will also be highlighted. It is important
Copyright © 2009, Elsevier Inc. All rights reserved.
Current Metabolomics Platforms: Basic Tools and General Features
to emphasize from the outset that metabolic profiling tools are in a rapid phase of development, such that our description of currently useful methods will likely represent a “snapshot in time” that will evolve rapidly over the next several years.
■
181
100 P
L A %
CURRENT METABOLOMICS PLATFORMS: BASIC TOOLS AND GENERAL FEATURES As described in other chapters in this volume, genome-wide analysis of genetic variation (“genomics”) and surveys of changes in mRNA abundance by microarray technologies (“transcriptomics”) are examples of robust and relatively mature technologies for investigation of mechanisms underlying biological variability. For example, mRNA profiling has developed to a point that microarray cores have become common in both the academic and private sectors, with many thousands of RNA profiling experiments performed on a yearly basis. This is not yet the case for laboratories that focus on comprehensive metabolic analysis. One reason for this is the complexity inherent in attempting to measure small molecule metabolites (intermediary metabolites as discussed in this chapter are defined as nonpeptidic and with a molecular weight of 3000) in a quantitatively rigorous fashion. Underlying issues include the variable lability of metabolites (e.g., amino acids are often quite stable, whereas organic acid or lipid species can be quite labile), the chemical diversity of different metabolite classes, problems encountered in efficient extraction of metabolites from different biological matrices (e.g., tissues, blood, urine, etc.), and the wide dynamic concentration range of metabolite species (ranging from sub-nanomolar to milimolar). Thus, it is not surprising that there is currently no single technology for measurement of all of the metabolites in the “metabolome”. This means that the small number of existing laboratories that engage in comprehensive metabolic analysis have usually been constructed on an “ad hoc” basis, often driven by the direct technical experience and/or scientific goals and philosophies of the founding scientists. This leads to varying choices in instrument platforms, and places an even greater burden on individual laboratories for rigor in quantitative analysis, since standardized methods do not yet exist for the majority of analytes. The two major instrument platforms for measuring metabolite levels in biological samples are nuclear magnetic resonance (NMR) and mass spectrometry (MS) (Dettmer et al., 2007; German et al., 2005; Hollywood et al., 2006; Lindon et al., 2007; Watson, 2006). Both technologies can also be used for metabolic flux analysis following provision of stable isotope-labeled metabolic fuels (e.g., U-13C glucose) to cells and organisms, as discussed later. In general, when applying these technologies to assays of metabolite levels, research groups in the field tend to fall into two camps that adopt either “unbiased”/“top-down” approaches or “targeted”/”bottom-up” methods as their core technologies. Unbiased profiling involves use of either NMR
F
A
F
L VP GG
V
S S
M
Y Y
M H
E D D
E
0 140
160
180
200 m/z
220
240
260
Figure 15.1 Tandem mass spectrum of butyl esters of acidic and neutral amino acids in serum. Data were obtained using a neutral loss of 102 Da scan function. The figure illustrates the use of internal standards for “targeted” quantitative profiling of metabolites. Each amino acid in serum (shown by black letters) is measured against its own stable-isotope-labeled internal standard (shown in red letters). G, glycine, A, alanine, S, serine, P, proline, V, valine, L, leucine, M, methionine, H, histidine, F, phenylalanine, Y, tyrosine, D, aspartate, E, glutamate.
or MS for measurement of as many metabolites as possible in a biological specimen simultaneously, regardless of the chemical class of the metabolites. When applied to unbiased profiling, both NMR and MS have advantages and limitations, as discussed below, and neither technology can presently be used for surveying all of the metabolites in a sample in a quantitative fashion. In contrast, targeted profiling focuses on quantification of discrete clusters of chemically related metabolites (“modules”) using various combinations of chromatographic separations technologies and MS instruments that are most compatible with the analyte class. For example, electrospray ionization (ESI) and tandem mass spectrometry (MS/MS) are used for quantitative measurements of multiple amino acids and urea cycle intermediates or acylcarnitines (byproducts of mitochondrial oxidation of fatty acids, amino acids and glucose) of varying chain length and degree of saturation, whereas gas chromatography-MS (GCMS) is used routinely for measurement of multiple species of free fatty acids and organic acids. For each separate module, accurate quantification of the analytes is facilitated by addition of a cocktail of stable isotope labeled standards to the biological sample prior to the extraction and derivatization steps that may be necessary for the particular MS approach (see below and example shown in Figure 15.1). Ideally, each unknown analyte will be paired with its labeled cognate (usually an M+3 heavy isotope, e.g., methionine
182
CHAPTER 15
TABLE 15.1
■
Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms
Summary of mass spectrometry (MS) methods used for metabolic profiling
Technique
Mass Analyzer
Reference
Metabolites
GC/MS
Quadrupole
Fiehn et al. (2000) Roessner et al. (2001) Gullberg et al. (2004) Pope et al. (2007) Denkert et al. (2006)
Fatty acids, alcohols, sterols, aliphatic and aromatic acids, amino acids, sugars, sugar alcohols, phosphates, polyamines
An et al. (2004) Haqq et al. (2005) Allen et al. (2003) Scholz et al. (2004) Oikawa et al. (2006)
Acylcarnitines, amino acids, coenzyme-As,
Bajad et al. (2006) Sabatine et al. (2005) Han and Gross (2003) Murphy et al. (2005) Lafaye et al. (2003) Lafaye et al. (2004) Plumb et al. (2005) Lenz et al. (2007) Rainville et al. (2007)
Phoshpholipids, sphingolipids, eicosanoids, amino acids, sulfoconjugates, carboxylic acids, amines, glucuronides, glycosides, glutathione conjugates
Sato et al. (2004) Williams et al. (2007) McNally et al. (2006) McNally et al. (2007) Soga et al. (2006) Soga et al. (2006)
Amino acids, sugar nucleotides, carboxylic acids, sugar phosphates, amines
TOF
FIA or Infusion
QqQ TOF Q-TOF FT-ICR
LC/MS
QqQ
QIT TOF Q-TOF CE/MS
Quadrupole QqQ-IT TOF Q-TOF
Direct Ionization
MALDI-TOF MALDI-TOF/TOF DESI-QIT
Yu et al. (2006) Fraser et al. (2007) Sun et al. (2007) Cooks et al. (2006) Pan et al. (2007)
Amino acids, fatty acids, carotenoids, nucleosides, carboxylic acids
Abbreviations: GC, Gas Chromatography; MS, Mass Spectrometry; LC, Liquid Chromatography; CE, Capillary Electrophoresis; TOF, Time of Flight; FT-ICR, Fourier Transform Ion Cyclotron Resonance;Q, Quadrupole; QqQ Triple Quadrupole; QIT Quadrupole Ion Trap; MALDI, Matrix Assisted Laser Desorption Ionization; DESI, Desorption Electrospray Ionization.
with methyl-D3 methionine, pyruvate with 13C3-pyruvate, etc.) to control for differences in analyte loss during chemical extraction and/or derivatization of individual analyte species and to compensate for ionization suppression effects. In practice, the limited range of stable isotope-labeled metabolites available from commercial suppliers and their relatively high cost constitute a current impediment to more broadscale development and deployment of these technologies. To overcome this problem in the future, it may be necessary for metabolomics core laboratories to import chemical synthesis capabilities for production of comprehensive libraries of internal standards.
Table 15.1 provides a summary of various MS-based metabolic profiling technologies that are currently used for measurement of different metabolite classes.
COMPARISON OF NMR AND MS TECHNOLOGIES FOR UNBIASED METABOLIC PROFILING Nuclear magnetic resonance spectroscopy is theoretically an excellent tool for unbiased metabolic profiling of all small
MS Methods for Targeted Metabolic Profiling
molecule metabolites, since the method is based on detection of any molecules that contain carbon or hydrogen (German et al., 2005; Lindon et al., 2007). Moreover, the method is non-destructive, in that analyses can be conducted directly in bodily fluids, cells, and even in intact tissues without the need for chemical extraction or derivatization of the individual analytes. These advantages are off-set by significant technical challenges, including poor sensitivity (it is generally not possible to detect metabolites at sub-micromolar levels with NMR) and the difficulty of deconvoluting spectra of complex metabolite mixtures such as typically found in biological matrices like plasma, urine, or tissue extracts. When NMR is applied to such samples, all metabolites within the detection limits of the technology resolve as hundreds to thousands of proton magnetic resonance peaks. The complexity of these data and the absence of comprehensive libraries of NMR spectra make it very difficult to assign the different resonances to specific metabolites. Thus, although NMR spectra are rich with information, the full harvesting of these datasets is not yet possible. Instead, peaks within NMR datasets are often treated as independent statistical variables, and the data analyzed by methods such as principal components analysis (PCA) to identify clusters of resonances that characterize different biological or disease states. Mass spectrometry has the immediate advantage of much higher sensitivity compared to NMR, and the most advanced MS platforms such as Fourier transform ion cyclotron mass spectrometers (FT-ICR-MS) have the capacity to detect metabolites at concentrations as low as the femtomole range (Dettmer et al., 2007). Moreover, modern MS platforms have very high-mass resolution (100,000), can resolve literally thousands of individual small molecules without the need for chromatography, and can achieve mass accuracies of 1 ppm. Seemingly, current MS platforms could therefore provide a means of circumventing problems encountered when using NMR methods for unbiased metabolic profiling of complex biological matrices, as described earlier. In practice, however, MS techniques usually require a sample extraction step that can cause metabolite losses. Moreover, different kinds of MS instruments use varying methods for the generation and detection of ions, and in concert, different kinds of metabolites have discrete ionization properties and are therefore variably resolved and detected in instruments that rely on mass and charge for analyte separation (Dettmer et al., 2007). Thus, while several current MS instrument platforms can be used for resolution and detection of large numbers of metabolites (e.g., high-performance liquid chromatography-MS, Q-TOF-MS, and FT-ICR-MS), the variable lability, solubility, recovery, and ionization of the different analyte species, coupled with the lack of broad-coverage spectral libraries makes these unbiased methods inherently nonquantitative and useful mainly for detecting gross changes in spectral peaks in samples of different biological origin. Similar to NMR, masses in these complex MS spectra are often not assigned to a specific analyte, and the data are instead analyzed by PCA or other multi-factorial statistical tools. However, an
■
183
advantage of MS-based unbiased profiling methods is that when specific peaks or clusters of peaks consistently differentiate biological states (e.g., a metastatic tumor from a benign tumor), their detection by high-resolution MS platforms provides an immediate clue for analyte identification by providing an accurate mass estimate of the analyte. These mass estimates can be used to query databases such as KEGG, HMP, and ChemSpider for chemical formulae that match the estimated mass, thereby usually providing a strong starting point for identifying the analyte, and ultimately facilitating a return to more mechanistically oriented studies. The number of possible formulae is also significantly reduced by high-mass accuracy; for example, the number of possibilities for a mass 500 at a mass accuracy 5 ppm with an elemental palate CHNOS is 23. A mass accuracy of 2 ppm would reduce the number to 10 and an accuracy of 1 ppm would yield only 4 possibilities.
MS METHODS FOR TARGETED METABOLIC PROFILING For scientists with core interests in biological mechanisms rather than biomarkers, knowledge about the identity of metabolites being surveyed and their exact concentration greatly enhances the value of the datasets. This means that some metabolic profiling laboratories are built in a bottom-up fashion, with emphasis on targeted and quantitative MS-based analyses. Although these approaches tend to utilize simpler tools such as ESI-MS/MS and GC-MS, significant challenges must still be overcome in building rigorous and fully vetted analysis modules for groups of metabolites. For example, GC-MS, when coupled with the use of stable isotope standards, can provide sufficient chromatographic resolution and analyte-specific detection to allow quantification of groups of metabolites in a class (e.g., fatty acids of different change length and degree of saturation or organic acids/TCA cycle intermediates). However, because the chromatography is performed in the gas phase at high-temperature, analytes must be volatile and have sufficient thermal stability to survive the analysis. To help stabilize the analytes under study, reactive carboxyl, sulfhydryl, amine, or hydroxyl groups are derivatized by akylation, oximation, acylation, or silylation (Dettmer et al., 2007; German et al., 2005). These methods, while effective, add complexity beyond that already introduced by sample extraction to the analytical procedures, and can result in batchto-batch variation if not properly controlled. Similarly, analysis of acylcarnitines or amino acids/urea cycle intermediates by ESI-MS/MS requires different derivatization steps for each module – for example, treatment with acidic methanol for acylcarnitines and n-butanol for amino acids/urea cycle intermediates (Haqq et al., 2005). Nevertheless, with care, these modules can provide quantitative data for highly informative clusters of analytes (e.g., fatty acids, fatty acylCoAs, and acylcarnitines of different chain length and degree of saturation, amino acids/urea
184
CHAPTER 15
■
Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms
cycle intermediates, and organic acids), with coefficients of variation in replicate assays in the range of 10%.
EXAMPLES OF NMR-BASED METABOLIC PROFILING IN DISEASE RESEARCH The most common application of 1H-NMR-based metabolic profiling in mammals and humans to date has been for evaluation of toxic effects of xenobiotic drugs. Work in this area has been reviewed by others (Lindon et al., 2007), so we focus instead on newer applications in the realm of disease research. Coronary artery disease (CAD) is the leading cause of death in the United States and, in concert with the epidemic of obesity and diabetes, is rapidly becoming the leading cause of death in many developing countries. The genetic predilection of CAD is well established; family history has repeatedly been shown to be a robust, independent risk factor for CAD. Many of the commonly accepted risk factors for CAD are metabolic, such as lipid abnormalities and diabetes, making this disease a particularly logical target for comprehensive metabolic profiling technologies. Indeed, 1H NMR was recently reported to detect metabolic profiles that predicted the presence and severity of CAD (Brindle et al., 2002). The study involved comparison of a relatively small group of patients with triple vessel coronary heart disease (n 36) to angiographically normal subjects (n 30). Partial least squares-discriminant analysis was used to identify peaks in the major lipid regions of the spectra that appeared to provide separation between the groups. The specific lipid species involved were not identified by this analysis, although it was suggested that choline-containing metabolites were particularly diagnostic. However, a subsequent study using very similar techniques demonstrated that the predictive value of the NMR-based metabolic profiles was weak when other factors such as gender and use of medical interventions such as statins were taken into account (Kirschenlohr et al., 2006). This second group of authors demonstrated that the 1H NMR technique could identify male versus female subjects with 100% accuracy, but was much less able to identify statin users or subjects with CVD, despite expectations of substantial changes in lipid profile in the former group. This study makes it clear that 1H NMR is currently not a substitute for the more invasive procedure of angiography in detection of CVD. An intriguing and more promising recent application of 1 H NMR-based metabolic analysis has been to study the influence of intestinal bacteria (microbiota) on pathophysiologic outcomes of high-fat (HF) feeding (Dumas et al., 2006). This involved feeding of a HF diet to a mouse strain (129S6) known to be susceptible to HF-induced hepatic steatosis and insulin resistance, and for comparison, to a strain (BALBc) with relative resistance to these effects of diet. Metabolic profiling of plasma and urine samples revealed low circulating levels of plasma phosphatidylcholine (PChol) and high levels of methylamines
in urine in the 129S6 strain. Since the conversion of choline to methylamines is accomplished by gut microbiota rather than mammalian enzymes (Zeisel et al., 1983), and choline deficient diets are associated with hepatic steatosis (Buchman et al., 1995), the authors propose that the increased propensity of the 129S6 strain for hepatic steatosis and insulin resistance could be due to a reduced PC pool necessary for the assembly of VLDL particles, leading to deposition of triglycerides in liver. As with its application to diagnosis of CVD, the value of NMR for detecting the contributions of bacterial metabolism to whole animal metabolic profiles and disease development remains to be confirmed or refuted by an independent study. However, in this instance, other studies suggest that this could indeed be an important pathway. Thus, treatment of germ-free mice with microbiota from the cecum of conventionally raised animals produced an increase in body fat content and appearance of insulin resistance within 14 days of transfer (Backhed et al. 2004). In addition, more recent studies have revealed that genetically obese rodents or obese humans have changes in relative abundance of two main intestinal bacterial groups, the Bacteroidetes and the Firmicutes, relative to lean controls, and that the “obese microbiome” has an increased capacity to harvest energy from the diet (Turnbaugh et al., 2006). The mechanism by which the obese microbiome influences peripheral metabolism of the host seems to involve changes in circulating levels of fasting-induced adipose factor (FIAF), a circulating lipoprotein lipase inhibitor, as well as changes in activity of the important metabolic regulatory proteins PGC-1 and 5 AMPactivated kinase (AMPK) in skeletal muscle and liver (Backhed et al., 2007). An intriguing area for future study is to attempt to define metabolic profiles for specific gastrointestinal bacteria such that their relative contributions to changes in overall metabolic profile can be monitored and fully understood in the context of fuel homeostasis in the host.
EXAMPLES OF TARGETED MSBASED METABOLIC PROFILING FOR UNDERSTANDING OF DISEASE MECHANISMS Development of Tools The past decade has seen a rapid increase in the use of MS-based analysis platforms for metabolic profiling. Many of the early applications of MS technologies were in the area of plant biology (Fiehn et al., 2000) or for detecting inborn errors of metabolism in newborn children (Frazier et al., 2006; Shekhawat et al., 2005). Application of tandem mass spectrometry (MS/MS) to newborn screening has allowed detection of more than 40 different genetic diseases of lipid and amino acid metabolism (see Chapter 41). In newborn screening, less emphasis is placed on absolute quantification of multiple analytes in a module, since the screen only requires detection of changes
Examples of Targeted MS-Based Metabolic Profling for Understanding of Disease Mechanisms
in a single or discrete cluster of analytes with respect to established laboratory norms. For example, a defect in HMG-CoA lyase results in large and specific increases in C5-OH-carnitine and methylglutaryl-carnitine species detected by MS/MS, whereas defects in long-chain L-3-hydroxyacyl CoA dehydrogenase (LCHAD) or mitochondrial trifunctional protein (MTP) are associated with increases in C16-OH and C18OH-carnitine metabolites (Shekhawat et al., 2005). Similarly, defects in amino acid catabolizing enzymes such as phenylalaine hydroxylase to cause phenylketonuria, or the branchedchain -ketoacid dehydrogenase complex to cause Maple Syrup urine disease are readily detected by dramatic increases in phenylalanine or branched-chain amino acid (BCCA) levels, respectively. In more recent years, many of the MS-based targeted metabolic profiling techniques developed for diagnosing inborn errors of metabolism have been adopted and refined for studies of metabolic regulatory mechanisms, disease detection, and mechanisms of disease pathogenesis. Selective examples, meant to be illustrative rather than comprehensive, are provided. One approach has been to combine several targeted MSbased assay modules as an assemblage of tools that can report on several critical metabolic pathways (An et al., 2004; Haqq et al., 2005; Koves et al., 2008). Using a combination of GC/MS and MS/MS, we are currently able to perform quantitative analysis of approximately 150 metabolites in six groups: 1. free fatty acids of varying chain length and degrees of saturation; 2. total fatty acids (free+esterified); 3. acylcarnitines (representing products of mitochondrial fatty acid, amino acid, and glucose oxidation); 4. acyl CoAs of varying chain length and degree of saturation; 5. organic acids (TCA cycle intermediates and related metabolites), and 6. amino acids, including urea cycle intermediates. Although the total number of analytes measured with these tools is small relative to estimates of 5000 total metabolites in the “metabolome”, they are nevertheless highly useful for understanding changes in metabolic fuel selection under different physiologic and pathophysiologic circumstances. Moreover, expansion of the platform to include a broader range of analytes of interest in disease pathogenesis would appear possible in the near term, via adoption of published methods for analysis of sphingolipids (including ceramides), phospholipids, and prostaglandins and their metabolites (eicosanoids) (Dettmer et al., 2007; Han and Gross, 2005; Merrill et al., 2005; Watson, 2006). However, the reader should appreciate that development of new modules is not a trivial or inexpensive undertaking, as it requires acquisition or synthesis of stable isotope-labeled internal standards to cover most if not all of the individual analytes in the module, development of extraction procedures that are efficient for multiple analytes in a class, and demonstration of quantitative reproducibility of the method. The advantage gained is that such tools can be
■
185
applied to understanding of metabolic regulatory mechanisms in isolated cells, animal models of disease, and human disease states, as summarized in the following specific examples. Metabolic Profiling for Studies of Metabolic Regulatory and Signaling Mechanisms in Cultured Cells An example of integration of NMR- and MS-based tools for understanding of metabolic regulation and signaling mechanisms comes from recent work on the process of glucose-stimulated insulin secretion (GSIS) in pancreatic islet -cells (Boucher et al., 2004; Jensen et al., 2006; Joseph et al., 2006; Joseph et al., 2007; Lu et al., 2002; Ronnebaum et al., 2006; Ronnebaum et al., 2008). Stimulation of islet -cells with glucose causes increases in insulin secretion within seconds to minutes, and this response is mediated by signals that are generated by -cell glucose metabolism. A commonly accepted idea is that increases in the rate of glucose metabolism in -cells leads to increases in ATP:ADP ratio, which causes inhibition of ATP-sensitive K channels (KATP channels), membrane depolarization, and activation of voltage-gated calcium channels. The resulting influx of extracellular Ca2 is then thought to stimulate exocytosis of insulin containing secretory granules. However, this model clearly does not provide a complete description of signals that regulate GSIS, since pharmacologic or molecular inhibition of KATP-channel activity does not abrograte glucose regulation of insulin secretion (Henquin et al., 2003; Nenquin et al., 2004). To investigate non-KATP-channel related signals for GSIS, a set of insulinoma (INS-1)-derived cell lines with varying capacities for GSIS were developed (Hohmeier et al., 2000). Application of 13C NMR-based isotopomer analysis and MS-based profiling of intermediary metabolites led to the discovery of a critical link between pyruvate carboxylase (PC)mediated pyruvate exchange with TCA cycle intermediates (“pyruvate cycling”) and GSIS, and demonstration that these pathways are dysregulated in lipid-cultured and dysfunctional -cells (Boucher et al., 2004; Jensen et al., 2006; Joseph et al., 2006; Joseph et al., 2007; Lu et al., 2002; Ronnebaum et al., 2006). More recent studies have focused on identification of the specific pyruvate cycling pathways that may be involved in generation of signals for insulin secretion. One important pathway appears to involve export of citrate and/or isocitrate from the mitochondria via the citrate/isocitrate carrier (CIC), and subsequent conversion of isocitrate to -ketoglutarate ( -KG) by the cytosolic NAPD-dependent isoform of isocitrate dehydrogenase (ICDc) (Joseph et al., 2006; Ronnebaum et al., 2006). siRNA-mediated suppression of CIC or ICDc cause substantial impairment of GSIS in both insulinoma cell lines and primary rat islets. These studies suggest that a metabolic byproduct of pyruvate/isocitrate cycling may be an important amplifying signal for control of GSIS. Possible mediators that are under investigation include NADPH, -KG or its metabolites, or GTP generated by the succinyl CoA dehydrogenase reaction, all of which are direct or downstream products of the ICDc reaction.
186
CHAPTER 15
■
Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms
[U-13C] glucose
pyruvate PDH PC acetyl-CoA
PC ME 1 TCA
TCA αKG
glutamate C2
Figure 15.2 Use of 13C NMR for measurement of metabolic flux in cells. To measure flux of pyruvate into the TCA cycle via the oxidative enzyme pyruvate dehydrogenase (PDH) versus the anaplerotic enzyme pyruvate carboxylase (PC), [U-13C] glucose is administered to cells in culture. After 3 hours of incubation, cells are extracted and the NMR spectra of different carbons of glutamate are obtained. If all of the 13C-labeled glucose is converted to pyruvate to enter the TCA cycle through PDH, the spectrum of carbon 2 of glutamate will appear as shown at the bottom right of the figure. In contrast, if the labeled pyruvate enters the TCA cycle both through PDH and PC in equal proportions, the glutamate C2 spectrum will appear as shown at the top right of the figure. Arrows show significant difference in peak height in the two spectra. The change in the spectra under these two scenarios occurs because pyruvate that enters the TCA cycle via PC can engage in pyruvate recycling pathways that results in dilution of 13C (black circles) with natural abundance 12 C (gray circles), thus changing the pool of glutatmate mass isotopomers. Spectra of other glutamate carbons are obtained in the same fashion, and the aggregate data are deconvoluted to derive estimates of flux through PDH versus PC (Boucher et al., 2004; Lu et al., 2002; Sherry et al., 2004).
In the foregoing studies, measurement of pyruvate cycling flux was accomplished by incubation of -cells in low (3 mM) or high (12 mM) concentrations of U-13C glucose for several hours, followed by extraction of cells and analysis of the glutamate spectrum by NMR (see e.g., Figure 15.2). Specific resonances for each of the carbons of glutamate are affected by the population of mass isotopomers (glutamate with varying mixtures of 13C and 12C at each of the 5 carbons of the molecule), and this information can be used to calculate flux through the oxidative (PDH) and anaplerotic (PC) entry points of the tricarboxylic acid (TCA) cycle (Boucher et al., 2004; Lu et al., 2002; Sherry et al., 2004). When coupled to measurements of oxygen consumption (respiration), these rates can be expressed in
absolute rather than relative terms (Boucher et al., 2004; Sherry et al., 2004). These methods revealed that the capacity for GSIS in variously glucose responsive INS-1-derived cell lines was tightly correlated with PC-catalyzed pyruvate cycling activity, but not PDH-catalyzed glucose oxidation. To gain deeper understanding of the specific pyruvate cycling pathway involved in GSIS, these NMR-based methods of flux anlaysis were combined with MS/MS and GC/MS methods for profiling of acylcarnitines and organic acids, respectively. These experiments revealed that siRNA-mediated suppression of PC activity has little influence on pyruvate cycling rates or GSIS due to a compensatory rise in C2-acylcarnitine, which reflects the acetyl CoA pool; acetyl CoA is an allosteric activator of PC, which offsets the decrease in PC protein. These studies therefore uncovered a compensatory mechanism for preservation of pyruvate cycling activity and GSIS even when PC levels are reduced (Jensen et al., 2006). In another study, organic acid profiling was used to show that suppression of CIC caused a fall in cytosolic citrate levels, as would be expected if a major pathway for export of citrate from the mitochondria to the cytosol is blocked (Joseph et al., 2006). Finally, suppression of ICDc caused a decrease in pyruvate cycling flux, as measured by NMR, coupled with a rise in lactate levels measured by GC/MS, suggesting that in the face of suppressed pyruvate cycling activity, pyruvate is converted to lactate by default (Ronnebaum et al., 2006). These examples illustrate two important application of MS-based metabolic profiling in cellular research: (1) use in validation of specific genetic engineering or pharmacologic manipulations of cellular systems; (2) integration with flux analysis to provide a complete picture of changes in metabolic pathways under varying experimental conditions. Targeted MS-Based Metabolic Profiling Applied to Animal Models of Disease Targeted MS-based methods have also been used to gain understanding of metabolic control mechanisms in animal models of disease. Lipid infusion or the ingestion of a HF diet results in insulin resistance and eventual development of diabetes, but the mechanism by which this occurs has not been elucidated. In rats fed a HF diet, whole-animal, as well as muscle and liver insulin resistance are ameliorated in response to hepatic overexpression of a gene that affects lipid partitioning, malonyl-CoA decarboxylase (MCD) (An et al., 2004). Metabolic profiling of 37 acylcarnitine species by tandem mass spectrometry (MS/MS) revealed a decrease in the concentration of a lipid-derived metabolite, -OH-butyrylcarnitine (HB) in muscle of MCD-overexpressing animals, that was likely due to a change in intramuscular ketone metabolism. These findings spurred further investigation that led to a new model of diet-induced insulin resistance in muscle, involving accumulation of mitochondrially derived byproducts of lipid oxidation rather than diversion of lipid metabolites into cytosolic biosynthetic pathways (Muoio and Newgard, 2008). Supporting this idea are studies showing that chronic exposure
Examples of Targeted MS-Based Metabolic Profling for Understanding of Disease Mechanisms
of muscle to elevated lipids results in an increase rather than a decrease in expression of genes in the fatty acid -oxidative pathway (Koves et al., 2005). Importantly, this lipid-induced upregulation of the enzymatic machinery for -oxidation of fatty acids in muscle is not coordinated with upregulation of downstream metabolic pathways such as the TCA cycle and electron transport chain. This results in incomplete metabolism of fatty acids in the -oxidation pathway, accumulation of lipid-derived metabolites in the mitochondria, and a decrease in the levels of TCA cycle intermediates, as revealed by quantitative MS/MS and GC/MS analysis (Koves et al., 2005; Koves et al., 2008) (Figure 15.3). These abnormalities are reversed by exercise intervention in mice fed a HF diet, in association with increased TCA cycle activity and restoration of insulin sensitivity and glucose tolerance (Koves et al., 2005). Also supporting the model, knockout mice lacking MCD, which promotes fatty acid -oxidation, had markedly lower acylcarnitine levels in muscle and were protected against diet-induced insulin resistance, despite high levels of long-chain acyl CoAs (Koves et al., 2008). Conversely, transgenic mice with muscle-specific overexpression of PPAR, a nuclear receptor that activates -oxidative genes, developed both local and systemic glucose intolerance (Finck et al., 2005, Figure 15.3). The point to emphasize in the context of the current chapter is that the initial insights that led to this new model of muscle insulin resistance came mainly from metabolic profiling of liver and muscle tissue samples. These initial findings were then used to generate hypotheses that were tested by genetic engineering, nutritional modulation, and physical activity studies, illustrating that in addition to its traditional role as a disease screening tool, metabolic profiling can be used as a discovery tool for metabolic control mechanisms and disease pathogenesis. Application of MS-Based Metabolic Profiling for Understanding of Human Disease Pathogenesis Targeted MS-based metabolic profiling has also been increasingly applied to studies of human diseases and conditions. Specific examples of applications to the areas of obesity and cardiovascular disease (CVD) are discussed here, but these tools are also being applied to diverse areas such as cancer (Fan et al., 2006) and mental disorders (Kaddurah-Daouk et al., 2007). It has been appreciated for many years that obesity is associated with impaired metabolic regulation, including development of insulin resistance in muscle, liver, and adipose tissue, as well as loss of mass and function of pancreatic islet -cells (Muoio and Newgard, 2008). However, the specific mechanisms that link overnutrition and obesity to global metabolic dysfunction remain incompletely defined. We recently completed a metabolic analysis of 74 obese (median BMI of 36.6 kg/m2) and 67 lean (median BMI of 23.2 kg/m2) human subjects using the MS-based metabolic, hormonal and physiologic profiling tools described earlier (Haqq et al., 2005; Newgard et al., 2008). PCA was used to detect a metabolic signature comprised of changes
■
187
in amino acid, acylcarnitine, and organic acid metabolites that suggests that pathways of BCAA catabolism are overloaded in obese humans. To test the significance of this for development of obesity-related disease, we fed normal rats on HF, high-fat with supplemented BCAA (HF/BCAA) or standard chow (SC) diets. Despite a low rate of body weight gain equivalent to the SC group, HF/BCAA rats were equally insulin resistant as HF rats, demonstrating a contribution of BCAA to development of insulin resistance independent of body weight. However, the effects of BCAA on body weight and insulin sensitivity seem to require the presence of fat, since animals fed on SC diet with supplemented BCAA gain weight normally and have normal insulin sensitivity. These findings demonstrate a synergistic interaction of dietary protein and fat in the development of insulin resistance, and provide a demonstration of how broad-based metabolic profiling tools can identify unanticipated contributions of dietary and metabolic factors to disease development. MS-based metabolic profiling has also been applied to cardiovascular disease. In one study, liquid chromatography/triple quadrupole high-sensitivity MS analysis was applied to plasma samples of 18 subjects with exercise-inducible ischemia compared to 18 normal controls (Sabatine et al., 2005). Blood samples were taken immediately before and immediately after exercise, and levels of 173 known and several more minor intermediary metabolites were measured. A number of metabolites were found to be discordant between the two groups, including lactate, byproducts of AMP metabolism, metabolites of the citric acid cycle, and several of the unknowns. Using the six most discordantly regulated metabolites, a metabolic ischemia risk score was created and found to be statistically correlated to the probability of ischemia. However, it should be noted that the number of subjects was quite small in this study, necessitating follow up studies that either confirm or refute these interesting initial findings. In another recent study, MS/MS and GC/MS-based profiling of fatty acids, acylcarnitines, and amino acids was performed on plasma samples from 117 subjects from eight multiplex families with familial early-onset CVD (Shah et al., 2008). Using SOLAR and adjusting for variables such as diabetes, hypertension, dyslipidemia, body-mass-index (BMI), age and sex, high heritabilities were found for several amino acids (arginine, glutamate, alanine, ornithine, valine, leucine/isoleucine, h2 0.50–0.85, P 0.00004 7.2 1015), free fatty acids (arachidonic, linoleic, h2 0.57–0.59, P 0.0004 0.00005), and acylcarnitines (h2 0.54–0.86, P 0.002 0.0000002). Moreover, PCA was used to identify several metabolite clusters, with several of these factors found to be highly heritable. Interestingly, GENECARD families showed two distinct metabolite profiles, tracking with their clinical characteristics and reflective of factors identified from PCA, suggesting different genetic backgrounds and consequent variation in control of key metabolic pathways that converge on CVD. However, once again, the utility of these profiles for predicting CVD in other high-risk families or in the general population will require further study.
188
CHAPTER 15
■
Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms
(a)
C20, C18:2, C18:1, C16 Fatty Acids
37 acylcarnitine metabolites
Intermediates (C3-C14) CATs CATs C2 (acetylcarnitine)
βOX
CATs
TCA
CoA
AJD11-3 dm031303a24 1 (1.694) Sm (Mn, 20.75) 100
IS
6
218
417
C2
Mitochondria
Acylcarnitines %
C4OH
305
C18:1
262 249
440 418
235
276
414
290
221 246 260 274
Tissue or serum extracts
438 441 456 462
306
430
0 200 220 240 260 280 300 320 340 360 380 400 420 440 460 480 500 m/z
Muscle acylcarnitines (nmol/mg protein)
(b)
75
3.0
2.5 MCD (Active) 2.0
MCDmut (Inactive)
1.5
* 1.0
0.5
C2 C3 C4 C5:1 C5 *C4-OH C6 C5-OH C4-DC C8:1 C8 C5-DC C6-DC C10:3 C10:2 C10:1 C10 C8-DC C12 C12-OH C10-DC C14:2 C14:1 C14 C14-OH C12-DC C16 C16-OH C18:2 C18:1 C18 C18:1-OH C18-OH C20
0
Figure 15.3 Use of tandem mass spectrometry to profile acylcarnitines in an animal model of insulin resistance. (a). Acylcarnitines provide a profile of products of mitochondrial fatty acid oxidation. As long-chain fatty acyl CoAs enter the mitochondria through carnitine palmitoyltransferase I (CPT1), they are oxidized by successive removal of two carbon acetyl CoA units to generate CoA esters ranging from 2 to 20 carbons in length and with varying degrees of sauration. These CoA intermediates are converted to carnitine esters, which are readily extracted from tissues or measured in plasma. The extracts are processed and analyzed by ESI-MS/MS to produce spectra such as that shown at the right of the panel. (b). Acylcarnitines in muscle samples taken from high-fat fed, insulinresistant rats (black bars) or high-fat fed rats rendered insulin sensitive by hepatic expression of malonyl CoA decarboxylase (MCD) (white bars). Note the dramatic decrease in concentration of one particular acylcarnitine, -hydroxybutyryl carnitine (labeled C4-OH). Data are from An et al., 2004.
References
INTEGRATION OF METABOLIC PROFILING WITH OTHER “OMICS” TECHNOLOGIES Genetic linkage and association studies have the power to establish a causal link between gene loci and physiological traits (see Chapter 8). These studies can make novel connections between biological processes that would not otherwise be predictable based on current knowledge. Numerous quantitative trait loci (QTL) influencing disease-related phenotypes have been identified through gene mapping and positional cloning. While it has become relatively straightforward to map a phenotype to a broad-genomic region, identification of the individual gene(s) and variants responsible for the phenotype remains difficult. Moreover, traditional QTL mapping is based on association with physiological phenotype, but often does not reveal the molecular pathways leading to that phenotype. One way to overcome these limitations is to broadly expand the types of phenotypes analyzed in genetic screens. For example, with microarray technology, one can measure the abundance of virtually all mRNAs in a segregating sample. Importantly, mRNA abundance shows sufficient heritability in outbred populations or experimental crosses to allow mapping of gene loci that control gene expression, termed expression QTL (eQTL) (Schadt et al., 2005). When eQTL co-localize with a physiological QTL, one can hypothesize a shared regulator and offer a potential pathway leading to the physiological trait. The pathway between a QTL and a physiological trait often involves changes in the steady-state levels of metabolic intermediates, in addition to changes in mRNA abundance. Recent studies in plants (Wentzell et al., 2007) and in animal models of disease (Ferrara et al., 2008) have demonstrated that metabolite profiles show sufficient heritability to allow mapping of metabolic quantitative trait loci (mQTL), thereby facilitating integration of metabolic profiles with genomic and transcriptomic analyses. For example, one recent study mapped expression QTLs controlling glucosinolate content in Arabidopsis Bay–Sha recombinant lines, and found a strong correlation between loci controlling enzymes involved in glucosinolate synthesis and the levels of metabolites within this particular pathway (Wentzell et al., 2007). Similarly, analysis of mRNA expression and metabolic profiles in liver samples obtained in an F2 intercross between the diabetes-resistant C57BL/6 leptinob/ob and the diabetessusceptible BTBR leptinob/ob mouse strains was used to show that
■
189
liver metabolites map to distinct genetic regions, thereby demonstrating that tissue metabolite profiles are heritable (Ferrara et al., 2008). This study also integrated liver microarray and metabolic profiling data to construct putative regulatory networks for control of specific metabolic processes in liver. As a proof of principle of the practical significance of this multi-disciplinary approach, a regulatory network that links gene expression and metabolic changes in the context of glutamate metabolism was identified, and the validity of the approach was validated by experiments showing that genes in the network respond to changes in glutamine and glutamate availability. Although further development of these methods will be required, they already appear to have the potential to unearth regulatory networks that are altered to contribute to chronic, complex, and highly prevalent diseases and conditions such as obesity, diabetes, and cardiovascular disease.
FUTURE DIRECTIONS In the post-sequence era, biologists and translational investigators alike have gained a new appreciation for metabolic analysis as a critical tool for assessing the physiologic and pathophysiologic impact of genetic variation. The current growth in instrument and methods development in the field of metabolomics is built on the foundation of decades of analytical biochemistry. The major difference between then and now is that the current emphasis is on methods that allow simultaneous measurement of multiple analytes in a biological sample, whereas earlier work was often focused on one or a small number of metabolites per assay. Despite these advances, no single profiling method currently allows simultaneous analysis of all of the metabolites in the metabolome. Ultimate achievement of this goal will require continued intensive development of deeper libraries of chemical standards, instrument platforms with broad-sensitivity range and high-mass accuracy, and likely, integration of MS and NMR methods to gain full analyte coverage. These advances must be coupled to continued development of computational methods for analysis of complex datasets and their integration with equally complex genomic, transcriptomic, and proteomic profiles. The examples provided in this chapter of new scientific insights gained by application of current tools suggest a broad horizon and provide strong encouragement for further technology development in this area.
REFERENCES Allen, J., Davey, H.M., Broadhurst, D., Rowland, J.J., Oliver, S.G. and Kell, D.B. (Jun 2003). High-throughput classification of yeast mutants for functional genomics using metabolic footprinting. Nat Biotechnol 21(6), 692–696.
An, J., Muoio, D.M., Shiota, M., Fujimoto, Y., Cline, G.W., Shulman, G.I., Stevens, R., Millington, D. and Newgard, C.B. (2004). Hepatic expression of malonyl CoA decarboxylase reverses muscle, liver, and whole-animal insulin resistance. Nat Med 10, 268–274.
190
CHAPTER 15
■
Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms
Backhed, F., Ding, H., Wang, T., Hooper, L.V., Koh, G.Y., Nagy, A., Semenkovich, C.F. and Gordon, G.I. (2004). The gut microbiota as an environmental factor that regulates fat storage. Proc Natl Acad Sci USA 101, 15718–15723. Backhed, F., Manchester, J.K., Semenkovich, C.F. and Gordon, J.I. (2007). Mechanisms underlying the resistance to diet-induced obesity in germ-free mice. Proc Natl Acad Sci USA 104, 979–984. Bajad, S.U., Lu, W., Kimball, E.H., Yuan, J., Peterson, C. and Rabinowitz, J.D. (2006). Separation and quantitation of water soluble cellular metabolites by hydrophilic interaction chromatography-tandem mass spectrometry. J Chromatogr 1125, 76–88. Boucher, A., Lu, D., Burgess, S., Telemaque-Potts, S., Jensen, M., Mulder, H., Wang, M-Y., Unger, R.H., Sherry, A.D. and Newgard, C.B. (2004). Mechanism of lipid-induced impairment of glucose-stimulated insulin secretion and its reversal by an analogue of malate. J Biol Chem 279, 27263–27271. Brindle, J.T., Antti, H., Holmes, E., Tranter, G., Nicholson, J.K., Bethell, H.W.L., Clarke, S., Schofield, P.M., McKilligin, E., Mosedale, D.E. et al. (2002). Rapid and noninvasive diagnosis of the presence and severity of coronary heart disease using 1HNMR-based metabonomics. Nat Med 8, 1439–1444. Buchman, A.L., Dubin, M.D., Moukarzel, A.A., Jenden, D.J., Roch, M., Rice, K.M., Gornbein, J. and Ament, M.E. (1995). Hepatology 22, 1399–1403. Cooks, R.G., Ouyang, Z.,Takats, Z. and Wiseman, J.M. (2006). Detection Technologies. Ambient mass spectrometry. Science 311, 1566–1570. Denkert, C., Budczies, J., Kind, T., Weichert, W., Tablack, P., Schouli, J., Niesporek, S., K¯onsgen, D., Dietel, M. and Fiehn, O. (Nov 15 2006). Mass spectrometry-based metabolic profiling reveals different metabolite patterns in invasive ovarian carcinomas and ovarian borderline tumors. Cancer Res 66(22), 10795–10804. Dettmer, K.,Aronov, P.A. and Hammock, B.D. (2007). Mass spectrometrybased metabolomics. Mass Spectrom Rev 26, 51–78. Dumas, M.-E., Barton, R.H., Toye, A., Cloarec, O., Blancher, C., Rothwell, A., Fearnside, J., Tatoud, R., Blanc, V., Lindon, J.C. et al. (2006). Metabolic profiling reveals a contribution of gut microbiota to fatty liver phenotype in insulin-resistant mice. Proc Natl Acad Sci USA 103, 12511–12516. Fan, T.W., Higashi, R.M. and Lane, A.N. (2006). Integrating metabolomics and transcriptomics for probing SE anticancer mechanisms. Drug Metab Rev 38, 707–732. Ferrara, C.T., Wang, P., Neto, E.C., Stevens, R.D., Bain, J.R., Wenner, B.R., Ilkayeva, O.R., Keller, M.P., Blasiole, D.A., Kendziorski, C.M., Yandell, B.S., Newgard, C.B. and Attie, A.D. (Mar 2008). Genetic networks of liver metabolism revealed by integration of metabolic and transcriptomic profiling. PLoS Genet 4(3), e1000034. Fiehn, O., Kopka, J., Dormann, P., Altmann, T., Trethewey, R.N. and Willmitzer, L. (2000). Metabolite profiling for plant functional genomics. Nat Biotechnol 18, 1157–1161. Finck, B.N. et al. (2005). A potential link between muscle peroxisome proliferator-activated receptor-a signaling and obesity-related diabetes. Cell Metabol 1, 133–144. Fraser, P.D., Enfissi, E.M., Goodfellow, M., Eguchi, T. and Bramley, P.M. (2007). Metabolite profiling of plant carotenoids using the matrixassisted laser desorption ionization time-of-flight mass spectrometry. Plant J 49, 552–564. Frazier, D.M., Millington, D.S., McCandless, S.E., Koeberl, D.D., Weavil, S.D., Chaing, S.H. and Muenzer, J. (2006). The tandem mass
spectrometry newborn screening experience in North Carolina: 1997–2005. J Inherit Metab Dis 29, 76–85. German, J.B., Hammock, B.D. and Watkins, S.M. (2005). Metabolomics: Building on a century of biochemistry to guide human health. Metabolomics 1, 3–9. Gullberg, J., Jonsson, P., Nordstrom, A., Sjostrom, M. and Moritz, T. (2004). Design of experiments: An efficient strategy to identify factors influencing extraction and derivatization of Arabidopsis thaliana samples in metabolomic studies with gas chromatography/mass spectrometry. Anal Biochem 331, 283–295. Han, X. and Gross, R.W. (2003). Global analyses of cellular lipidomes directly from crude extracts of biological samples by ESI mass spectrometry: A bridge to lipidomics. J Lipid Res 44, 1071–1079. Han, X. and Gross, R.W. (2005). Shotgun lipidomics: Electrospray ionization mass spectrometric analysis and quantitation of cellular lipodomes directly from crude extracts of biological samples. Mass Spectrom Rev 24, 367–412. Haqq, A., Lien, L.F., Boan, J., Arlotto, M., Slentz, C.A., Muehlbauer, M.J., Rochon, J., Gallup, D., McMahon, R.L., Bain, J.R. et al. (2005). The Study of the Effects of Diet on Metabolism and Nutrition (STEDMAN) Project: Rationale and Design. Contem Clin Trials 26, 616–625. Henquin, J.C., Ravier, M.A., Nenquin, M., Jonas, J.C. and Gilon, P. (2003). Hierarchy of the beta-cell signals controlling insulin secretion. Eur J Clin Invest 33, 742–750. Hohmeier, H.E., Mulder, H., Chen, G., Henkel-Rieger, R., Prentki, M. and Newgard, C.B. (2000). Isolation of INS-1-derived cell lines with robust ATP-sensitive K+ channel-dependent and -independent glucose-stimulated insulin secretion. Diabetes 49, 424–430. Hollywood, K., Brison, D.R. and Goodacre, R. (2006). Metabolomics: Current technologies and future trends. Proteomics 6, 4716–4723. Jensen, M.V., Joseph, J., Ilkayeva, O., Burgess, S., Lu, D., Ronnebaum, S., Odegaard, M., Becker, T.C., Sherry, A.D. and Newgard, C.B. (2006). Compensatory responses to pyruvate carboxylase suppression in islet beta-cells: Preservation of glucose-stimulated insulin secretion. J Biol Chem 281, 22342–22351. Joseph, J.W., Jensen, M.V., Ilkayeva, O., Palmieri, F., Alarcon, C., Rhodes, C.J. and Newgard, C.B. (2006). The mitochondrial citrate carrier plays a regulatory role in glucose-stimulated insulin secretion. J Biol Chem 281, 35624–35632. Joseph, J.W., Odegaard, M.L., Ronnebaum, S.M., Burgess, S.C., Muehlbauer, J., Sherry, A.D. and Newgard, C.B. (2007). Normal flux through ATP-citrate lyase or fatty acid synthase is not required for glucose-stimulated insulin secretion. J Biol Chem. 282, 31592–31600. Kaddurah-Daouk, R., McEvoy, J., Baillie, R.A., Lee, D., Yao, J.K., Doraiswamy, P.M. and Krishnan, K.R. (2007). Metabolomic mapping of atypical antipsychotic effects in schizophrenia. Mol Psychiatry 12, 934–945. Kirschenlohr, H.L., Griffin, J.L., Clarke, S.C., Rhydwen, R., Grace, A.A., Schofield, P.M., Brindle, K.M. and Metcalfe, J.C. (2006). Proton NMR analysis of plasma is a weak predictor of coronary artery disease. Nat Med 12, 705–710. Koves, T.R., Li, P., An, J., Slentz, D., Ilkayeva, O., Akimoto, T., Dohm, G.L., Yan, Z., Newgard, C.B. and Muoio, D.M. (2005). PGC1-mediated metabolic remodeling of skeletal muscle mimics exercise training and reverses lipid-induced mitochondrial inefficiency. J Biol Chem 280, 33588–33598.
References
Koves, T.R., Ussher, J.R., Noland, R.C., Slentz, D.S., Mosedale, M., Ikayeva, O.J., Bain, J., Stevens, R., Dyck, J.R., Newgard, C.B., Lopaschuk, G. and Muoio, D.M. (2008). Mitochondrial overload and incomplete fatty acid oxidation contribute to skeletal muscle insulin resistance. Cell Metab 7, 45–56. Lafaye, A., Junot, C., Ramounet-Le Gall, B., Fritsch, P., Tabet, J.C. and Ezan, E. (2003). Metabolite profiling in rat urine by liquid chromatography/electrospray ion trap mass spectrometry. Application to the study of heavy metal toxicity. Rapid Commun Mass Spectrom 17, 2541–2549. Lafaye, A., Junot, C., Ramounet-Le Gall, B., Fritsch, P., Ezan, E. and Tabet, J.C. (2004). Profiling of sulfoconjugates in urine by using precursor ion and neutral loss scans in tandem mass spectrometry. Application to the investigation of heavy metal toxicity in rats. J Mass Spectrom 39, 655–664. Lenz, E.M., Williams, R.E., Sidaway, J. et al. (2007). The application of microbore UPLC/oa-TOF-MS and 1H NMR spectroscopy to the metabonomic analysis of rat urine following the intravenous administration of pravastatin. J Pharm Biomed Anal 44, 845–852. Lindon, J.C., Holmes, E. and Nicholson, J.K. (2007). Metabonomics in pharmaceutical R & D. FEBS J 274, 1140–1151. Lu, D., Mulder, H., Zhao, P., Burgess, S.C., Jensen, M.V., Kamzolova, S., Newgard, C.B. and Sherry, A.D. (2002). 13C NMR isotopomer analysis reveals a connection between pyruvate cycling and glucose-stimulated insulin secretion (GSIS). Proc Natl Acad Sci USA 99, 2708–2713. McNally, D.J., Hui, J.P., Aubry, A.J. et al. (2006). Functional characterization of the flagellar glycosylation locus in Campylobacter jejuni 81–176 using a focused metabolomics approach. J Biol Chem 281, 18489–18498. McNally, D.J., Aubry, A.J., Hui, J.P. et al. (2007). Targeted metabolomics analysis of Campylobacter coli VC167 reveals legionaminic acid derivatives as novel flagellar glycans. J Biol Chem 282, 14463–14475. Merrill, A.H., Jr., Sullards, M.C., Allegood, J.C., Kelly, S. and Wang, E. (2005). Sphingolipidomics: high-throughput, structure-specific, and quantitative analysis of sphingolipids by liquid chromatography tandem mass spectrometry. Methods 36, 207–224. Muoio, D.M. and Newgard, C.B. (2008). Mechanisms of disease: Molecular and metabolic mechanisms of insulin resistance and cell failure in type 2 diabetes. Nat Rev Mollar Cell Biol 9, 193–205. Murphy, R.C., Barkley, R.M., Zemski, Berry, K., Hankin, J., Harrison, K., Johnson, C., Krank, J., McAnoy, A., Uhlson, C. and Zarini, S. (2005). Electrospray ionization and tandom mass spectrometry of eicosanoids. Anal Biochem 346, 1–42. Nenquin, M., Szollosi, A., Aguilar-Bryan, L., Bryan, J. and Henquin, J.C. (2004). Both triggering and amplifying pathways contribute to fuel-induced insulin secretion in the absence of sulfonylurea receptor-1 in pancreatic beta-cells. J Biol Chem 279, 32316–32324. Newgard, C.B., Bain, J.R., Muehlbauer, M.J., McMahon, R.L., Rochon, J., Gallup, D., Stevens, R., Ilkayeva, O., Millington, D., Butler, M.D. et al. (2007). Comprehensive metabolic profiling of obese versus lean humans reveals a signature of overloaded branch chain amino acid catabolism. Submitted for publication. Oikawa, A., Nakamura, Y., Ogura, T. et al. (2006). Clarification of pathway-specific inhibition by Fourier transform ion cyclotron resonance/mass spectrometry-based metabolic phenotyping studies. Plant Physiol 142, 398–413.
■
191
Pan, Z., Gu, H., Talaty, N. et al. (2007). Principal component analysis of urine metabolites detected by NMR and DESI-MS in patients with inborn errors of metabolism. Anal Bioanal Chem 387, 539–549. Plumb, R.S., Granger, J.H., Stumpf, C.L. et al. (2005). A rapid screening approach to metabonomics using UPLC and oa-TOF mass spectrometry: Application to age, gender and diurnal variation in normal/Zucker obese rats and black, white and nude mice. Analyst 130, 844–849. Pope, G.A., MacKenzie, D.A., Defernez, M. et al. (2007). Metabolic footprinting as a tool for discriminating between brewing yeasts. Yeast 24, 667–679. Rainville, P.D., Stumpf, C.L., Shockcor, J.P., Plumb, R.S. and Nicholson, J.K. (2007). Novel application of reversed-phase UPLC-oaTOF-MS for lipid analysis in complex biological mixtures: A new tool for lipidomics. J Proteome Res 6, 552–558. Ronnebaum, S., Joseph, J.W., Burgess, S.C., Lu, D., Ilkayeva, O., Stevens, R., Becker, T.C., Muehlbauer, J., Sherry, A.D., Newgard, C.B. et al. (2006). A pyruvate cycling pathway involving cytosolic NAPD-dependent isocitrate dehydrogenase regulates glucosestimulated insulin secretion. J Biol Chem 281, 31041–31049. Roessner, U., Luedemann, A., Brust, D. et al. (2001). Metabolic profiling allows comprehensive phenotyping of genetically or environmentally modified plant systems. Plant Cell 13, 11–29. Sabatine, M.S., Liu, E., Morrow, D.A., Heller, E., McCarroll, R., Wiegand, R., Berriz, G.F., Roth, F.P. and Gerszten, R.E. (2005). Metabolomic identification of novel biomarkers of myocardial ischemia. Circulation 112, 3868–3875. Sato, S., Soga, T., Nishioka, T. and Tomita, M. (Oct 2004). Simultaneous determination of the main metabolites in rice leaves using capillary electrophoresis mass spectrometry and capillary electrophoresis diode array detection. Plant J 40(1), 151–163. Schadt, E.E. et al. (2005). An integrative genomics approach to infer causal associations between gene expression and disease. Nat Genet 37, 710–717. Shah, S.H., Hauser, E.R., Bain, J.R., Muehlbauer, M.J., Stevens, R.D., Wenner, B.R., Sketch, M., Dowdy, E., Granger, C.B., Ginsburg, G. S. et al. (2007). High heritability of metabolomic profiles in families burdened with early-onset cardiovascular disease. Submitted for publication. Shekhawat, P.S., Matern, D. and Strauss, A.W. (2005). Fetal fatty acid oxidation disorders, their effect on maternal health and neonatal outcome: Impact of expanded newborn screening on their diagnosis and management. Ped Res 57, 78R–86R. Scholz, M., Gatzek, S., Sterling, A., Fiehn, O. and Selbig, J. (2004). Metabolite fingerprinting: detecting biological features by independent component analysis. Bioinformatics 20, 2447–2454. Sherry, A.D., Jeffrey, F.M. and Malloy, C.R. (2004). Analytical solutions for (13)C isotopomer analysis of complex metabolic conditions: Substrate oxidation, multiple pyruvate cycles, and gluconeogenesis. Metab Eng 6, 12–24. Soga, T., Baran, R., Suematsu, M. et al. (2006). Differential metabolomics reveals ophthalmic acid as an oxidative stress biomarker indicating hepatic glutathione consumption. J Biol Chem 281, 16768–16776. Sun, G., Yang, K., Zhao, Z., Guan, S., Han, X. and Gross, R.W. (2007). Shotgun metabolomics approach for the analysis of negatively charged water-soluble cellular metabolites from mouse heart tissue. Anal Chem 79, 6629–6640.
192
CHAPTER 15
■
Comprehensive Metabolic Analysis for Understanding of Disease Mechanisms
Turnbaugh, P.J., Ley, R.E., Mahowald, M.A., Magrini, V., Mardis, E.R. and Gordon, J.I. (2006). An obesity-associated gut microbiome with increased capacity for energy harvest. Nature 444, 1027–1031. Vaidyanathan, S., Kell, D.B. and Goodacre, R. (2002). Flow-injection electrospray ionization mass spectrometry of crude cell extracts for high-throughput bacterial identification. J Am Soc Mass Spectrom 13, 118–128. Vaidyanathan, S., Gaskell, S. and Goodacre, R. (2006). Matrixsuppressed laser desorption/ionisation mass spectrometry and its suitability for metabolome analyses. Rapid Commun Mass Spectrom 20, 1192–1198. Watson, A.D. (2006). Lipidomics: A global approach to lipid analysis in biological systems. J. Lipid Res 47, 2101–2111. Wentzell, A.M., Rowe, H.C., Hansen, B.G., Ticconi, C., Halkier, B. A. and Kliebenstein, D.J. (2007). Linking metabolic QTLs with
network and cis-eQTLs controlling biosynthetic pathways. PLoS Genet 3, 1687–1701. Williams, B.J., Cameron, C.J., Workman, R., Broeckling, C.D., Sumner, L.W. and Smith, J.T. (2007). Amino acid profiling in plant cell cultures: An inter-laboratory comparison of CE-MS and GC-MS. Electrophoresis 28, 1371–1379. Yu, H., Lopez, E., Young, S.W., Luo, J., Tian, H. and Cao, P. (2000). Quantitative analysis of free fatty acids in rat plasma using matrixassisted laser desorption/ionization time-of-flight mass spectrometry with meso-tetrakis porphyrin as matrix. Anal Biochem 354, 182–191. Zeisel, S.H., Wishnok, J.S. and Blusztajn, J.K. (1983). J Pharmacol Exp Ther 225, 320–324.
CHAPTER
16 Comprehensive Analysis of Gene Function: RNA interference and Chemical Genomics Bjorn T. Gjertsen and James B. Lorens
INTRODUCTION Geneticists have for decades mutagenized cells and model organisms to induce phenotypic changes essential for our understanding of gene function (Bier 2005; Peters et al., 2003). Mutations engendering these cell behavioral changes have been mapped and the cognate gene products identified. Pharmaceutical scientists have for decades conducted diversity screens to identify chemical entities that affect biological function (Landro et al., 2000). Large libraries of small molecule compounds that cover a diverse chemical space have been interrogated in high-throughput screening schemes. Compounds that inhibit critical protein functions have been revealed. In recent years molecular biologists have combined these principles, developing different types of dominant genetic effectors (“perturbagens”) that affect gene function at multiple levels (Bredel and Jacoby, 2004; Kawasumi and Nghiem, 2007; Knight and Shokat, 2007; Rognan, 2007). This conflation of high-throughput chemical diversity screening and trans-dominant genetics is particularly evident in the recent advent of RNA interference (RNAi) and chemical genomic approaches with promising clinical translation. For example, the Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
identification of tumor-specific gene functional-dependence for oncogenic growth led to the development of imatinib mesylate, a small molecule inhibitor of the Bcr-Abl fusion protein in chronic myelogeous leukemia (Hunter, 2007; Sherbenou and Druker, 2007). This type of synthetic lethality via small molecule or modulation of gene function through RNAi is a promising strategy for future therapeutic development (Kaelin, 2005;Westbrook et al., 2005). Gene function analysis needs to be considered in conjunction with currently available technology. Recent technological development allows us to elucidate gene function more extensively, particularly in mammalian systems.Two recent approaches – RNAi and chemical genomics – provide new opportunities to elucidate the role of specific genes in complex biological systems. RNAi has rapidly evolved from a curious genetic phenomenon to the most widely used gene function analysis tool in less than a decade (Fire et al., 1998; Mello and Conte, 2004). Chemical genomics has established the use of small molecule libraries and high-throughput screening, a domain previously unique to the pharmaceutical industry, to delineate gene functions in diverse biological systems (Knight and Shokat, 2007). Copyright © 2009, Elsevier Inc. All rights reserved. 193
194
CHAPTER 16
TABLE 16.1
■
Comprehensive Analysis of Gene Function
Genetic effectors*
Effector type
Delivery
Library
Effects
Protein Full-length Truncated
Vector
cDNA fragmented cDNA
Gain Gain/Loss
Fusion
cDNA
Gain/Loss
Protein domains
cDNA
Gain/Loss
Random oligo
Loss (Gain)
Random oligo
Loss (Gain)
Oligo/split intein
Loss (Gain)
cDNA
Loss
Ribozyme
random oligo
Loss
shRNA
fragmented cDNA/oligo
Loss Loss
siRNA
oligo
Peptide Free
Vector, internalization motif
Scaffolded (GFP, protease, scFv, singlechain antibody) Cyclic peptide RNA Antisense
Vector, transfection
* Dominant genetic effectors or “perturbagens” are trans-acting molecules that alter gene function by interfering with a specific gene product the different stages of gene function (e.g., gene expression, protein-product function). Genetic effectors can be derived from protein, peptide or RNA and exert either gain- or loss-of-function effects on gene function. Libraries of genetic effectors are generally based on expressed (mRNA/cDNA), specific or random oligonucleotide sequences. These can be delivered directly to cells by transfection or expressed by gene vectors.
GENE FUNCTION ANALYSIS: AN OVERVIEW The assignment of function to a gene product entails an interruption of the genotype to phenotype flow. This can be accomplished at every stage: DNA mutation, mRNA destruction or inhibition of protein function. Classical genetic approaches involve cis-changes in DNA sequence or structure (insertion, deletions) that are generally technically challenging in mammalian cells (e.g., specific gene knockout). Trans-dominant genetics offers an alternative approach (Gudkov et al., 1994; Xu et al., 2001). Dominant genetic effectors or “perturbagens” are trans-acting molecules that alter gene function by interfering with a specific gene product at different levels (gene expression, protein-product function). Perturbagens exhibit different modes of action and include dominant negative proteins, siRNA, antisense RNA, ribozymes, inhibitory linear or cyclic
peptides, single chain antibodies or chemical inhibitors (Table 16.1) (Lorens et al., 2001). These dominant inhibitory molecules can be taken up by target cells or expressed by various vectors. Each effector type has distinct advantages and optimal uses. Hence, the assignment of gene function invariably enlists a combination of approaches. Recent efforts in gene functional analysis have focused on chemical genetic approaches based on the availability of many new classes of small chemical inhibitors of protein function, and RNAi. RNAi is a sequence-specific post-transcriptional gene-silencing mechanism used by endogenous small regulatory RNAs. Experimentally induced gene silencing has been used extensively for gene functionalization experiments.
RNA INTERFERENCE RNAi is initiated by the presence of double-stranded RNA (dsRNA) in the cytosol (Figure 16.1). The dsRNA is recognized by the RNAse III endonuclease Dicer complex and cleaved into characteristic short, interfering RNAs (siRNA), each 21–23 nucleotides with a core 19 base-paired region, 2 nucleotide 3OH-overhangs and 5-P-ends. Subsequent gene-silencing effects are triggered by siRNA. One of the siRNA strands (guide strand), identified by a less stable 5-duplex region, is preferentially assimilated by an Argonaute (Ago) family protein into the RNA Interference Silencing Complex (RISC). The other (passenger strand) is rejected and destroyed. The guide strand endows the RISC complex with sequence specificity, by hybridizing to cognate mRNA. The RISC:siRNA:mRNA complex localizes to processing bodies (P-bodies), cytoplasmic sites of cellular RNA degradation and storage. The RISC interaction with mRNA via guide strand hybridization leads to one of two outcomes, dictated by the stability of the hybridization and the presence of specific proteins in the RISC complex. Degradation of the mRNA requires a perfect match between the short RNA and the mRNA. Perfect base pairing promotes endonucleolytic cleavage in middle of guide:mRNA duplex (10 nucleotides from the guide strand 5end) by Ago2/RISC complexes. The products of RISC cleavage are degraded by bulk RNA degradation enzymes. In contrast, imperfect hybridization, characteristic of microRNA, promotes translational repression by RISC complexes. The suppression of translation is tolerant of significant mismatches outside of a short “seed” region (6 bp) at the 5 end of the guide RNA. The relative rates of cleavage versus translation inhibition may promote this difference, with more stable complexes kinetically favorable for mRNA degradation. Design of siRNA The design of effective siRNA is based on empirical criteria determined by statistical surveys of siRNA collections and improved understanding of the biochemistry of the RNAi process (Elbashir et al., 2001; Hsieh et al., 2004). A key consideration is that the antisense sequence be utilized as the guide strand. Sequence motifs in the siRNA, mirrors of the cognate mRNA
RNA interference
Dicer
■
195
siRNA
PAZ
Exportin shRNA
PAZ
m7 Target mRNA
Assembly of RISC
Argonaute
Guide strand AAAA
PIWI
Cleavage of passenger strand
Cleavage of mRNA m7
mRNA degradation AAAA
Figure 16.1 The basic approach to RNA interference. RNAi pathways are guided by small RNAs that include siRNA and miRNAs. This figure indicates how retroviral RNAi works through formation of a shRNA exported from the nucleus to the Dicer complex in the cytoplasm. The double-stranded hairpin RNA is cleaved by the Dicer enzyme complex into siRNA. These siRNAs are incorporated into Ago and the RISC. After cleavage of the RNA passenger strand the siRNA guide strand recognizes target sites to direct mRNA cleavage, carried out by the catalytic domain of Ago. RNAi therapeutics developed to harness the siRNA pathway typically involves the delivery of synthetic siRNA into the cell cytoplasm. The miRNA pathway begins with endogenously encoded primary microRNA transcripts (pri-miRNAs), or retrovirally transduced that encode for similar transcripts as indicted in this figure. For more details see de Fougerolles et al., 2007.
secondary structure, influence overall silencing efficiency and may reflect accessibility constraints. Important determinant siRNA features include overall hybridization energy and H-bonding patterns, asymmetric stability at the 5 ends (e.g., AU rich 5 end of guide strand), low duplex stability adjacent to the cleavage site (positions 9–14) and preferred nucleotides at specific positions (e.g., U at position 1 and 17; A at the cleavage site, position 10). RNAi specificity is achieved by limiting siRNA complementarity to non-target transcripts. The rule of thumb is that a given siRNA sequence should have at least three mismatches to non-target transcripts to abrogate silencing, with the central duplex region surrounding the cleavage site (position 9–12) (Birmingham et al., 2006) being particularly sensitive to mismatches. However, recent studies demonstrate that much of the off-target effects are engendered by the seed region, a hexamer stretch at position 2–8 of the guide strand that mediates translational repression by microRNA (miRNA) in 3UTR of mRNA. The presence of more than one seed region sequence in an mRNA 3UTR significantly increases the probability of gene silencing. An in silico analysis of published siRNA sequences concluded that the majority has a significant risk of non-specific gene knock-down (Snove and Holen, 2004). Chemical modification of seed region bases in (e.g., 2-O-methyl
ribosyl substitution at position 2) synthetic siRNA oligonucleotides can reduce this miRNA-like silencing (Jackson et al., 2006). Hence, as off-target silencing is largely sequence specific, RNAi-based results must be validated by at least two independent siRNA sequences derived from the target transcript, phenotypic “rescue” by co-expression of a siRNA-resistant cDNA version of the target gene or by corroborating evidence from other techniques (Cullen, 2006; Echeverri et al., 2006). Delivery of RNAi Effectors Several different methods can be used to deliver the active siRNA into the cell, depending on the organism of interest. In the nematode Caenorhabditis elegans, simply feeding the worm with bacteria that are expressing dsRNA encoding a target gene is sufficient to trigger long-lasting knock-down, as the long dsRNA is converted into active siRNAs intracellularly by Dicer (Fire et al., 1998). In mammalian cells, long dsRNA activates the antiviral interferon pathway leading to non-specific suppression of gene expression (de Veer et al., 2005). siRNAs or short (less than 30 bp) dsRNA precursors of siRNAs can be produced in vitro by chemical synthesis or enzymatic cleavage of longer RNAs and then introduced into cells by transfection. The transfection of
196
CHAPTER 16
■
Comprehensive Analysis of Gene Function
siRNA pools appears to enhance silencing specificity as the concentration of each individual siRNA is low, reducing the magnitude of off-target effects for any particular sequence. Alternatively, the siRNA can be produced within the cell by expression from a vector. These vector-based systems generally use RNA polymerase III promoter cassettes to produce a short primary transcript that folds into a hairpin structure (shRNA) and utilize the endogenous miRNA processing pathway: shRNA is cleaved within the nucleus by the microprocessor complex and exported to the cytoplasm where Dicer converts these into siRNA (Brummelkamp et al., 2002, Chang et al., 2006). Recently, miRNA-based expression scaffolds were developed to allow shRNA expression from more widely used RNA polymerase II promoter systems. These RNAi triggers are designed to be perfectly matched to the target mRNA and can result in potent and sustained transcript degradation. A number of conditional expression systems, mainly based on tetracyclinemediated gene regulation principles, have also been developed, allowing regulated gene silencing. RNAi Loss-of-Function Library Screens Forward genetic screens are conducted in mammalian cells using libraries of dominant genetic effectors introduced into cells to produce a population of individually “mutagenized” cells where
RNAi library
RNAi Classical genetics
Forward (Hypothesis-generating) Phenotype of interest Reverse (Hypothesis-based)
Gene (or protein) of interest
Chemical genetics Library of small molecules
Targeted small molecules
Figure 16.2 Chemical genomics in medicine. Random DNA mutagenesis is a classical genetics approach to create phenotypes that mimic disease in a cell or organism. Random mutagenesis may be replaced by library of interfering RNA, and targeted mutagenesis may be replaced by specific interfering RNA sequences that target the mRNA of interest. In chemical genomics a small molecule compound is used to achieve the same phenotype, or to correct a defined disease-related phenotype. After Kawasumi and Nghiem, 2007.
specific cellular functions are interrupted by the dominantly acting protein, peptide, RNA or organic chemical (Figure 16.2). Individual cells in which the genetic effector has caused a particular phenotypic response can then be selected and the genetic effector responsible identified. By defining the interactions of these effectors, potential targets for the development of therapeutic agents can be identified. Several different groups have produced collections of chemically synthesized siRNAs targeting all predicted human genes. Typically 3–5 different siRNAs are designed for each gene, with the expectation that at least half will knock-down gene expression more than 70%. For use in screens, these siRNAs are transfected into cells, usually as siRNA-lipid complexes, or by pre-spotting the siRNAs onto cell culture surfaces (“reverse transfection”) (Wheeler et al., 2005). These libraries have certain limitations: the approach is limited to transfectable cell types, and the transfected reagents produce only transient RNAi. Alternatively, retroviral vector shRNA libraries can be used. The distinct advantage of this system is the stable integration of viral vectors into host cell chromosomes and subsequent long-term expression of the shRNA. In a recent example, MacKeigan and colleagues conducted synthetic lethal screens to identify kinases and phosphatases that regulate apoptosis and chemoresistance (MacKeigan et al., 2005). HeLa cells were transfected with siRNAs derived from 650 human kinase and 222 phosphatase genes. Remarkably, siRNAs for 73 kinases (11%) and 70 phosphatases (32%) induced greater than twofold increases in apoptosis level relative to controls. These 143 “survival kinases and phosphatases” hence all contribute to HeLa cell survival, emphasizing the potential for novel synthetic interactions in transformed cells. Moffat and colleagues recently reported the development of a large arrayed lentiviral RNAi library (TRC1) and highcontent screen for mitosis (Moffat et al., 2006). A selected subset of the TRC1 library (4903 shRNA constructs) was used to target 1028 human genes (mostly kinases/phosphatases) with single, distinct shRNA-expressing lentiviruses in each well. Automated fluorescence microscopy and image analysis was used to screen for effects on mitosis. Over 100 genes met the screening criteria and were verified to affect mitotic progression using standard microscopy. Brummelkamp and colleagues used a combined chemical genetic and RNAi library screen to define genes required for p53-dependent cell cycle arrest (Brummelkamp et al., 2006). Treatment of breast carcinoma cells with nutlin-3, a small molecule inhibitor of the p53-negative regulator MDM2 (Vassilev et al., 2004), results in upregulation of p53 protein levels and cell cycle arrest. Cells were transduced with a retroviral shRNA library targeting 8000 genes (3 shRNA per gene) and divided into two populations, one that was treated with nutlin-3 and one receiving vehicle. shRNA sequences were isolated by PCR from colonies arising in the respective cell populations and compared by microarray (barcode) analysis. A small number of sequences were specifically enriched in the treated cell population, including the positive control p53 targeting sequence, and
Chemical Genomics
the p53 binding protein, p53BP1, that are required for p53 activation in nutlin-3 treated cells (Brummelkamp et al., 2006). In Vivo Use of siRNA The notion of using siRNAs as small molecule inhibitors of any gene is a particularly exciting prospect for in vivo gene analysis, drug target validation and novel therapeutic modalities. Significant obstacles remain, however, regarding efficient in vivo delivery of highly charged dsRNA. Unmodified siRNAs are rapidly degraded by serum RNAses and do not easily traverse cellular membranes. Chemical modification of specific bases can provide resistance to RNAse but can also compromise base pairing. Hence these modification strategies must balance stability and potency, such as distributed 2-O-methyl modifications. Systemically administered siRNAs are conjugated to targeting groups (e.g., proteins, cholesterol), encapsulated by lipids or carried by nanoparticles to improve cellular uptake. Alternatively various viral vectors can be used to express shRNA in target cells. Indeed, siRNA-based drugs are already in early clinical development. Off-Target Effects of siRNA Unintended gene regulation by siRNA can be induced by the delivery method and activation of innate cellular immunity. Offtarget effects may also be caused by base pair complementarities that alter dozens to hundreds of genes up to fourfold. Off-target effects are present in both in synthetic produced RNA duplexes and in siRNA generated in intact cells by vector-based expression of hairpin RNAs (Kettner-Buhrow et al., 2006). An in silico analysis of published siRNA concluded that the majority of reported siRNA have a risk of non-specific gene knock-down (Snove and Holen, 2004), but suggested that most of these redundancies can be removed in future siRNA design. Birmingham and colleagues have analyzed off-target genes by expression profiling in human cells, and have isolated the off-targeting to the 3 untranslated region (Birmingham et al., 2006). Modification of the siRNA, a 2-O-methyl ribosyl substitution at position 2 in the guide strand, is suggested to decrease the silencing of partially complementary transcripts (Jackson et al., 2006). Successful use of siRNA in the study of gene function should therefore carefully consider both the bioinformatic rules used for target gene complimentary analysis, as well as the chemical synthesis design.
CHEMICAL GENOMICS In recent years the increasing availability of large sets of bioactive small molecule compounds and the ability to systematically interrogate their effects on specific protein targets or cellular functions has engendered the field of chemical genomics (Kawasumi and Nghiem, 2007; Knight and Shokat, 2007). Based on the tenet of chemical genetics, that small molecules can be used to alter gene function and thus mimic mutant alleles, chemical genomics endeavors to discover and characterize small
■
197
molecules that modulate every gene. This is based either on the ability to design de novo a specific inhibitor of a given protein or the ability to survey a large diverse set of compounds for this activity. This latter approach is analogous to creating random mutations in DNA and necessarily requires significant followup to identify the compound target (“mutation”) (Hoyne and Goodnow, 2006; Inamdar, 2001). Hence most chemical genomic screens have focused on a specific target protein. While the generation of mutations in DNA is straightforward, the chemical synthesis of diverse structures is much more difficult. The advent of combinatorial chemistry has significantly improved our ability to generate very large collections of molecules (Dolle et al., 2005; Huwe, 2006). Natural compounds and other “privileged” bioactive scaffolds that affect known proteins or biological functions can form the basis for more targeted libraries (Ortholand and Ganesan, 2004). Hence, a practical description of chemical genomics must include tools that allow scientists to rapidly characterize a large pool of small molecules against target proteins. This implies an inversion of the drug discovery process, where the conventional industry strategy up to now has been to find drugs for targets, instead of finding targets for drugs. Chemical genomics is thus a modern conflation of well establish pharmacology principles, advances in genomics technology and effective chemical synthesis methodologies (Figure 16.2). Chemical Genomic Examples Combinatorial chemistry has made it possible to generate large libraries of small molecules in a short time, while automated high-throughput screening technologies facilitate the rapid interrogation of these compounds. Much of this work, conducted by the pharmacological industry, has been targeted to the discovery of novel chemical entities that modulate important signaling enzymes and cell death regulators (Roland and Dolle, 2004). Synthesis on beads allows transfer of minute amounts of a library of compounds to cell-based screens (MacBeath, 2001). We describe three different examples of chemical biology that illustrate the potential of this approach. FK506 The first example is FK506 (rapamycin, sirolimus), a natural macrolide antibiotic from the bacterium Streptomyces hygroscopicus, discovered as an antifungal agent from natural compound screening in the 1970s. During the 1990s, it was approved as an immunosuppressant for use in kidney transplant. Through study of the FK506 binding protein, a new class of proteins, the immunophilins, were discovered (Schreiber, 1991). The immunophilins appear to be essential in regulation of T-cell functions, predominantly through inhibition of the mTOR (mammalian target of rapamycin) signaling pathway. The result is attenuated T-cell related immunoreactions and avoidance of graft-versushost reactions and rejection of transplanted organs like kidneys. The mTOR pathway is also involved in cell growth of various cancers, and FK506 has been such a successful inhibitor of
198
CHAPTER 16
■
Comprehensive Analysis of Gene Function
proliferation that has lead to clinical cancer trials of this and similar compounds (Sabatini, 2006). This exemplifies the strength of screening natural sources for bioactive compounds, and illustrates how these compounds may result in discovery of new protein classes and cellular pathways. Alzheimer’s Disease New diagnostic imaging modalities, in particular the positron emission tomography combined with computer tomography, allow functional imaging of tissues or organs based on highly specific molecular probes (see Chapter 44). Alzheimer’s disease is a devastating neurological degenerative disease characterized by progressive loss of cognitive function. Only clinical examination together with imprecise imaging examinations have been used to determine the diagnosis, while post-mortem histopathological examination has revealed its disease-related protein structural features with beta-amyloid senile plaques and neurofibrillary tangles. These pathological protein structures are related to neural cell death, but therapy development is seriously hampered by lack of precise in vivo disease monitoring in humans. Small molecules have been developed that bind to senile plaques and neurofibrillary tangles (Kepe et al., 2006). The development of these molecular imaging probes has been relatively slow, based on decades of empirical use of dyes binding to these structures. More extensive use of chemical libraries, in combination with animal models, may lead to even more precise diagnostics. Bcl-2 Binding Compounds The Bcl-2 (B-cell lymphoma protein) was discovered in nonHodgkin lymphoma, where a recurrent chromosomal translocation had placed the Bcl-2 gene under control of the strong immunoglobulin heavy chain promoter (Adams and Cory, 2001). The actual mechanism of action of Bcl-2 first was elucidated when it was realized that Bcl-2 is homologous to the CED-4 gene in the worm Caenorhabditis elegans, a gene that is central in the worm’s genetic programs for cell death regulation. The function of Bcl-2 was explored by worm and mouse genetics, and the structure of this novel class of proteins mapped with high resolution. Through structure-guided design, small molecules have been designed that bind to a deep hydrophobic binding pocket in Bcl-2 (Bruncko et al., 2007). One such compound was developed based on a high-throughput NMR-screen of a chemical library in search for a lead compound that bound to the hydrophobic BH3-binding groove of Bcl-XL. Further structure analysis by NMR was used to avoid binding to serum albumin, as well as enhance the binding to Bcl-2, Bcl-XL and Bcl-w. One of the final drug compounds, ABT-737, bound to Bcl-2 in the sub-nanomolar range, illustrating the power of NMR-based high-throughput screening with parallel synthesis and structure-based design. Preclinical studies are promising for this new class of drugs that block the anticell death action of Bcl-2 (Fesik, 2005).
TABLE 16.2
Phenotypic screens*
Cellular function
Target phenotype
Read-out
Cell death
Survival
Growth/colony formation
Proliferation
Cell cycle arrest
Various dyes (e.g., tetrazolium salt) DNA staining Cell tracker fluorescence (flow cytometry)
Protein localization
Mislocalization
GFP-fusion Immunofluorescence
Protein modification
Proteasome activity Kinase activity
GFP-PEST
Promoter activation
Reporter genes (luciferase)
Transcription
Phospho-specific antibodies (microscopy, flow cytometry)
Flow cytometry surface marker *Phenotypic screens are designed to address a specific cellular function. General cellular functions (cell proliferation, cell death) involve multiple gene products, while more specific functions (protein localization, enzyme activity) involve fewer genes. Phenotypic screens can be focused on specific signaling events using gene expression reporters. Many detection methods can be used including enzymatic assays, microscopy and flow cytometry.
Phenotypic Screening Phenotypic screening can take advantage of forward and reverse genetics to identify a tentative disease target. Forward genetics is based on random modulation of the phenotype and then identification of the gene. Reverse genetics is based on a specific gene modification and then examination for a phenotype (Figure 16.2). An in vitro phenotype screen can comprise cells that are treated with a library of small molecules or siRNA and examined for activation or inhibition of the process of interest. High-throughput screening of defined compound libraries and cell responses that is detected by automated microscopy is a powerful approach when combined with a carefully designed strategy for verification. Such assays can be used to screen for compounds that alter simple intracellular translocations of labeled proteins or more complex cellular mechanisms like phagocytosis, phagosome maturation and bacterial invasion into human cells (Steinberg et al., 2007). Importantly, cell-based phenotypic screens present the target proteins in their native configuration, often a confounding challenge for screens based on purified proteins (Table 16.2). Combination of drug target gene expression signatures and high-throughput gene expression analysis has been used to determine new compounds that modulate cancer phenotypes,
Gene Function Studies
here exemplified by inhibition of androgen-receptor signaling (Hieronymus et al., 2006). The cellular mRNA from a defined set of genes were bar coded and amplified through PCR, and quantified by two color flow cytometer after hybridization to uniquely colored polystyrene bead. This gene expression analysis classified the two androgen-receptor signaling inhibiting compounds, celastrol and gedunin, as HSP90-inhibitors (Hieronymus et al., 2006). Most high-throughput screening approaches utilize relatively straightforward readouts. In contrast, high-content screening seeks to capture multiple parameters simultaneously. This “high resolution” examination of cell responses can reveal unique characteristics of drug action, though at the expense of throughput. Hence, smaller focused libraries are generally used. In a search for novel cancer cell apoptosis-inducing agents from cyanobacteria, Herfindal and colleagues carefully compared nuclear morphologies in treated normal human hepatocytes (Herfindal et al., 2005). The result was identification of specific hepatototoxic strains of algae not involving the known hepatotoxins microcystins and nodularin. A similar approach using primary blood platelets and neural derived cell lines has identified specific neural and platelet targeting toxins (Selheim et al., 2005). Structure elucidation of the active natural compound is necessary to fully understand the mechanisms involved (Antonopoulou et al., 2002) and provide a basis for future drug development. Parallel Chemical and Genome-wide RNAi Screens Parallel screening for siRNA and small molecule inhibitors is a particularly fruitful combination to map constituents of a cellular function (Figure 16.2). Further siRNA hits provide an effective starting point to reveal the targets of small molecule hits. In a study focused on cytokinesis, the final step of cell division, by searching for siRNAs and small molecules that generate binuclear cells (Eggert et al., 2004). Comparative analysis of phenotypes revealed distinct categories of drug action, providing a basis for further target validation.
GENE FUNCTION STUDIES Therapeutic development invariably involves experiments in animal models. Gene function analysis in the intact organism is a critical component of validation. Genetically tractable organisms like fly, worm and zebrafish allow the researcher to study gene function within short periods of time. Mice have been used for decades, more recently in large institutional set ups, for example to dissect the genetic basis for human neurological disorders (Oliver and Davies, 2005). But due to obvious differences, many gene functions of importance in animals may be irrelevant or absent in humans and vice versa. This is exemplified by cancer genetics where the phenotype of certain oncogenes in mouse does not mimic the disease spectra in humans with a similar genetic defect (Rangarajan and Weinberg, 2003).
■
199
Clinical Trials and Epidemiology Experimental early phase human clinical trials allow a chemical genetic elucidation of gene function. Many patients are treated with one specific drug during the disease course. These compounds have defined target and off-target effects, and gene function can be studied in sentinel cells or tissues, like peripheral leukocytes, or the diseased tissues. There is increasing use of gene expression studies in patients who undergo therapy. One example is examination of gene expression analysis within hours after intensive chemotherapy against acute myeloid leukemia (Anensen et al., 2006). In this study peripheral blood cells, in these samples representing more than 95% pure cancer cells, were harvested from 2 h to 6 h after start of chemotherapy. As early as 2 h after start of chemotherapy, the p53 protein was activated, and the 100 most expressed genes were dominated by known p53 target genes. The tumor suppressor p53 is mutated in more than 40% of all cancers, but in leukemia the gene mutation frequency is below 10%. However, mutated p53 in AML is associated with highly chemoresistant disease. Together, this indicates that normal p53 activation is of paramount importance in the first intensive chemotherapy in acute myeloid leukemia, and that other therapeutic strategies should probably be searched for in cases were p53 is mutated. In some countries or regions there are extensive health registries, including general health status and careful registration of diseases like cancer, cardiovascular disease and multiple sclerosis. These comprehensive registries take advantage of unique personal identifiers and limited population mobility. As a result of governmental financed insurance plans, corresponding detailed information of prescribed drugs is available in populationbased registries (Andersen, 2006). Impacting most fields of modern medicine, epidemiological studies have been performed by linking different registries. Increased risk for miscarriage in pregnant users of non-steroidal anti-inflammatory drugs have been identified by combining data from the prescription registry, birth registry and hospital discharge registry (Nielsen et al., 2001). Protective association of non-steroidal anti-inflammatory drugs against risk for colon, rectal, stomach and ovarian cancer has been described (Sørensen et al., 2003). Finally, registries were used to identify increased mortality of myocardial infarction in patients on cyclooxygenase-2 inhibitors (Gislason et al., 2006). The molecular cause of these associations will in the future be addressed by linkage of genetics and proteomics data from population-based biobanks with the health registries (Pukkala et al., 2007; Stoltenberg, 2005). Combination of drug prescription registries, health registries and biobanks could facilitate retrospective genomic screens for function of drugs and drug combination. A European registry and biobank with more than 600,000 twin pairs and a corresponding biobank may be such a unique source for retrospective chemical genomics in humans (Muilu et al., 2007). Genomics combined with epidemiological data may be an important future tool to determine relationships between genes and drugs that are inaccessible in animal models or require large number of individuals
200
CHAPTER 16
Comprehensive Analysis of Gene Function
■
(a)
Diseases
Genes
Drugs
Genome-wide mRNA-expression profiles N
N
(b)
N
N O P HO OH
Na P OH
O
OH
Unannotated chemical
cmap hypothesis
Disease state
cmap hypothesis
Experimental verification
N3
O
O
O
O
N
N
N
N
N
N
O O
OH
New inhibitors
(c)
Experimental verification
Clinical evaluation
Figure 16.3 A universal functional bioassay forms the Connectivity Map tool. (a) Based on the assumption that all transcripts are known, and simultaneous measurement of these transcript are possible through standardized gene expression array data collection, it may be possible to portrait all induced and organic biological conditions in a common analytic space. Similarities can thereby be determined between these induced and biological conditions, and allow the foundation for Connectivity Maps. (b) An uncharacterized molecule is identified as an active compound in a particular cellular pathway through the Connectivity Map. (c) A particular disease state is coupled with perturbations in certain molecular pathways, and suggests a small molecule as experimental therapeutic. Both (b) and (c) illustrates that the Connectivity Map is a tool for the experimental researcher, where in vitro and in vivo experimental verification of the proposed in silico molecular connection is needed. Modified from Lamb (2007).
for detection. Hypotheses generated in genetic epidemiology may then be tested experimentally by chemical genomics.
CONCLUSIONS New developments in the understanding of RNAi will allow design of siRNA molecules with more potent and specific mRNA targeting characteristics for all genes. The growing access to small
compound collections, liquid handling robotics, miniaturization of assay formats and streamlined readout detection methods will undoubtedly lead to both less time-consuming, labor-intensive and expensive siRNA and chemical genomic screening. As recently proposed, connectivity mapping based on geneexpression signatures may be able to successfully connect disease-related pathways and bioactive small molecules (Lamb et al., 2006). Lamb and colleagues propose a large-scale Connectivity Map project similar to the Human Genome Project, where an
References
open resource can be used for investigators to find connections between small molecule action, physiological processes and disease (Figure 16.3). Several proof-of-principle reports have emerged, for example, identifying novel compounds for established pathways or suggest known drugs and pathways to resolve known drugresistance causing mechanism (Hieronymus et al., 2006; Wei et al., 2006). Connectivity Map technology, even with its limitations (Lamb et al., 2006, for review see Lamb, 2007), may be a useful tool for the experimentalist to focus his or her interest. We have entered the post-genome sequence era. But without having the key to the cipher of gene function, the outstanding effort in genome sequencing will remain just an impressive
■
201
code, without helping us to resolve the enigmas of our most common diseases and making genomic medicine a reality for the patients.
ACKNOWLEDGEMENTS This work was supported by the Norwegian Research Council Functional Genomics Program (FUGE) grant numbers 151859 and 183675. We thank Lars Helgeland Emmet McCormack, David Micklem, Maja Mujic and Line Wergeland for fruitful discussions and for providing illustrations for Figure 16.3.
REFERENCES Adams, J.M., Cory, S. (2001). Life-or-death decisions by the Bcl-2 protein family. Trends Biochem Sci 26, 61–66. Andersen, M. (2006). Is it possible to measure prescribing quality using only prescription data?. Basic Clin Pharmacol Toxicol 1, 314–319. Anensen, N., Oyan, A.M., Bourdon, J.C., Kalland, K.H., Bruserud, O. and Gjertsen, B.T. (2006). A distinct p53 protein isoform signature reflects the onset of induction chemotherapy for acute myeloid leukemia. Clin Cancer Res 12, 3985–3992. Antonopoulou, S., Oikonomou, A., Karantonis, H.C., Fragopoulou, E. and Pantazidou, A. (2002). Isolation and structural elucidation of biologically active phospholipids from Scytonema julianum (cyanobacteria). Biochem J 367, 287–293. Bier, E. (2005). Drosophila, the golden bug, emerges as a tool for human genetics. Nat Rev Genet 6, 9–23. Birmingham, A., Anderson, E.M., Reynolds, A., Ilsley-Tyree, D., Leake, D., Fedorov, Y., Baskerville, S., Maksimova, E., Robinson, K., Karpilow, J. et al. (2006). 3 UTR seed matches, but not overall identity, are associated with RNAi off-targets. Nat Method 3, 199–204. Bredel, M. and Jacoby, E. (2004). Chemogenomics: An emerging strategy for rapid target and drug discovery. Nat Rev Genet 5(4), 262–275. Brummelkamp, T.R., Bernards, R. and Agami, R. (2002). A system for stable expression of short interfering RNAs in mammalian cells. Science 296, 550–553. Brummelkamp, T.R., Fabius, A.W., Mullenders, J., Madiredjo, M., Velds, A., Kerkhoven, R.M., Bernards, R. and Beijersbergen, R.L. (2006). An shRNA barcode screen provides insight into cancer cell vulnerability to MDM2 inhibitors. Nat Chem Biol 2, 202–206. Bruncko, M., Oost, T.K., Belli, B.A., Ding, H., Joseph, M.K., Kunzer, A., Martineau, D., McClellan, W.J., Mitten, M., Ng, S.C. et al. (2007). Studies leading to potent, dual inhibitors of Bcl-2 and Bcl-xL. J Med Chem 50, 641–662. Chang, K., Elledge, S.J. and Hannon, G.J. (2006). Lessons from Nature: microRNA-based shRNA libraries. Nat Method 3, 707–714. Cullen, B.R. (2006). Enhancing and confirming the specificity of RNAi experiments. Nat Method 3, 677–681. de Fougerolles, A., Vornlocher, H.P., Maraganore, J. and Lieberman, J. (2007). Interfering with disease: a progress report on siRNA-based therapeutics. Nat Rev Drug Discov 6, 443–453.
Dolle, R.E., Le Bourdonnec, B., Morales, G.A., Moriarty, K.J. and Salvino, J.M. (2005). Comprehensive survey of combinatorial library synthesis: 2005. J Comb Chem 8(5), 597–635. Echeverri, C.J., Beachy, P.A., Baum, B., Boutros, M., Buchholz, F., Chanda, S.K., Downward, J., Ellenberg, J., Fraser, A.G., Hacohen, N. et al. (2006). Minimizing the risk of reporting false positives in large-scale RNAi screens. Nat Method 3, 777–779. Eggert, U.S., Kiger, A.A., Richter, C., Perlman, Z.E., Perrimon, N., Mitchison, T.J. and Field, C.M. (2004). Parallel chemical genetic and genome-wide RNAi screens identify cytokinesis inhibitors and targets. PLoS Biol 2, 2135–2143. Elbashir, S.M., Harborth, J., Lendeckel, W., Yalcin, A., Weber, K. and Tuschl, T. (2001). Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian cells. Nature 411, 494–498. Fesik, S.W. (2005). Promoting apoptosis as a strategy for cancer drug discovery. Nat Rev Cancer 5, 876–885. Fire, A., Xu, S., Montgomery, M.K., Kostas, S.A., Driver, S.E. and Mello, C.C. (1998). Potent and specific genetic interference by double-stranded RNA in Caenorhabditis elegans. Nature 391, 806–811. Fuchs, F. and Boutros, M. (2006). Cellular phenotyping by RNAi. Brief Funct Genomic Proteomic 5, 52–56. Gislason, G.H., Jacobsen, S., Rasmussen, J.N., Rasmussen, S., Buch, P., Friberg, J., Schramm, T.K., Abildstrom, S.Z., Køber, L., Madsen, M. et al. (2006). Risk of death or reinfarction associated with the use of selective cyclooxygenase-2 inhibitors and nonselective nonsteroidal antiinflammatory drugs after acute myocardial infarction. Circulation 113(25), 2906–2913. Gudkov,A.V., Kazarov,A.R.,Thimmapaya, R.,Axenovich, S.A., Mazo, I.A. and Roninson, I.B. (1994). Cloning mammalian genes by expression selection of genetic suppressor elements: Association of kinesin with drug resistance and cell immortalization. Proc Natl Acad Sci USA 91, 3744–3748. Herfindal, L., Oftedal, L., Selheim, F., Wahlsten, M., Sivonen, K. and Doskeland, S.O. (2005). A high proportion of Baltic Sea benthic cyanobacterial isolates contain apoptogens able to induce rapid death of isolated rat hepatocytes. Toxicon 46, 252–260. Hieronymus, H., Lamb, J., Ross, K.N., Peng, X.P., Clement, C., Rodina,A., Nieto, M., Du, J., Stegmaier, K., Raj, S.M. et al. (2006). Gene expression signature-based chemical genomic prediction identifies a novel class of HSP90 pathway modulators. Cancer Cell 10, 321–330.
202
CHAPTER 16
■
Comprehensive Analysis of Gene Function
Hsieh, A.C., Bo, R., Manola, J., Vazquez, F., Bare, O., Khvorova, A., Scaringe, S. and Sellers, W.R. (2004). A library of siRNA duplexes targeting the phosphoinositide 3-kinase pathway: Determinants of gene silencing for use in cell-based screens. Nucleic Acids Res 32, 893–901. Huwe, C.M. (2006). Synthetic library design. Drug Discov Today 11(15-16), 763–767. Hoyne, G.F. and Goodnow, C.C. (2006). The use of genomewide ENU mutagenesis screens to unravel complex mammalian traits: Identifying genes that regulate organ-specific and systemic autoimmunity. Immunol Rev 210, 27–39. Hunter, T. (2007). Treatment for chronic myelogenous leukemia: The long road to imatinib. J Clin Invest 117, 2036–2043. Inamdar, M.S. (2001). Functional genomics the old-fashioned way: Chemical mutagenesis in mice. Bioessays 23, 116–120. Jackson, A.L., Burchard, J., Schelter, J., Chau, B.N., Cleary, M., Lim, L. and Linsley, P.S. (2006). Widespread siRNA “off-target” transcript silencing mediated by seed region sequence complementarity. RNA 12, 1179–1187. Kaelin, W.G., Jr (2005). The concept of synthetic lethality in the context of anticancer therapy. Nat Rev Cancer 5, 689–698. Kawasumi, M. and Nghiem, P. (2007). Chemical genetics: Elucidating biological systems with small-molecule compounds. J Invest Dermatol 127(7), 1577–1584. Kepe, V., Huang, S.C., Small, G.W., Satyamurthy, N. and Barrio, J.R. (2006). Visualizing pathology deposits in the living brain of patients with Alzheimer’s disease. Methods Enzymol 412, 144–160. Kettner-Buhrow, D., Dittrich-Breiholz, O., Schneider, H., Wolter, S., Resch, K. and Kracht, M. (2006). Small interfering RNAs generated by recombinant dicer induce inflammatory gene expression independent from the TAK1-NFkappaB-MAPK signaling pathways. Biochem Biophys Res Commun 347, 566–573. Knight, Z.A. and Shokat, K.M. (2007). Chemical genetics: Where genetics and pharmacology meet. Cell 128(3), 425–430. Lamb, J., Crawford, E.D., Peck, D., Modell, J.W., Blat, I.C., Wrobel, M.J., Lerner, J., Brunet, J.P., Subramanian, A., Ross, K.N. et al. (2006). The Connectivity Map: Using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929–1935. Lamb, J. (2007). The connectivity map: a new tool for biomedical research. Nat Rev Cancer 7, 54–60. Landro, J.A., Taylor, I.C., Stirtan, W.G., Osterman, D.G., Kristie, J., Hunnicutt, E.J., Rae, P.M. and Sweetnam, P.M. (2000). HTS in the new millennium: The role of pharmacology and flexibility. J Pharmacol Toxicol Method 44(1), 273–289. Lorens, J.B., Sousa, C., Bennett, M.K., Molineaux, S.M. and Payan, D.G. (2001). The use of retroviruses as pharmaceutical tools for target discovery and validation in the field of functional genomics. Curr Opin Biotechnol 12, 613–621. MacBeath, G. (2001). Chemical genomics: What will it take and who gets to play? Genome Biol, 2, comment2005.1–comment2005.6. MacKeigan, J.P., Murphy, L.O. and Blenis, J. (2005). Sensitized RNAi screen of human kinases and phosphatases identifies new regulators of apoptosis and chemoresistance. Nat Cell Biol 7, 591–600. Mello, C.C. and Conte, D. (2004). Revealing the world of RNA interference. Nature 431, 338–342. Moffat, J., Grueneberg, D.A.,Yang, X., Kim, S.Y., Kloepfer, A.M., Hinkle, G., Piqani, B., Eisenhaure, T.M., Luo, B., Grenier, J.K. et al. (2006).
A lentiviral RNAi library for human and mouse genes applied to an arrayed viral high-content screen. Cell 124, 1283–1298. Muilu, J., Peltonen, L. and Litton, J.E. (2007). The federated database – a basis for biobank-based post-genome studies, integrating phenome and genome data from 600,000 twin pairs in Europe. Eur J Hum Genet 15(7), 718–723. Nielsen, G.L., Sørensen, H.T., Larsen, H. and Pedersen, L. (2001). Risk of adverse birth outcome and miscarriage in pregnant users of non-steroidal anti-inflammatory drugs: Population based observational study and case-control study. BMJ 322(7281), 266–270. Oliver, P.L., Davies, K.E. (2005). Analysis of human neurological disorders using mutagenesis in the mouse. Clin Sci 108, 385–397. London. Ortholand, J.Y. and Ganesan, A. (2004). Natural products and combinatorial chemistry: Back to the future. Curr Opin Chem Biol 8(3), 271–280. Peters, J.L., Cnudde, F. and Gerats, T. (2003). Forward genetics and mapbased cloning approaches. Trends Plant Sci 8(10), 484–491. Pukkala, E., Andersen, A., Berglund, G., Gislefoss, R., Gudnason, V., Hallmans, G., Jellum, E., Jousilahti, P., Knekt, P., Koskela, P. et al. (2007). Nordic biological specimen banks as basis for studies of cancer causes and control – more than 2 million sample donors, 25 million person years and 100,000 prospective cancers. Acta Oncol 46(3), 286–307. Rangarajan, A. and Weinberg, R.A. (2003). Opinion: Comparative biology of mouse versus human cells: Modelling human cancer in mice. Nat Rev Cancer 3, 952–959. Rognan, D. (2007). Chemogenomic approaches to rational drug design. Br J Pharmacol 152, 38–52. Roland, E. and Dolle, J. (2004). Comprehensive Survey of Combinatorial Library Synthesis: 2003. Comb Chem 6, 623–679. Sabatini, D.M. (2006). mTOR and cancer: insights into a complex relationship. Nat Rev Cancer 6, 729–734. Schreiber, S.L. (1991). Chemistry and biology of the immunophilins and their immunosuppressive ligands. Science 251, 283–287. Selheim, F., Herfindal, L., Martins, R., Vasconcelos, V. and Doskeland, S. O. (2005). Neuro-apoptogenic and blood platelet targeting toxins in benthic marine cyanobacteria from the Portuguese coast. Aquat Toxicol 74, 294–306. Sherbenou, D.W. and Druker, B.J. (2007). Applying the discovery of the Philadelphia chromosome. J Clin Invest 117, 2067–2074. Snove, O, Jr and Holen, T. (2004). Many commonly used siRNAs risk off-target activity. Biochem Biophys Res Commun 319, 256–263. Steinberg, B.E., Scott, C.C. and Grinstein, S. (2007). High-throughput assays of phagocytosis, phagosome maturation, and bacterial invasion. Am J Physiol Cell Physiol 292, C945–C952. Stoltenberg, C. (2005). Merging genetics and epidemiology: what is in it for public health?. Scand J Public Health 33, 1–3. Sørensen, H.T., Friis, S., Nørgård, B., Mellemkjaer, L., Blot, W.J., McLaughlin, J.K., Ekbom, A. and Baron, J.A. (2003). Risk of cancer in a large cohort of nonaspirin NSAID users: A populationbased study. Br J Cancer 88(11), 1687–1692. Vassilev, L.T., Vu, B.T., Graves, B., Carvajal, D., Podlaski, F., Filipovic, Z., Kong, N., Kammlott, U., Lukacs, C., and Klein, C., et al. (2004). In vivo activation of the p53 pathway by small-molecule antagonists of MDM2. Science 303, 844–848. de Veer, M.J., Sledz, C.A. and Williams, B.R. (2005). Detection of foreign RNA: Implications for RNAi. Immunol Cell Biol 83, 224–228. Westbrook, T.F., Stegmeier, F. and Elledge, S.J. (2005). Dissecting cancer pathways and vulnerabilities with RNAi. Cold Spring Harb Symp Quant Biol 70, 435–444.
Recommended Resources
Wheeler, D.B., Carpenter, A.E. and Sabatini, D.M. (2005). Cell microarrays and RNA interference chip away at gene function. Nat Genet 37(Suppl), S25–S30. Wei, G., Twomey, D., Lamb, J., Schlis, K., Agarwal, J., Stam, R.W., Opferman, J.T., Sallan, S.E., den Boer, M.L., Pieters, R. et al. (2006). Gene expression-based chemical genomics identifies
203
rapamycin as a modulator of MCL1 and glucocorticoid resistance. Cancer Cell 10, 331–342. Xu, X., Leo, C., Jang, Y., Chan, E., Padilla, D., Huang, B.C., Lin, T., Gururaja, T., Hitoshi, Y., Lorens, J.B. et al. (2001). Dominant effector genetics in mammalian cells. Nat Genet 27, 23–29.
RECOMMENDED RESOURCES http://www.ncgc.nih.gov/ http://www.broad.harvard.edu/chembio/icg/overview/index.htm http://pubchem.ncbi.nlm.nih.gov/ http://nmf.jax.org/index.html http://www.gsc.riken.go.jp
■
http://www.ncbi.nlm.nih.gov/sites/entrez?dbomim/ http://www.genomeutwin.org/ http://www.ukbiobank.ac.uk/ Database of Genotypes and Phenotypes (dbGaP): (http://view.ncbi.nlm.nih.gov/sites/entrez?dbgap/
This page intentionally left blank
Section
Informatic and Computational Platforms for Genomic Medicine
17. 18. 19. 20. 21.
3
Bioinformatic and Computational Analysis for Genomic Medicine Fundamentals and History of Informatics for Genomic and Personalized Medicine Electronic Medical Records in Genomic Medicine Practice and Research Clinical Decision Support in Genomic and Personalized Medicine Online Health Information Retrieval by Consumers and the Challenge of Personal Genomics
CHAPTER
17 Bioinformatic and Computational Analysis for Genomic Medicine Atul J. Butte
INTRODUCTION The past 10 years have seen development and deployment of a variety of measurement tools in molecular science that enable the large-scale parallel quantitative assessment of molecular state. The premier example of such measurement tools is the RNA expression detection microarray, which provides quantitative measurements of expression of over 40,000 unique RNAs within cells (Chee et al., 1996; DeRisi et al., 1996). Yet, as illustrated in previous chapters in this volume, microarrays represent just one of at least 30 or so measurement or experimental modalities available to investigators in molecular biology and genomics, including metabolite quantification (Edwards et al., 2001), DNA polymorphism measurements (Johnson et al., 2001), protein quantification (Espina et al., 2003; Ghaemmaghami et al., 2003; Gygi et al., 1999; Liotta et al., 2003), protein activity (Hestvik et al., 2003), protein interactions with small molecules (Jessani et al., 2002) and DNA (Lee et al., 2002; Odom et al., 2004; Ren et al., 2000;Wang et al., 2001; Weinmann et al., 2002), and many others. Though these molecular technologies are commonly described as being “high-throughput” or “genome-wide” in nature, we will use the term high-bandwidth to describe them in this chapter. An analogy explains our reasoning: high-bandwidth is to a supermarket, which serves a large variety of different foods irrespective of speed, while high-throughput is to a fast-food restaurant, serving a few types of
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 206
foods quickly. Genome-wide technologies are biased towards genes, while we are covering measurements of many other types besides genes. The service role for bioinformatics in research using these new tools is undeniable. Each of these high-bandwidth modalities yields a sizeable amount of raw data, so distilling these raw data and filtering out measurement noise through the proper use of bioinformatics methods is crucial. Bioinformatics clearly plays a role in the storage, retrieval, and sharing of these measurements from local and international repositories and in relating these measurements to clinical outcomes. However, what we will argue in this chapter is that the role for bioinformatics in genomic and personalized medicine is well beyond that of providing a service and lies more broadly in enabling new and interesting questions to be asked in biomedical, translational, and clinical research. In this chapter, we will present several examples that illustrate how the use of specific bioinformatics methods has enabled a revolution of discovery in medical diagnostics, prognostics, therapeutics, anatomy, and nosology. We will then review a few specific analytic methods and standardized vocabularies useful in studying and identifying high-bandwidth measurements for genomic and personalized medicine. After covering a few freely available software tools valuable for conducting analyzes, we will end with some thoughts about where the future discoveries in genomic medicine, enabled by bioinformatics, may come from.
Copyright © 2009, Elsevier Inc. All rights reserved.
Vignettes: How Specific Bioinformatics Methods can Change the Practice of Medicine
VIGNETTES: HOW SPECIFIC BIOINFORMATICS METHODS CAN CHANGE THE PRACTICE OF MEDICINE While many high-bandwidth measurement technologies have been used to enable basic discoveries in the life sciences, certainly the most intriguing and provocative use of these measurement modalities has been in enabling novel types of medical discoveries. Here, we will present case examples of how these discoveries were enabled by bioinformatics techniques and have demonstrably altered our abilities in four distinct areas of medicine: diagnosis, therapeutics, histopathology, and nosology (Table 17.1). For these vignettes, we will particularly focus on the use of microarrays, as this is currently the most commonly used highbandwidth measurement modality. Diagnosis Diagnosis is a process in which a physician, when presented with a patient holding a chief complaint, gathers information through questioning, history taking, physical examination, and testing, compares the results of those queries with patterns established for known diseases, and assigns the disease with the highest likelihood. High-bandwidth testing in molecular biology has impacted every step of the diagnostic process. For example, those being evaluated for a disease may not yet even have a complaint, as DNA
TABLE 17.1
■
207
polymorphisms are discovered and used to predict future disease (Kohane et al., 2006). Medical history taking becomes problematic when considering the genetic context of a patient, which is increasingly known; for example, many infants now have one or more genes sequenced at birth as part of state-mandated newborn screening (Khoury et al., 2003). History taking has potential to become even more convoluted as more adults have genes (and even whole genomes) sequenced. The bioinformatics challenge in storing, providing, and teaching patients and parents how to use this type of medical information, while protecting against others finding this information, has only just started to be addressed (Adida and Kohane, 2006; 2004; Centers for Disease Control and Prevention, 2004; Henneman et al., 2006; Roche and Annas, 2001, 2006). The nature of the physical examination itself has changed. Positive and negative physical findings (e.g., acanthosis nigricans, or darkening of the skin of the neck) are traditionally considered as probes into the process of disease (e.g., type 2 diabetes mellitus). With the democratization of genomics throughout all fields of medicine, physical findings can now be viewed collectively as phenotype and as a probe into the inner workings of the genome (Freimer and Sabatti, 2003). Two different diseases with similar sets of physical findings may indeed appear quite similar when viewed from the perspective of genes and gene products (Brunner and van Driel, 2004). But certainly the greatest impact on diagnosis from highbandwidth molecular testing has been felt on our ability to redefine our patterns of disease using these new measurements. One
Bioinformatics techniques have demonstrably altered our abilities in four distinct areas of medicine
Area of the practice of medicine Diagnosis Chief complaint Medical history Medical record keeping Physical examination Differential diagnosis Disease sub-type Therapeutics Finding novel effects for existing drugs Side effects Finding novel therapies for untreatable conditions Histopathology Metastasis of unknown origin Pathological mechanism of action of disease Nosology Finding differences between diseases Finding similarities between diseases Linking cellular models to human diseases
Example of bioinformatics altering practice Predicting future disease from polymorphisms before complaint Genetic screening at birth should become part of a patient’s life-long electronic medical record Patient-owned Internet-based genomic records Findings collectively viewed as phenotype and matched against patterns Molecular measurements can be compared against case-examples Hierarchical clustering to determine novel subtypes from molecular measurements Linking drug effects to molecular measurements Predicting drug adverse effect from pre-clinical molecular measurements Matching molecular profiles of disease with molecular profiles of drug effect Matching molecular measurements from metastasis with profiles of known primary cancers Identifying differentially expressed genes and linking back to known biological pathways Principal components of variance of disease sample microarrays Common signatures across microarray measurements of disease Comparing cellular and disease microarray measurements
208
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
of the earliest examples of this impact was in the work of Golub and colleagues published in 1999 (Golub et al., 1999). Acute lymphomas were first divided into acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML) in the 1960s based on histochemical testing, and more recently by antibody-based testing against cell-surface molecules, but there was no single established test to make this diagnosis. Golub was the first to show that gene expression microarrays could be used to measure gene
expression levels from samples from acute lymphomas, and using a bioinformatics technique known as supervised machine learning, trained a prediction system to find and use those genes that uniquely distinguished ALL from AML (Figure 17.1). He then used this predictor to accurately classify additional new leukemia cases. Many others have used alternative computational and bioinformatics techniques to build predictors since this landmark study, while many additional diagnostic dilemmas have similarly
ALL
AML C-myb (U22376) Proteasome iota (X59417) MB-1 (U05259) Cyclin D3 (M92287) Myosin light chain (M31211) RbAp48 (X74262) SNF2 (D26156) HkrT-1 (S50223) E2A (M31523) Inducible protein (L47738) Dynein light chain (U32944) Topoisomerase II β (Z15115) IRF2 (X15949) TFIIEβ (X63469) Acyl-Coenzyme A dehydrogenase (M91432) SNF2 (U29175) (Ca2)-ATPase (Z69881) SRP9 (U20998) MCM3 (D38073) Deoxyhypusine synthase (U26266) Op 18 (M31303) Rabaptin-5 (Y08612) Heterochromatin protein p25 (U35451) IL-7 receptor (M29696) Adenosine deaminase (M13792) Fumarylacetoacetate (M55150) Zyxin (X95735) LTC4 synthase (U50136) LYN (M16038) HoxA9 (U82759) CD33 (M23197) Adipsin (M84526) Leptin receptor (Y12670) Cystatin C (M27891) Proteoglycan 1 (X17042) IL-8 precursor (Y00787) Azurocidin (M96326) p62 (U46751) CyP3 (M80254) MCL1 (L08246) ATPase (M62762) IL-8 (M28130) Cathepsin D (M63138) Lectin (M57710) MAD-3 (M69043) CD11c (M81695) Ebp72 (X85116) Lysozyme (M19045) Properdin (M83652) Catalase (X04085)
3
2.5
low
2
1.5
1
0.5
0
0.5
Normalized Expression
1
1.5
2
2.5
3 high
Figure 17.1 Todd Golub and colleagues measured genome-wide expression levels in samples from patients with ALL and AML. The top 50 genes distinguishing ALL and AML are displayed, with genes elevated in expression shown in pink, and genes decreased in expression shown in blue. From Golub et al., 1999 Science 286:531. Reprinted by permission from AAAS.
Vignettes: How Specific Bioinformatics Methods can Change tjhe Practice of Medicine
Therapeutics Bioinformatics, coupled with the exponential growth in data collected using high-bandwidth measurement modalities, has the potential to revolutionize the way patients are treated with pharmaceuticals. Much has already been written on how highbandwidth measurements have generally altered the process of drug discovery (Debouck and Goodfellow, 1999). Here, we will cover specific case examples on how the conceptualization of pharmacology can be radically changed through development and use of biomedical informatics methods. In 2000, Butte and colleagues focused on generating hypotheses of functional relationships between pairs of genes and pharmaceuticals, with the hope that specific genes could be found that altered a cancer’s ability to respond to specific drugs. To do this, they started with two databases, both related to the NCI60, a standardized set of 60 human cancer cell lines used by the National Cancer Institute Developmental Therapeutics Program (Weinstein et al., 1997). One database contained the baseline microarray measurements of the expression level of 6701 genes measured in these cell lines. The other database contained a set of drug susceptibility measurements for the same cell lines, across
209
(a)
spi-1PU.1 CD86B7-2 RAD50 CD21 Gertminal center kinase
Casein kinase l, 2 Diacylglycerol kinase delta Arachidonate 5-lipoxygenase CD22 JNK3 Myosin-IC KCNN3 Ca activated K channel PI3-kinase p110 catalytic, isoform WIPWASP interacting protein JAW1 APS adapter protein Protocadherin 43 Terminal deoxynucleotide transferase Focal adhesion kinase BCL-7A BCL-6 FMR2 A-myb CD10 OGG18-oxyguanine DNA glycosylase LMO2 CD38 CD27 Ick IRS-1 RDC-1 ABR OP-1 RGS13 PKC delta MEK1 SIAH-2 IL-4 receptor alpha chain APRPMA-responsive peptide GADD34 IL-10 receptor beta chain c-myc NIK ser/thr kinase BCL-2 MAPKK5 kinase PBEFpre-B enhancing factor TNF alpha receptor II Cyclin D2 Deoxycytidylate deaminase IRF-4 CD44 FLIPFLICE-like inhibitory protein SLAPsrc-like adapter protein DRIL1Dead ringer-like 1 Trk3Neurotrophic tyr kinase receptor IL-16 SP100 nuclear body protein LYSP100 K channel, shaker-related, member 3 ID2 NET tyrosine kinase IL-2 receptor beta chain
(b)
All patients
1.0
GC B-like Probability
been addressed using gene expression microarrays, including melanoma (Bittner et al., 2000) and prostate cancer (Singh et al., 2002). Other types of high-bandwidth modalities have been used to assist in other diagnostic questions, including proteomics (Petricoin et al., 2002; Robinson et al., 2002) and metabolomics (Clayton et al., 2006) (see also Chapters 14 and 15). While enabling physicians to make more efficient and accurate decisions between diseases itself is remarkable, the potential for high-bandwidth molecular testing in diagnosis is even greater. Beyond just improving on older tests for established diagnostic dilemmas, high-bandwidth studies have also been used to redefine and even divide diagnoses into novel sub-types. The earliest example of this was in the work of Alizadeh and colleagues in 2000 (Alizadeh et al., 2000). Alizadeh used gene expression microarrays to measure gene expression in samples from patients diagnosed with diffuse large B-cell lymphoma, a cancer of B-lymphocytes. This type of cancer was previously divided into low-, intermediate-, and high-grade categories based on growth pattern and immunohistochemistry. After applying a bioinformatics technique called hierarchical clustering (described later in this chapter) to the microarray data, Alizadeh discovered that the patient samples of B-cell lymphoma could be divided into an equal split of two subtypes (Figure 17.2). Most importantly, patients with these two different sub-types of lymphoma retrospectively demonstrated significant differences in survival by Kaplan–Meier analysis. This was a landmark study because it showed how high-bandwidth molecular measurements could provide a next level of resolution to resolve sub-types of disease not otherwise discernable by physicians. This type of study has been replicated for other types of cancers, including lung cancer (Bhattacharjee et al., 2001) and breast cancer (Sorlie et al., 2001), has been replicated using other types of measurements, including proteomics (Welsh et al., 2003), and is starting to enter clinical trials (Potti et al., 2006).
■
19 patients, 6 deaths 0.5 Activated B-like 21 patients, 16 deaths P 0.01
0.0 0
2
4 6 8 10 Overall survival (years)
12
Figure 17.2 (a) Ash Alizadeh and colleagues measured genome-wide expression levels in samples from patients with diffuse large B-cell lymphoma and found samples clustered into two sub-types based on expression patterns. (b) Patients in these two sub-types have a significant difference in survival (Alizadeh et al., 2000). Reprinted by permission from Macmillan Publishers Ltd: Nature 403:503, copyright 2000.
210
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
nearly 5000 anti-cancer agents. They then used a database technique known as join to combine these two databases through the cell lines in common. After joining the two tables, they searched for baseline RNA expression levels in the cell lines that correlated with the inhibition of growth of these same cell lines by thousands of anti-cancer agents. They found only one network containing an association between a gene expression and a measure of anti-cancer agent susceptibility (Figure 17.3). The association suggested that increased expression of lymphocyte cytosolic protein-1 (LCP1) is associated with increased susceptibility to the anti-cancer agent NSC 624044, a thiazolidine carboxylic acid derivative. Though a specific role for LCP1 in tumorogenicity had been postulated and though other thiazolidine carboxylic acid derivatives were known to inhibit tumor cell growth, there was no known relationship between this specific anti-cancer agent and gene in the biomedical literature. The significance of this work was that drug efficacy measurements could be joined to microarray measurements and potentially other types of
genomic measurements, enabling the discovery of hidden associations between drugs and genes. Fliri and colleagues joined databases in a similar manner in 2005, but instead of joining using cell lines, they used the drugs themselves (Fliri et al., 2005). Specifically, they took a database of known side-effects of drugs and joined it to a database of ligand binding assay results as measured in the presence of those same drugs. With this method, they were able to show that drugs with similar side-effect profiles actually had similar ligandbinding assays. The significance of this finding is that pre-clinical studies, such as ligand-binding assays and others, could be used to predict clinical side-effects, which, when severe enough, can lead to withdrawal of a drug from the market. Lamb and colleagues in 2006 built a database from microarrays measured across the genome after the application of 164 different small molecules to a breast cancer cell line (Lamb et al., 2006). They then queried this database of drug responses using responses measured from other experiments for which a detailed
BioDn-3 AFFX-M27830_5 D32129_f AFFX-M27830_5 BioDn-3 AFFX-M27830_5 HG3597-HT3800_f M14199 P624044 J02923
U43901_rna1
P169517
HG2743-HT2845
P685105 P637436
M83216 L18877_f HG2743-HT2846
L18920_f
U03735_f
Figure 17.3 Relevance networks calculated by joining drug activity data with gene microarray measurements. White rectangles indicate genes, and green rectangles indicate anti-cancer agents. The only discovered network involving both genes and drugs is shown in the middle. The activity of four anti-cancer agents against multiple cancer cell lines correlates with the expression pattern of a single gene across those same cancer cell lines. J02923 indicates the gene LCP1, while drug identifiers correspond to National Cancer Institute NSC identifiers (Butte et al., 2000).
Vignettes: How Specific Bioinformatics Methods can Change the Practice of Medicine
mechanism had not been known, using a measure of similarity known as a Kolmogorov-Smirnov statistic (Figure 17.4). For example, when queried with the gene expression differences between dexamethasone-treatable and dexamethasone-resistant ALL, they found that the genome-wide difference between these two types of leukemia most closely matched that of the genome-wide response to the drug sirolimus. The significance of this approach is that existing drugs could be associated with unexplained diseaserelated processes, and might even be used to treat those diseases. But most importantly, this work highlights the importance of building structured searchable repositories of experimental data. Histopathology Histopathology is the study of diseased tissue by sectioning, staining, and multi-resolution microscopy. Given high-bandwidth measurements from samples of potentially diseased tissue, bioinformatics methods can be applied to determine whether such samples do quantitatively resemble a disease (van de Rijn and Gilks, 2004). For example, Ramaswamy and colleagues tackled the problem of finding the original source of a cancer, given just a metastatic sample (Ramaswamy et al., 2001). They used a method called support vector machines, each of which was designed to recognize one specific type of cancer (Figure 17.5). This work was significant in that it was one of the first to show how multiple types of cancers could be distinguished using high-bandwidth measurements, and that unknown samples could be accurately assigned a primary diagnosis. Similar work has been published using alternative bioinformatics methods (Su et al., 2001). Fibroblasts are cells in the connective tissue that create much of the extracellular matrix between tissues and proliferate during wound healing. Though fibroblasts from different parts of the body may have similar properties and resemble each
Biological state of interest (signature)
■
other under a microscope, Chang and colleagues demonstrated that when fibroblasts are considered by high-bandwidth measurements of gene expression, significant differences between fibroblasts are seen depending on where in the body they were originally obtained. In other words, cells that appear exactly the same by microscope appear quite differently by microarray, consistent with their original source (Chang et al., 2002). Microarrays have also been used to uncover histopathological findings that might have otherwise been missed. Sarwal and colleagues used microarrays to study the process of acute rejection after kidney transplantation (Sarwal et al., 2003). After hierarchically clustering their samples, they noticed a set of B-cell specific genes that corresponded to one sub-type of acute rejection. This was particularly interesting because B-cells had not previously been thought to be involved in acute rejection. Staining for a B-cell specific cell surface protein showed large aggregations of B-cells in samples of acute rejection. The significance of this finding was the realization that while histochemistry can offer a broad survey of the cellular makeup in a sample, these cells will still remain invisible unless illuminated with the right stain. The proper stain can be found using gene expression microarrays coupled with the right bioinformatics techniques, especially the prior annotation of genes as being specific for a certain cell type. Nosology In the mid 1700s, Carl Linnaeus promoted the binomial nomenclature to classify living things into a hierarchy, or taxonomy. Though we still organize species into taxonomies based on anatomic and physiological similarities, our modern-day use of DNA sequencing and genome sequence has enabled the reorganization of the position of species within established taxonomical trees.
Reference database (profiles)
Connections
positive up
query
output
down negative strong positive
weak positive
211
null
strong negative
Figure 17.4 A reference database of high-bandwidth molecular measurements can be made after the application of various drugs, then queried with an unknown biological response, yielding those drugs sharing a similar response. From Lamb et al., 2006, Science 313:1929. Reprinted by permission from AAAS.
212
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
This is because DNA sequence similarity analysis provides a quantitative measurement of how different two species are (Baldauf et al., 2000). Linnaeus was also a co-founder of systematic nosology, or the classification of disease, for which he is not as well known (Linnâe and Schrèoder, 1763). Genomic data and bioinformatics methods have now advanced to the point that we can begin to modernize the classification of disease itself, similar to how DNA sequencing has modernized taxonomy. Crescenzi and Giuliani were one of the first to demonstrate this principle in 2001 (Crescenzi and Giuliani, 2001). They gathered samples representative of 60 cancer cell lines and used a statistical method known as principal components analysis to find combinations of genes that capture the significance variance across cancers. They then showed how analyzes using just the top five principal components could reproduce the findings from the entire dataset. In other words, Crescenzi and Giuliani were able to show that when measured using high-bandwidth measurements,
the 60 varied types of cancers differed in only a few ways that could be modeled. Instead of modeling the significant differences between cancers, Rhodes and colleagues searched for commonalities in 2004 (Rhodes et al., 2004). After collecting 40 published microarray datasets totaling more than 3700 samples of cancer, Rhodes determined a genome-wide signature representative of neoplastic transformation. As expected, most of the genes in this cancer signature were known to participate in the cell cycle, the biological process involved in cell replication. A similar study by Bild and colleagues found commonly deregulated biological pathways across cancers that correspond with worsening survival (Bild et al., 2006). This type of work is significant because it suggests that a quantitative classification of cancers, or one based on actual measurements, might be more useful than the traditional anatomic or even histopathological classification of cancer (Figure 17.6).
SAMPLES BR BL CNS CO
LE
LU
LY
ME ML OV PA PR RE UT
HIERARCHICAL CLUSTERING CEA
DATASET
SOM
MOLECULAR MARKERS
GENES
(16,063 Genes, 218 Human Tumor Samples)
BR PR LU CO LY BL ML UT LE RE PA OV ME CNS
PSA
ER 3
2 1 0 1 2 3 standard deviation from mean
Figure 17.5 Over 200 samples of 14 different types of tumors were clustered to indicate similarities in tumor types and then classified to find genes uniquely implicating each type of tumor (Ramaswamy et al., 2001).
Control
Myc
Control
Ras
50
50
150
150
200 250
200 250 2 4 6 8 10 12 14 16 18 20
E2F3
30
150
40
200
50
250
350
300
Src
60 70 2 4 6 8 10 12 14 16 18
Control
β-catenin
10 20 30 40 50 60 70 80 90 100
20
100
300 2 4 6 8 10 12 14 16 18 20
Control 10
50
100 100
Control
2 4 6 8 10 12 14 16 18
2 1.5 1 0.5 0 0.5 1 1.5 2
2 4 6 8 10 12 14 16 18
Figure 17.6 Cells made to over-express particular oncogenes were measured using gene expression microarrays, and a “signature” or pattern of gene expression changes for each oncogene was then computed and used to probe samples of cancer (Bild et al., 2006). Reprinted by permission from Macmillan Publishers Ltd: Nature 439:353, copyright 2006.
Analytic Methods
ANALYTIC METHODS The field of computer science involved in the search for patterns and the induction of rules from data is called machine learning. Current methodologies to analyze high-bandwidth measurements can broadly be divided into two categories: supervised machine learning approaches, or analysis to determine genes or proteins that fit a predetermined pattern; and unsupervised machine learning approaches, or analysis to characterize of the components of a dataset, without the a priori input or knowledge of a right answer (Butte, 2002). Supervised machine learning methods are commonly used to find those individual or sets of measured elements with measurements that (1) are significantly different between defined groups of samples, and (2) accurately predict a characteristic of the sample. An example of an application of a supervised machine learning method is the vignette described above in which Golub and colleagues found genes that distinguish ALL and AML. There are many published supervised methods that find measured elements that accurately predict characteristics of samples, such as distinguishing one type of disease from another, or a malignant disease from a benign one (Dudoit et al., 2002). There are methods that find individual measured elements, such as the nearest-neighbor approach (Golub et al., 1999), and those that find sets of multiple elements, such as decision trees (Quinlan, 1992), neural networks (Rumelhart et al., 1986), and support vector machines (Brown et al., 2000; Chow et al., 2001; Furey et al., 2000). Unsupervised machine learning methods are used to find internal structure or relationships within a dataset, instead of trying to determine how best to predict a “correct” answer. Two of the example vignettes provided above – in which Crescenzi determined those genes that best explained the differences between cancers and in which Alizadeh found sub-types of Bcell lymphoma – are good examples of the use of unsupervised machine learning methods. Within unsupervised learning, there are three classes of techniques, including (1) measured element discovery, or finding elements (e.g., genes) with interesting properties without specifically searching for a specific predefined pattern, such as by principal component analysis (Alter et al., 2000; Fiehn et al., 2000; Hilsenbeck et al., 1999; Raychaudhuri et al., 2000; Wen et al., 1998); (2) cluster discovery, or finding groups of measured features or samples with similar patterns, such as by nearestneighbor clustering (Ben-Dor et al., 2000; Golub et al., 1999), self-organizing maps (Tamayo et al., 1999;Toronen et al., 1999), k-means clustering, and one and two-dimensional hierarchical clustering (Eisen et al., 1998; Ross et al., 2000); and (3) network discovery, or finding graphs representing associations between measured elements, such as by using Boolean networks (Liang et al., 1998; Szallasi and Liang, 1998; Wuensche, 1998), Bayesian networks (Friedman et al., 2000), and relevance networks (Butte and Kohane, 1999; Butte and Kohane, 2000; Butte et al., 2000). Several recent articles have reviewed many of the available methods (Butte, 2002; Quackenbush, 2004, 2006). Here we will
■
213
consider two of the most commonly used supervised and unsupervised methods used in those publications with the greatest impact for medicine. Case-Control Studies The majority of experiments performed using high-bandwidth measurements still typically use only a handful of cases in two or three conditions, and the experimenter typically has the goal of finding those measured elements (e.g., genes) that are significantly different between these groups. Significance has and continues to be evaluated in a multitude of different methods, including parametric (Tusher et al., 2001), non-parametric (Butte et al., 2001), Bayesian (Baldi and Long, 2001), and many others. Analysis of variance has been used for measurements from samples with multiple characteristics (Pavlidis and Noble, 2001). Signal processing (Butte et al., 2002), trend detection (Tseng et al., 2005), Bayesian (Ramoni et al., 2002), non-parametric (Reis et al., 2001), and many other methods have been used for measurements from samples across time (Simon et al., 2005). The challenge in using any of these methods across the set of measured elements (e.g., genes) is compensating for the high number of measurements made. For example, even a standard t-test could be used on every gene measured in two sets of microarrays from two conditions. The null hypothesis for a particular gene would be that the gene expression level is not significantly different across the two conditions, and a p-value can be calculated to assess the ability to reject that null hypothesis for each gene. But in such a case, it would be incorrect to use the “standard” threshold of p 0.05 to determine those genes that are significantly different. Traditionally, such a threshold indicates the likelihood of calling a truly negative finding positive, otherwise known as the false positive rate or a type 1 error in statistics. For a single test, a 5% false positive rate may not be unreasonable, but when multiplied across 40,000 transcripts, such a false positive rate would be expected to yield 2000 falsely positive genes, too many to validate. With high-bandwidth measurements, one is typically more interested in the false discovery rate, rather than the false positive rate. The false discovery rate is the expected proportion of accepted type 1 errors within a list of tested hypotheses, and controls for the number of tests being made. A false discovery rate can actually be calculated for every measured element. In essence, such a rate would indicate the likelihood of false discoveries if that element were considered positive. In practice, this is called the q-value for that element (Storey and Tibshirani, 2003). Those elements with low p-values also have low q-values. Though q-values and false discovery rates are gaining in popularity, it is important to note that there have been many other proposed methods to compensate for multiple-hypothesis testing (Dudoit et al., 2003b). Sub-Type Discovery The discovery of disease sub-types using high-bandwidth measurements has been the prime motivator for personalized disease
214
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
therapeutics, where a sample of disease from an individual could be potentially measured and the most appropriate set of therapies potentially prescribed. Typically, the discovery of sub-types involves a two-step process. First, an unsupervised machine learning method is applied yielding subsets of samples (e.g., patients) with similar measurements. Many choices are available in this step, including the specific clustering method used (e.g., hierarchical clustering (Eisen et al., 1998), self-organizing maps (Tamayo et al., 1999), or others), the mathematical definition of similarity used (e.g., Pearson correlation coefficient, Spearman rank correlation, Euclidean distance, or others), and the threshold similarity at which clusters are demarcated. While this first step will yield potential sub-types of disease, the utility of these sub-types still needs to be determined. Typically, utility is defined using a marker of clinical outcome, such as survival, disease-free survival, or response duration. The two or more sub-types of disease identified in the first step are then tested statistically against the marker, using a Kaplan–Meier estimator or Cox Proportional Hazards Model. The most important point to note with these types of analyzes is the censored nature of the samples. A particular patient that is lost to follow-up can only “contribute” his or her survival up to the length of follow-up. Recent methodological work has provided numerous new methods that combine both of these steps, enabling the single-step discovery of sets of measured elements (e.g., genes) that best discriminate for censored survival or
other outcomes (Bair and Tibshirani, 2004; Nguyen and Rocke, 2002; Park et al., 2002).
WHERE DATA FOR STUDIES MAY BE FOUND Corresponding with the successful application of highbandwidth molecular measurement modalities, the amount of data in international repositories has grown exponentially, because top-tier journals require the public availability of such data and because of increased calls for transparency of raw data in publications (Nature, 2002; Ball et al., 2004; Perou, 2001). For example, the Gene Expression Omnibus (GEO) is an international repository for gene expression data, developed and maintained by the National Library of Medicine (Wheeler et al., 2004). As of this writing, GEO holds 103,000 samples (i.e., microarrays) from over 4300 experiments involving 189 species, across over 2100 types of microarrays, with 2.6 billion individual gene measurements (Figure 17.7). More impressively, GEO has been gaining data at 300% per year. GEO is not the only international repository for microarray data. ArrayExpress is a similar database of gene expression measurements supported by the European Bioinformatics Institute (Parkinson et al., 2005). As of this writing, ArrayExpress contains 42,424 samples from 1437 experiments. The Stanford Microarray Database (SMD) was the first international repository
Figure 17.7 Screen shot of the home page for the National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO).
Bioinformatics Vocabularies and Ontologies
for microarray data (Ball et al., 2005). As of this writing, SMD holds 61,573 samples involving 37 species, with a total of 1.84 billion individual gene measurements. It is not clear how much overlap there is in experiments with GEO, but it is likely that over 150,000 unique samples are available for study. Recently, international repositories for proteomic experimental datasets have been instituted (Prince et al., 2004). PeptideAtlas, supported by the Institute for Systems Biology (ISB), is an example of one such repository (Desiere et al., 2006; Deutsch et al., 2005). As of this writing, PeptideAtlas holds the raw and processed mass spectra for 76 experiments. The Proteomics Identifications Database (PRIDE), supported by EBI, currently holds data from over 1655 experiments (Jones et al., 2006). Though GEO, ArrayExpress, SMD, and PeptideAtlas are already incredible resources for gene expression measurements, these measurements are poorly indexed. Even for microarray data stored in standardized formats like MIAME and MAGE-ML (Brazma et al., 2001; Spellman et al., 2002), the contextual annotations needed to determine relevance of a dataset are unfortunately still represented by unstructured narrative text. Determining the phenotypes, diseases, and environmental contexts studied by these experiments is no longer a tractable manual process. Most repository websites provide only a free-text based search facility for experimental annotations. A case example demonstrates the significance of the problem. At the time of this writing, searching on the NCBI GEO website yields 27 datasets annotated with the term “breast cancer.” However, a search for “breast tumor” yields only 8 datasets. A search for the plural, “breast tumors,” yields 9 datasets, while a search for “breast carcinoma” yields only 6 datasets. Breast cancer is one of the most important cancers affecting women, but searching for existing microarray datasets relevant to this important disease yields arbitrary results and is fraught with false negatives. The basic problem is that data in these repositories are not consistently annotated with terms from a standardized vocabulary. While some initial work has been done in re-annotating these datasets using a standardized vocabulary (Butte and Kohane, 2006), investigators interested in finding relevant data will have to try multiple terms to comprehensively find data. Another caveat is that much of the most relevant clinical data associated with high-bandwidth measurements may not be stored in international repositories. For example, a dataset of microarray measurements on patients with varying lengths of survival after cancer diagnosis may be available in GEO, but the survival times may not be. Such clinical data may be available on an investigator’s laboratory website or in an associated publication, but is often not publicly available at all.
BIOINFORMATICS VOCABULARIES AND ONTOLOGIES A cell biologist interested in finding clinical samples relevant to a disease being studied in the lab will not find those samples unless she can “speak the same language” as the pathologists
■
215
storing and indexing the samples. A physician interested in learning more about genes that are significantly different in his experiments will not be able to learn more about the pathways implicated by those genes, unless he can “speak the same language” as those geneticists providing annotative information for those genes. In the application of bioinformatics to genomic and personalized medicine, it is often the case that shifting focus like this crucially depends on the use of a standardized vocabulary. The formal modeling and representation of concepts of knowledge, the terms and other attributes used to describe these concepts, and the relationships of these concepts, is called a ontology. The importance of using proper ontologies in biomedical research has been described in recent publications (Blake, 2004; Soldatova and King, 2005). Here, we will describe some of the more commonly used ontologies used in the research of genomic medicine (Table 17.2). Clinical vocabularies The largest disease ontology (or nosology) in use today is the Systemized Nomenclature of Medicine-Clinical Term (SNOMED-CT) ontology. SNOMED-CT has a lineage spanning 75 years and is used by pathologists world-wide (Chute, 2000). SNOMED-CT has over 340,000 biomedical concepts in 18 hierarchies with 1.3 million relationships that classify diseases based on syndromic and pathophysiological mechanisms (Cho et al., 1998). For example, relations in SNOMED-CT assert that clear cell carcinoma of the kidney is a malignant tumor of the kidney (i.e., a type of tumor of a particular organ) and separately is also
TABLE 17.2 Some of the ontologies and vocabularies in use in bioinformatics Ontology and vocabulary
Use
Systemized Nomenclature of Medicine-Clinical Term (SNOMED-CT)
Pathological based classification of disease
International Classification of Diseases (ICD) with Clinical Modification
Diagnosis codes for billing, public health, and epidemiology
Gene Ontology (GO)
Molecular functions, biological processes, and cellular components of proteins
Open Biomedical Ontologies (OBO)
Includes Gene Ontology and many other ontologies describing cell types and anatomy
InterDom
Protein–protein interactions
InterPro
Protein domains
Unified Medical Language System (UMLS)
Unified compendium of 140 biomedical vocabularies
216
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
a malignant neoplasm of the retroperitoneum (i.e., a type of tumor in a particular anatomical location). Another disease classification more familiar to physicians is the International Classification of Diseases (ICD) with Clinical Modification, a nosology with origins in the 1850s. ICD-9-CM has nearly 7000 codes, while ICD-10-CM has over 14,000 codes to describe causes of morbidity and mortality (World Health Organization, 2005). Both ICD-9-CM and ICD-10-CM are in use by various health care providers, the US Federal Government, and the World Health Organization. Though most health care institutions in the United States use ICD-9, ICD-10 was recently approved for use by the US Congress. To continue the example above, relations in ICD-9-CM assert that a malignant neoplasm of the kidney is a malignant neoplasm of the genitourinary organs. Importantly, however, the limited resolution in ICD-9-CM does not permit the specific diagnosis of clear cell carcinoma of the kidney. There are several important differences between the SNOMED-CT and the ICD nosologies. As shown above, SNOMED-CT has greater expressiveness than ICD to represent diseases. SNOMED provides a rich set of relationships between medical concepts. ICD provides only a hierarchical relationship between concepts. Despite their differences, it is important to note that both SNOMED-CT and ICD are viewed as administrative nosologies, typically used for billing, reporting, electronic medical records, and decision support systems. However, important sources of stored tissues in medical institutions, such as frozen tissue banks, may only be accessible through SNOMED or ICD codes, and these coding systems commonly serve as an index into these repositories. Genomic Ontologies The Gene Ontology (GO) is a taxonomy that is used to describe the normal molecular function of proteins, the cellular components in which proteins operate, and the larger biological processes in which they participate (Ashburner et al., 2000). At the time of this writing, GO currently holds 7470 molecular functions, 1823 cellular components, and 12,250 biological processes. Over 186,000 genes across a variety of species have been assigned to one or more GO categories, but this essentially translates to one-third of human genes having some coded function. The easiest way to find the GO classification for a gene is to execute a query using Entrez Gene. For example, a query for the gene coding for the insulin receptor (INSR) yields 13 molecular functions (e.g., phosphoinositide 3-kinase binding), 7 biological processes (e.g., carbohydrate metabolism), and 2 cellular components. It is important to note that an actual publication detailing the evidence for each annotation is available for many GO annotations, and these publications can be easily referenced. It is also important to note that the GO annotations for genes can change quickly, as both the annotations for genes and GO itself continue to be “works in progress.” The GO is actually one of 61 ontologies in the set known as Open Biomedical Ontologies (OBO) (Mungall, 2004). These ontologies are open, in the sense that they can be freely downloaded and used without constraints. Other ontologies in OBO include ontologies to describe the gross anatomy of flies,
worms, and mice, cell types, and functional portions of proteins, as well as ontologies to describe cancers and other concepts related to humans. Other vocabularies used for gene and protein annotation include InterDom (Ng et al., 2003), InterPro (Mulder et al., 2003), and PRINTS (Attwood et al., 2003) for protein domains, and other gene identifiers include the stable NCBI Gene identifier (Wheeler et al., 2006) and the unstable NCBI UniGene identifier (Wheeler et al., 2000). Unified Medical Language System The Unified Medical Language System (UMLS) is the largest available compendium of biomedical vocabularies, containing 140 biomedical vocabularies with over 1.2 million concepts and 41 million relations between concepts (Bodenreider, 2004; Butte and Kohane, 2006). UMLS already unifies vocabularies used extensively in molecular biology and genomics, such as the Medical Subject Headings (MeSH), NCBI Taxonomy, and the GO, with medical vocabularies including the International Classification of Diseases and SNOMED Clinical Terms (Ashburner et al., 2000; ICD, 2003; Wheeler et al., 2004). The UMLS is a unified vocabulary, which means that concepts listed in multiple vocabularies are brought together. For example, the concept of fever, which holds the UMLS concept unique identifier of C0015967, is represented in 78 component vocabularies, such as D005334 in MeSH, 386661006 and 50177009 in SNOMED-CT, 780.6 in ICD-9-CM, 10016558 in the Medical Dictionary for Regulatory Activities Terminology (MedDRA), X25 in Perioperative Nursing DataSet, GO:0001660 in GO, and U001776 in the Library of Congress Subject Headings. Each of these individual codes represents the same concept, and thus have been unified to a single biomedical concept in UMLS. Because of its ability to span nearly every other relevant vocabulary, UMLS best serves as a bridging vocabulary, providing terms commonly used by both physicians as well as molecular biologists. Given the support for UMLS and its component vocabularies shown by the National Library of Medicine and the US Department of Health and Human Services, it is likely that terms and concepts from UMLS will constitute important portions of the future electronic health records of patients in the United States (see also Chapter 19). As samples from patients continue to be valuable for study using high-bandwidth measurements, tapping into the medical records associated with these samples will only grow in importance and UMLS can help enable these secondary research uses of medical records.
FREELY AVAILABLE BIOINFORMATICS TOOLS There is an incredible preponderance of software tools freely available for research in bioinformatics. As an example, as of this writing, over 300 articles are published each year specifically describing bioinformatics software tools in the journals BMC Bioinformatics, Bioinformatics, Nucleic Acids Research, and Genome Biology. The vast majority of these are freely available for
Freely Available Bioinformatics Tools
■
217
Figure 17.8 Screen shot of the MeV (Saeed et al., 2006). Reprinted from Methods in Enzymology, Volume 411, Saeed, A.I., Bhagabati, N.K., Braisted, J.C., Liang, W., Sharov, V., Howe, E.A., Li, J., “TM4 microarray software suite”, 134–93, copyright 2006, with permission from Elsevier.
researchers. Describing all of the available free tools is clearly intractable, though a recent book and paper serve as nice collections of most of the commonly used tools (Dudoit et al., 2003a; Parmigiani, 2003). Here instead, we will arbitrarily focus on a few important tools useful in the study of genomic and personalized medicine, dividing our set into two groups: those useful in the analysis of high-bandwidth data and those useful in the interpretation of results. Analytic Tools The Multiexperiment Viewer (MeV) is an easy-to-use analytic tool initially developed at the Institute for Genomic Research and currently maintained at the Dana Farber Cancer Institute (Saeed et al., 2006). It specifically supports 17 tools for the analysis of gene expression microarrays, and presents these tools using easy-touse on screen buttons (Figure 17.8). It runs under the Macintosh, Windows, and Linux operating systems, and the programming source code is available and can be modified by bioinformatics programmers. At the time of this writing, MeV has been cited by nearly 200 publications and has been extensively documented. GenePattern is a sophisticated analytic tool developed and maintained by the Broad Institute (Reich et al., 2006). GenePattern supports the analysis of multiple types of high-bandwidth measurements, including gene expression microarrays, proteomics, and single nucleotide polymorphisms. Each analytic and visualization tool is contained within a module, which can be chained together to represent a reproducible pipeline for analysis (Figure 17.9).
GenePattern runs under the Macintosh, Windows, and Linux operating systems. GenePattern is well-documented and supported, and as of this writing, approximately 30 papers have cited its use. GenePattern clearly excels in its advanced features. One particularly useful feature of GenePattern is that it offers a modular architecture allowing users to plug in new analytic tools or visualization methods as they become available. These tools can be created to interface with other bioinformatics analytic tools, including Matlab and R and Bioconductor (described below). GenePattern can distribute the work required for an analysis onto a parallel computer cluster, which, if available, can speed the time required to complete an analysis. In addition, GenePattern supports the output of measurements into the standardized formats necessary for submission to international repositories, a step that is increasingly required before a top-tier journal accepts a manuscript. At the opposite extreme of user-friendliness lies the statistical computing and graphics environment known as R (R Development Core Team, 2004). Based on the S statistical computing environment developed by Bell Labs in the late 1970s, R is the most flexible of the tools listed here. R is almost essentially “command line” driven, in that commands to execute statistical and visualization tools are typed and interpreted serially, but most users write scripts and sophisticated programs that can be run to generate the appropriate output and figures. There is essentially no user-interface (i.e., menus and windows) that can be used to perform analyzes; getting up-to-speed with R involves climbing a steep learning curve.
218
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
Analytical methodology
Pipeline representation and results
1 Heat map
Gene expression data set
Gene list significance
Pre-processing
2 Gene selection
Marker genes
1
Build predictor
Cross-validation results
2
3
Model
Cross-validation results
Test set prediction results
Test data set
Test set prediction results
3
Figure 17.9 The conceptual analytic “pipeline” needed to analyze high-bandwidth molecular data is represented on the left, while its computational equivalent is represented on the right. Using GenePattern, these steps can be coded and documented, yielding reproducible research results (Reich et al., 2006). Reprinted by permission from Macmillan Publishers Ltd, Nature Genetics 38, 500, copyright 2006.
Despite the severe user-interface handicaps, the most important advantage of R is its community of users. Because of its designed similarity to S, users of R can benefit from decades of developed analytic tools (some statistical tool plug-ins for R and S available on the Internet date back to the late 1980s). The most important set of plug-in tools for research in genomic medicine are those contained in the Bioconductor package, which offers tools for reading microarray data, processing raw data files, common supervised and unsupervised machine learning methods, and visualization methods (Gentleman et al., 2004). More than 50 books have been written that explain both commonly-used and cutting-edge statistical techniques using R (and S), including a few specifically for bioinformatics and Bioconductor (Gentleman et al., 2004; Parmigiani, 2003). R runs
under the Macintosh, Windows, and Linux operating systems and its source code is available. Intepretation After several analyses, it quickly becomes obvious that the rate-limiting step in experiments involving high-bandwidth measurements is neither the handling of the biological samples nor the actual statistical or numerical work, but instead the post-analytic work in determining what the results actually mean. First, detailed names and knowledge might not yet exist for genes and proteins that have been found to be significantly involved in an experiment, even though these elements may have already been measured for many years. This complicates the interpretation of results. The official gene name, predicted
Freely Available Bioinformatics Tools
■
219
TCA cycle glycolysis fatty-acid degradation triacylglycerol
view same data set on hundreds of MAPPs
lipoprotein lipase glycerol
fatty acid
glycerol kinase
fatty acid CoA ligase
L-glycerol-3-phosphate
acyl-CoA
glycerol-3-PO4 dehydrogenase
carnitine acetyltransferase carnitine palmitoyltransferase
dihydroxyacetone phosphate
acyl-CoA
triosephosphate isomerase
acyl-CoA dehydrogenases: short chain medium chain long chain very long chain
glyceraldehyde-3-phosphate
see glycolysis MAPP
legend upregulated, P 0.05 downregulated, P 0.05 no criteria met not found
see TCA cycle MAPP
acyl-CoA (n2)
acetyl-CoA
thiolase
trans-Δ2-enoyl-CoA
3-ketoacyl-CoA
enoyl-CoA hydratase
3-OH-acyl-CoA dehydrogenase
click on genes for annotation, data and hyperlinks
3-OH-acyl-CoA
Figure 17.10 GenMAPP has the ability to color-code proteins participating in biological pathways based on measurements (Dahlquist et al., 2002). Reprinted by permission from Macmillan Publishers Ltd, Nature Genetics 31, 19, copyright 2002.
protein domains, or GO classification may become available for a gene or protein as early as tomorrow, or as late as years from now. Operationally, this means that one is never done analyzing a set of microarray data. One has to develop the infrastructure to constantly reinvestigate genes and proteins from analyzes performed in the past. It may be next week, for example, that new knowledge about a protein that was significantly implicated in an analysis performed three months ago gets published, finally leading to a novel and important finding. The challenges in determining the proper analytic methods to use are usually only a short-term difficulty, and typically after the “functional genomics pipeline” has been established, the rate-limiting step shifts to the post-analytic challenges (Kohane et al., 2002). In the future, truly demonstrating a “return on investment” from high-bandwidth molecular measurements depends on taking findings past the measurement stage and integrating them with the rest of the research pipeline, including interpretation and validation. The list of elements resulting from an analysis should not be viewed as an end in itself; its real value only increases as that list moves through biological validation, ranging from the numerical verification of measurement levels with alternative techniques, to the ascertainment of the meaning
of the results, such as finding common mechanisms or biological pathways involving the genes or proteins. While tools that link measurement elements back to known biological pathways are still in their infancy, some have been shown to be quite useful. GenMAPP is a software tool that can visualize of highbandwidth molecular measurements using pre-drawn diagrams illustrating biological pathways (Dahlquist et al., 2002). Though GenMAPP comes with nearly a thousand useful biological pathways across numerous species, GenMAPP also comes with an editor enabling a biologist or bioinformatician to draw any new pathway (Figure 17.10). Because of this, the GenMAPP user base has developed into a large community of users who share pathways. GenMAPP runs only under the Windows operating system. At the time of this writing, GenMAPP has been cited by nearly 600 publications and has been extensively documented. The Kyoto Encyclopedia of Genes and Genomes (KEGG) is an extensive web-based database of biological pathways and functions under development for more than a decade (Kanehisa, 1997). Similar to GenMAPP, KEGG offers tools that can color-code diagrams of biological pathways based on quantitative measurements, instead using a web-based interface (Figure 17.11). KEGG offers more than 42,000 biological pathways for nearly 500 species.
220
CHAPTER 17
Bioinformatic and Computational Analysis for Genomic Medicine
■
INSULIN SIGNALING PATHWAY
Flotillin APS
CAP
CrkII
Cb1
GRF2
Exo70
GLUT4 vesicle
TC10
Glucose homeostasis
Translocation to PM
Glucose uptake
GLUT4
CIP4/2
Glucose p
p p AMPK aPKC
SKIP
GK PGC-1 p
p
py
DNA
ps
IRS
PI3K
PDK1/2
p
p
Akt
GSK-3β
PIP3
py
p Phosphatidyl inositol signaling system
Glycolysis/ Gluconeogenesis
FBP
FOXO1
IKK
p py
Glycolysis G6PC
JNK
INSR
PYK
DNA
PTP1B
INS
Lipogenesis Fatty acid biosynthesis (path 1)
PFK
SREBP-1c
p
SOCS
ACC FAS
SHIP
p
PEPCK
Glycogen GYS
Glycogenesis
p p
PP1
PHK
p
Starch and sucrose metabolism
PYG
TSC2 TSC1
p
PKA
PDE3
p
Antilipolysis
HSL
cAMP Rheb
LAR
p
p
Antiapoptosis
BAD
p mTOR
MAPK signaling pathway
p
p p GRB2
SOS
S6
Raptor
p SHC
p
p70S6K
Lipid homeostasis
Apoptosis
Ras
Raf
4EBP1 p
MEK1/2
Protein synthesis
eIF4E p
p MNK
ERK1/2 p
ElK1 DNA
Proliferation, differentiation
Figure 17.11 The insulin signaling pathway, as represented in KEGG as of this writing. A click on any protein yields additional data about that protein.
Beyond biological pathways, KEGG also provides knowledge on ligands, metabolites, small-molecule drugs, and diseases. At the time of this writing, KEGG has been cited by nearly 1000 publications. Other free web-based resources available to study genes by known function and protein domains include Entrez Gene and GeneCards (Safran et al., 2002;Wheeler et al., 2006).
NEW QUESTIONS FOR GENOMIC MEDICINE While this chapter has demonstrated the utility of bioinformatics and computational biology for answering specific questions for genomic medicine, it is clear that future revolutionary discoveries are going to come from those who can handle, manage, and exploit the exponential growth in measurement methods and data. Even as newer high-bandwidth systems become available for measuring proteins, lipids, and protein-interactions, it is clear that significant discoveries can be enabled by putting measurement-systems and data together, an area of research informally termed integrative biology. Here, we will describe some significant studies that involved the integration of large-scale measurements across modalities. In one integration study, Schadt and colleagues started with gene expression differences in liver between two inbred strains of mice. They used the most significantly different genes as traits
which were then mapped as quantitative trait loci using genetics. They first highlighted genes with known polymorphisms that affected their transcript levels, then showed loci associated with fat pad mass (and thus relevant to obesity) that would otherwise have been insignificant had the expression data not been considered (Schadt et al., 2003a, b). In other words, their association of genes with obesity required both microarray and genetic tools. Mootha and colleagues integrated four publicly available expression datasets with linkage data and proteins identified from mitochondria to ascertain the gene and mutation responsible for Leigh syndrome, French-Canadian type (Mootha et al., 2003). Stoll and colleagues integrated 125 phenotypes with linkage data from rats to determine candidate genes potentially involved in cardiovascular function (Stoll et al., 2001). Chiang and colleagues took genes known to be associated with Bardet-Biedl syndrome and found a set of species in which these genes were conserved and other species in which these genes did not exist. Using this comparative genomic information, Chiang was able to predict and discover a novel gene with a mutation for Bardet-Biedl syndrome (Chiang et al., 2004). English collected 49 genome-scale experiments all related to the study of obesity, but of varying types, including gene expression microarrays, proteomics, RNA interference, and genetic scans. She evaluated the sensitivity and specificity of each against a gold-standard list of known obesity-associated genes. While the sensitivity and specificity of each independent experiment was
References
poor, she found that a simple integrative model could statistically significantly outperform each of the independent experiments in rediscovering known obesity-associated genes. Moreover, she found that the likelihood of discovering a novel obesity-associated gene increased when pairs of genome-scale experiments were considered, instead of individual experiments. This work showed that integrating experiments performed better than single experiments at finding complex disease-associated gene variants, and that it is now possible to predict how likely a gene is to have a variant associated with a complex condition (English and Butte, 2007). These examples are given to demonstrate that exploration into the integration of multiple high-bandwidth modalities has started. Instead of focusing on the cell, or the genotype ranging from one locus to the entire genome, or on any single measurement modality, integrative biology allows one to think holistically and horizontally. A disease such as diabetes can lead to myocardial infarction, nephropathy, and neuropathy; to study diabetes in the era of genomic and personalized medicine would require reasoning from a disease to all its various complications to the genome and
■
221
back, across many vantage points and cellular models that bridge the genome sequence, gene expression, proteomics, metabolomics, and phenotype. As we have shown in this chapter, there is an enormous supply of prior data that can be quite potent, if it can be brought to bear on the right problems, with the right tools. Though it has yet to be formally proven, the future study of complex diseases is highly likely to benefit from integrative biology and bioinformatics.
ACKNOWLEDGEMENTS The work was supported by grants from the Lucile Packard Foundation for Children’s Health, National Library of Medicine (K22 LM008261), National Institute of General Medical Sciences (R01 GM079719), Howard Hughes Medical Institute, and the Pharmaceutical Research and Manufacturers of America Foundation. The author is a scientific advisor to Genstruct, Inc., a computational systems biology company.
REFERENCES Adida, B. and Kohane, I.S. (2006). GenePING: secure, scalable management of personal genomic data. BMC Genomics 7, 93. Alizadeh, A.A., Eisen, M.B., Davis, R.E., Ma, C., Lossos, I.S., Rosenwald, A., Boldrick, J.C., Sabet, H., Tran, T., Yu, X. et al. (2000). Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature 403, 503–511. Alter, O., Brown, P.O. and Botstein, D. (2000). Singular value decomposition for genome-wide expression data processing and modeling. Proc Natl Acad Sci USA 97, 10101–10106. Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D., Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K., Dwight, S.S., Eppig, J.T. et al. (2000). Gene ontology: Tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25–29. Attwood, T.K., Bradley, P., Flower, D.R., Gaulton, A., Maudling, N., Mitchell, A.L., Moulton, G., Nordle, A., Paine, K., Taylor, P. et al. (2003). PRINTS and its automatic supplement, prePRINTS. Nucleic Acids Res 31, 400–402. Bair, E. and Tibshirani, R. (2004). Semi-supervised methods to predict patient survival from gene expression data. PLoS Biol 2, E108. Baldauf , S.L., Roger, A.J., Wenk-Siefert, I. and Doolittle, W.F. (2000). A kingdom-level phylogeny of eukaryotes based on combined protein data. Science 290, 972–977. Baldi, P. and Long, A.D. (2001). A Bayesian framework for the analysis of microarray expression data: regularized t-test and statistical inferences of gene changes. Bioinformatics 17, 509–519. Ball, C.A., Brazma, A., Causton, H., Chervitz, S., Edgar, R., Hingamp, P., Matese, J.C., Parkinson, H., Quackenbush, J., Ringwald, M. et al. (2004). Submission of microarray data to public repositories. PLoS Biol 2, E317. Ball, C.A., Awad, I.A., Demeter, J., Gollub, J., Hebert, J.M., HernandezBoussard, T., Jin, H., Matese, J.C., Nitzberg, M., Wymore, F. et al. (2005). The Stanford Microarray Database accommodates additional microarray platforms and data formats. Nucleic Acids Res 33, D580–D582.
Ben-Dor, A., Bruhn, L., Friedman, N., Nachman, I., Schummer, M. and Yakhini, Z. (2000). Tissue classification with gene expression profiles. J Comput Biol 7, 559–583. Bhattacharjee, A., Richards, W.G., Staunton, J., Li, C., Monti, S., Vasa, P., Ladd, C., Beheshti, J., Bueno, R., Gillette, M. et al. (2001). Classification of human lung carcinomas by mRNA expression profiling reveals distinct adenocarcinoma subclasses. Proc Natl Acad Sci USA 98, 13790–13795. Bild, A.H., Yao, G., Chang, J.T., Wang, Q., Potti, A., Chasse, D., Joshi, M.B., Harpole, D., Lancaster, J.M., Berchuck, A. et al. (2006). Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439, 353–357. Bittner, M., Meltzer, P., Chen, Y., Jiang, Y., Seftor, E., Hendrix, M., Radmacher, M., Simon, R., Yakhini, Z., Ben-Dor, A. et al. (2000). Molecular classification of cutaneous malignant melanoma by gene expression profiling. Nature 406, 536–540. Blake, J. (2004). Bio-ontologies-fast and furious. Nat Biotechnol 22, 773–774. Bodenreider, O. (2004). The Unified Medical Language System (UMLS): Integrating biomedical terminology. Nucleic Acids Res 32(Database issue), D267–D270. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeckert, C., Aach, J., Ansorge, W., Ball, C.A., Causton, H.C. et al. (2001). Minimum information about a microarray experiment (MIAME)-toward standards for microarray data. Nat Genet 29, 365–371. Brown, M.P., Grundy, W.N., Lin, D., Cristianini, N., Sugnet, C.W., Furey, T.S., Ares, M, Jr. and Haussler, D. (2000). Knowledge-based analysis of microarray gene expression data by using support vector machines. Proc Natl Acad Sci USA 97, 262–267. Brunner, H.G. and van Driel, M.A. (2004). From syndrome families to functional genomics. Nat Rev Genet 5, 545–551. Butte, A. (2002). The use and analysis of microarray data. Nat Rev Drug Discov 1, 951–960.
222
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
Butte, A. and Kohane, I. (1999). Unsupervised knowledge discovery in medical databases using relevance networks. In Fall Symposium, American Medical Informatics Association (N. Lorenzi, ed.), Hanley and Belfus,Washington, DC, pp. 711–715. Butte, A.J. and Kohane, I.S. (2000). Mutual information relevance networks: Functional genomic clustering using pairwise entropy measurements. Pac Symp Biocomput, 418–429. Butte, A.J. and Kohane, I.S. (2006). Creation and implications of a phenome-genome network. Nat Biotechnol 24, 55–62. Butte, A.J., Tamayo, P., Slonim, D., Golub, T.R. and Kohane, I.S. (2000). Discovering functional relationships between RNA expression and chemotherapeutic susceptibility using relevance networks. Proc Natl Acad Sci USA 97, 12182–12186. Butte, A.J.,Ye, J., Niederfellner, G., Rett, K., Häring, H., White, M.F. and Kohane, I.S. (2001). Determining significant fold differences in gene expression analysis. Pac Symp Biocomput, 6–17. Butte, A.J., Bao, L., Reis, B.Y., Watkins, T.W. and Kohane, I.S. (2002). Comparing the similarity of time-series gene expression using signal processing metrics. J Biomed Inform 34, 396–405. Centers for Disease Control and Prevention (2004). Genetic testing for breast and ovarian cancer susceptibility: Evaluating direct-toconsumer marketing – Atlanta, Denver, Raleigh-Durham, and Seattle, 2003. MMWR Morb Mortal Wkly Rep 53, 603–606. Chang, H.Y., Chi, J.T., Dudoit, S., Bondre, C., van de Rijn, M., Botstein, D. and Brown, P.O. (2002). Diversity, topographic differentiation, and positional memory in human fibroblasts. Proc Natl Acad Sci USA 99, 12877–12882. Chee, M., Yang, R., Hubbell, E., Berno, A., Huang, X.C., Stern, D., Winkler, J., Lockhart, D.J., Morris, M.S. and Fodor, S.P. (1996). Accessing genetic information with high-density DNA arrays. Science 274, 610–614. Chiang, A.P., Nishimura, D., Searby, C., Elbedour, K., Carmi, R., Ferguson, A.L., Secrist, J., Braun, T., Casavant, T., Stone, E.M. et al. (2004). Comparative genomic analysis identifies an ADPribosylation factor-like gene as the cause of Bardet-Biedl syndrome (BBS3). Am J Hum Genet 75, 475–484. Cho, R.J., Fromont-Racine, M., Wodicka, L., Feierbach, B., Stearns, T., Legrain, P., Lockhart, D.J. and Davis, R.W. (1998). Parallel analysis of genetic selections using whole genome oligonucleotide arrays. Proc Natl Acad Sci USA 95, 3752–3757. Chow, M.L., Moler, E.J. and Mian, I.S. (2001). Identifying marker genes in transcription profiling data using a mixture of feature relevance experts. Physiol Genomics 5, 99–111. Chute, C.G. (2000). Clinical classification and terminology: Some history and current observations. J Am Med Inform Assoc 7, 298–303. Clayton, T.A., Lindon, J.C., Cloarec, O., Antti, H., Charuel, C., Hanton, G., Provost, J.P., Le Net, J.L., Baker, D., Walley, R.J. et al. (2006). Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature 440, 1073–1077. Crescenzi, M. and Giuliani, A. (2001). The main biological determinants of tumor line taxonomy elucidated by a principal component analysis of microarray data. FEBS Lett 507, 114–118. Dahlquist, K.D., Salomonis, N., Vranizan, K., Lawlor, S.C. and Conklin, B.R. (2002). GenMAPP, a new tool for viewing and analyzing microarray data on biological pathways. Nat Genet 31, 19–20. Debouck, C. and Goodfellow, P.N. (1999). DNA microarrays in drug discovery and development. Nat Genet 21, 48–50. DeRisi, J., Penland, L., Brown, P.O., Bittner, M.L., Meltzer, P.S., Ray, M., Chen, Y., Su, Y.A. and Trent, J.M. (1996). Use of a cDNA
microarray to analyse gene expression patterns in human cancer. Nat Genet 14, 457–460. Desiere, F., Deutsch, E.W., King, N.L., Nesvizhskii, A.I., Mallick, P., Eng, J., Chen, S., Eddes, J., Loevenich, S.N. and Aebersold, R. (2006). The PeptideAtlas project. Nucleic Acids Res 34, D655–D658. Deutsch, E.W., Eng, J.K., Zhang, H., King, N.L., Nesvizhskii, A.I., Lin, B., Lee, H., Yi, E.C., Ossola, R. and Aebersold, R. (2005). Human Plasma PeptideAtlas. Proteomics 5, 3497–3500. Dudoit, S., Yang, Y.H., Callow, M.J. and Speed, T.P. (2002). Statistical methods for identifying differentially expressed genes in replicated cDNA Microarray experiments. Statistica Sinica 12, 111–139. Dudoit, S., Gentleman, R.C. and Quackenbush, J. (2003a). Open source software for the analysis of microarray data. Biotechniques(Suppl), 45–51. Dudoit, S., Shaffer, J.P. and Boldrick, J.C. (2003b). Multiple hypothesis testing in microarray experiments. Statis. Sci. 18, 71–103. Edwards, J.S., Ibarra, R.U. and Palsson, B.O. (2001). In silico predictions of Escherichia coli metabolic capabilities are consistent with experimental data. Nat Biotechnol 19, 125–130. Eisen, M.B., Spellman, P.T., Brown, P.O. and Botstein, D. (1998). Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci USA 95, 14863–14868. English, S.B. and Butte, A.J. (2007). Evaluation and integration of 49 genome-wide experiments and the prediction of previously unknown obesity-related genes. Bioinformatics 23, 2910–2917. Espina, V., Mehta, A.I., Winters, M.E., Calvert, V., Wulfkuhle, J., Petricoin, E.F., 3rd, and Liotta, L.A. (2003). Protein microarrays: Molecular profiling technologies for clinical specimens. Proteomics 3, 2091–2100. Fiehn, O., Kopka, J., Dormann, P., Altmann, T., Trethewey, R.N. and Willmitzer, L. (2000). Metabolite profiling for plant functional genomics. Nat Biotechnol 18, 1157–1161. Fliri, A.F., Loging, W.T., Thadeio, P.F. and Volkmann, R.A. (2005). Analysis of drug-induced effect patterns to link structure and side effects of medicines. Nat Chem Biol 1, 389–397. Freimer, N. and Sabatti, C. (2003). The human phenome project. Nat Genet 34, 15–21. Friedman, N., Linial, M., Nachman, I. and Pe’er, D. (2000). Using Bayesian networks to analyze expression data. J Comput Biol 7, 601–620. Furey, T.S., Cristianini, N., Duffy, N., Bednarski, D.W., Schummer, M. and Haussler, D. (2000). Support vector machine classification and validation of cancer tissue samples using microarray expression data. Bioinformatics 16, 906–914. Gentleman, R.C., Carey, V.J., Bates, D.M., Bolstad, B., Dettling, M., Dudoit, S., Ellis, B., Gautier, L., Ge, Y., Gentry, J. et al. (2004). Bioconductor: Open software development for computational biology and bioinformatics. Genome Biol 5, R80. Ghaemmaghami, S., Huh, W.K., Bower, K., Howson, R.W., Belle, A., Dephoure, N., O’Shea, E.K. and Weissman, J.S. (2003). Global analysis of protein expression in yeast. Nature 425, 737–741. Golub, T.R., Slonim, D.K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J.P., Coller, H., Loh, M.L., Downing, J.R., Caligiuri, M.A. et al. (1999). Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring. Science 286, 531–537. Gygi, S.P., Rist, B., Gerber, S.A., Turecek, F., Gelb, M.H. and Aebersold, R. (1999). Quantitative analysis of complex protein mixtures using isotope-coded affinity tags. Nat Biotechnol 17, 994–999.
References
Henneman, L., Timmermans, D.R. and Van Der Wal, G. (2006). Public attitudes toward genetic testing: Perceived benefits and objections. Genet Test 10, 139–145. Hestvik, A.L., Hmama, Z. and Av-Gay, Y. (2003). Kinome analysis of host response to mycobacterial infection: A novel technique in proteomics. Infect Immun 71, 5514–5522. Hilsenbeck, S.G., Friedrichs, W.E., Schiff, R., O’Connell, P., Hansen, R.K., Osborne, C.K. and Fuqua, S.A. (1999). Statistical analysis of array expression data as applied to the problem of tamoxifen resistance. J Natl Cancer Inst 91, 453–459. ICD (2003). International Classification of Diseases: 9th revision, Clinical Modification (ICD-9-CM), 6th edition. Centers for Medicare & Medicaid Services,Washington DC. Jessani, N., Liu, Y., Humphrey, M. and Cravatt, B.F. (2002). Enzyme activity profiles of the secreted and membrane proteome that depict cancer cell invasiveness. Proc Natl Acad Sci USA 99, 10335–10340. Johnson, G.C., Esposito, L., Barratt, B.J., Smith, A.N., Heward, J., Di Genova, G., Ueda, H., Cordell, H.J., Eaves, I.A., Dudbridge, F. et al. (2001). Haplotype tagging for the identification of common disease genes. Nat Genet 29, 233–237. Jones, P., Cote, R.G., Martens, L., Quinn, A.F., Taylor, C.F., Derache, W., Hermjakob, H. and Apweiler, R. (2006). PRIDE: A public repository of protein and peptide identifications for the proteomics community. Nucleic Acids Res 34, D659–D663. Kanehisa, M. (1997). A database for post-genome analysis. Trends Genet 13, 375–376. Khoury, M.J., McCabe, L.L. and McCabe, E.R. (2003). Population screening in the age of genomic medicine. N Engl J Med 348, 50–58. Kohane, I.S., Kho, A.T. and Butte, A.J. (2002). Microarrays for an integrative genomics. MIT Press, Cambridge, Massachusetts. Kohane, I.S., Masys, D.R. and Altman, R.B. (2006). The incidentalome: A threat to genomic medicine. JAMA 296, 212–215. Lamb, J., Crawford, E.D., Peck, D., Modell, J.W., Blat, I.C., Wrobel, M.J., Lerner, J., Brunet, J.P., Subramanian, A., Ross, K.N. et al. (2006). The Connectivity Map: Using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929–1935. Lee, T.I., Rinaldi, N.J., Robert, F., Odom, D.T., Bar-Joseph, Z., Gerber, G.K., Hannett, N.M., Harbison, C.T., Thompson, C.M., Simon, I. et al. (2002). Transcriptional regulatory networks in Saccharomyces cerevisiae. Science 298, 799–804. Liang, S., Fuhrman, S. and Somogyi, R. (1998). Reveal, a general reverse engineering algorithm for inference of genetic network architectures. Pac Symp Biocomput, 18–29. Linnâe, C.V. and Schrèoder, J. (1763). Genera morborum, apud C. E. Steinert, Upsaliμ. Liotta, L.A., Espina, V., Mehta, A.I., Calvert, V., Rosenblatt, K., Geho, D., Munson, P.J., Young, L., Wulfkuhle, J. and Petricoin, E.F., 3rd (2003). Protein microarrays: Meeting analytical challenges for clinical applications. Cancer Cell 3, 317–325. Mootha, V.K., Lepage, P., Miller, K., Bunkenborg, J., Reich, M., Hjerrild, M., Delmonte, T., Villeneuve, A., Sladek, R., Xu, F. et al. (2003). Identification of a gene causing human cytochrome c oxidase deficiency by integrative genomics. Proc Natl Acad Sci USA 100, 605–610. Mulder, N.J., Apweiler, R., Attwood, T.K., Bairoch, A., Barrell, D., Bateman, A., Binns, D., Biswas, M., Bradley, P., Bork, P. et al. (2003). The InterPro Database, 2003 brings increased coverage and new features. Nucleic Acids Res 31, 315–318.
■
223
Mungall, C.J. (2004). Obol: Integrating language and Meaning in BioOntologies. Comp Funct Genom 5, 509–520. Nature (2002). Microarray standards at last. Nature 419, 323. Ng, S.K., Zhang, Z., Tan, S.H. and Lin, K. (2003). InterDom: A database of putative interacting protein domains for validating predicted protein interactions and complexes. Nucleic Acids Res 31, 251–254. Nguyen, D.V. and Rocke, D.M. (2002). Partial least squares proportional hazard regression for application to DNA microarray survival data. Bioinformatics 18, 1625–1632. Odom, D.T., Zizlsperger, N., Gordon, D.B., Bell, G.W., Rinaldi, N.J., Murray, H.L., Volkert, T.L., Schreiber, J., Rolfe, P.A., Gifford, D.K. et al. (2004). Control of pancreas and liver gene expression by HNF transcription factors. Science 303, 1378–1381. Park, P.J., Tian, L. and Kohane, I.S. (2002). Linking gene expression data with patient survival times using partial least squares. Bioinformatics 18(Suppl 1), S120–S127. Parkinson, H., Sarkans, U., Shojatalab, M., Abeygunawardena, N., Contrino, S., Coulson, R., Farne, A., Lara, G.G., Holloway, E., Kapushesky, M. et al. (2005). ArrayExpress – a public repository for microarray gene expression data at the EBI. Nucleic Acids Res 33, D553–D555. Parmigiani, G. (2003). The analysis of gene expression data : Methods and software. Springer, New York. Pavlidis, P. and Noble, W.S. (2001). Analysis of strain and regional variation in gene expression in mouse brain. Genome Biol 2. Perou, C.M. (2001). Show me the data! Nat Genet 29, 373. Petricoin, E.F., Ardekani, A.M., Hitt, B.A., Levine, P.J., Fusaro, V.A., Steinberg, S.M., Mills, G.B., Simone, C., Fishman, D.A., Kohn, E.C. et al. (2002). Use of proteomic patterns in serum to identify ovarian cancer. Lancet 359, 572–577. Potti, A., Mukherjee, S., Petersen, R., Dressman, H.K., Bild, A., Koontz, J., Kratzke, R., Watson, M.A., Kelley, M., Ginsburg, G.S. et al. (2006). A genomic strategy to refine prognosis in early-stage non-small-cell lung cancer. N Engl J Med 355, 570–580. Prince, J.T., Carlson, M.W., Wang, R., Lu, P. and Marcotte, E.M. (2004). The need for a public proteomics repository. Nat Biotechnol 22, 471–472. Quackenbush, J. (2004). Data standards for “omic” science. Nat Biotechnol 22, 613–614. Quackenbush, J. (2006). Microarray analysis and tumor classification. N Engl J Med 354, 2463–2472. Quinlan, J. (1992). C4.5: Programs for machine learning. Morgan Kaufmann, San Mateo, California. R Development Core Team (2004). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. Ramaswamy, S., Tamayo, P., Rifkin, R., Mukherjee, S., Yeang, C.H., Angelo, M., Ladd, C., Reich, M., Latulippe, E., Mesirov, J.P. et al. (2001). Multiclass cancer diagnosis using tumor gene expression signatures. Proc Natl Acad Sci USA 98, 15149–15154. Ramoni, M.F., Sebastiani, P. and Kohane, I.S. (2002). Cluster analysis of gene expression dynamics. Proc Natl Acad Sci USA 99, 9121–9126. Raychaudhuri, S., Stuart, J.M. and Altman, R.B. (2000). Principal components analysis to summarize microarray experiments: Application to sporulation time series. Pac Symp Biocomput, 455–466. Reich, M., Liefeld, T., Gould, J., Lerner, J., Tamayo, P. and Mesirov, J.P. (2006). GenePattern 2.0. Nat Genet 38, 500–501. Reis, B.Y., Butte, A.J. and Kohane, I.S. (2001). Extracting knowledge from dynamics in gene expression. J Biomed Inform 34, 15–27.
224
CHAPTER 17
■
Bioinformatic and Computational Analysis for Genomic Medicine
Ren, B., Robert, F., Wyrick, J.J., Aparicio, O., Jennings, E.G., Simon, I., Zeitlinger, J., Schreiber, J., Hannett, N., Kanin, E. et al. (2000). Genome-wide location and function of DNA binding proteins. Science 290, 2306–2309. Rhodes, D.R.,Yu, J., Shanker, K., Deshpande, N.,Varambally, R., Ghosh, D., Barrette, T., Pandey, A. and Chinnaiyan, A.M. (2004). Large-scale meta-analysis of cancer microarray data identifies common transcriptional profiles of neoplastic transformation and progression. Proc Natl Acad Sci USA 101, 9309–9314. Robinson, W.H., DiGennaro, C., Hueber, W., Haab, B.B., Kamachi, M., Dean, E.J., Fournel, S., Fong, D., Genovese, M.C., de Vegvar, H.E. et al. (2002). Autoantigen microarrays for multiplex characterization of autoantibody responses. Nat Med 8, 295–301. Roche, P.A. and Annas, G.J. (2001). Protecting genetic privacy. Nat Rev Genet 2, 392–396. Roche, P.A. and Annas, G.J. (2006). DNA testing, banking, and genetic privacy. N Engl J Med 355, 545–546. Ross, D.T., Scherf, U., Eisen, M.B., Perou, C.M., Rees, C., Spellman, P., Iyer, V., Jeffrey, S.S., Van de Rijn, M., Waltham, M. et al. (2000). Systematic variation in gene expression patterns in human cancer cell lines. Nat Genet 24, 227–235. Rumelhart, D. and McClelland, J. University of California SDPRG (1986). Parallel distributed processing: Explorations in the microstructure of cognition. MIT Press, Cambridge, Mass. Saeed,A.I., Bhagabati, N.K., Braisted, J.C., Liang,W., Sharov,V., Howe, E.A., Li, J.,Thiagarajan, M.,White, J.A. and Quackenbush, J. (2006). TM4 microarray software suite. Methods Enzymol 411, 134–193. Safran, M., Solomon, I., Shmueli, O., Lapidot, M., Shen-Orr, S., Adato, A., Ben-Dor, U., Esterman, N., Rosen, N., Peter, I. et al. (2002). GeneCards(TM) 2002: Towards a complete, objectoriented, human gene compendium. Bioinformatics 18, 1542–1543. Sarwal, M., Chua, M.S., Kambham, N., Hsieh, S.C., Satterwhite, T., Masek, M. and Salvatierra, O, Jr. (2003). Molecular heterogeneity in acute renal allograft rejection identified by DNA microarray profiling. N Engl J Med 349, 125–138. Schadt, E.E., Monks, S.A., Drake, T.A., Lusis, A.J., Che, N., Colinayo, V., Ruff , T.G., Milligan, S.B., Lamb, J.R., Cavet, G. et al. (2003a). Genetics of gene expression surveyed in maize, mouse and man. Nature 422, 297–302. Schadt, E.E., Monks, S.A. and Friend, S.H. (2003b). A new paradigm for drug discovery: Integrating clinical, genetic, genomic and molecular phenotype data to identify drug targets. Biochem Soc Trans 31, 437–443. Simon, I., Siegfried, Z., Ernst, J. and Bar-Joseph, Z. (2005). Combined static and dynamic analysis for determining the quality of timeseries expression profiles. Nat Biotechnol 23, 1503–1508. Singh, D., Febbo, P.G., Ross, K., Jackson, D.G., Manola, J., Ladd, C., Tamayo, P., Renshaw, A.A., D’Amico, A.V., Richie, J.P. et al. (2002). Gene expression correlates of clinical prostate cancer behavior. Cancer Cell 1, 203–209. Soldatova, L.N. and King, R.D. (2005). Are the current ontologies in biology good ontologies? Nat Biotechnol 23, 1095–1098. Sorlie, T., Perou, C.M., Tibshirani, R., Aas, T., Geisler, S., Johnsen, H., Hastie, T., Eisen, M.B., van de Rijn, M., Jeffrey, S.S. et al. (2001). Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci USA 98, 10869–10874. Spellman, P.T., Miller, M., Stewart, J.,Troup, C., Sarkans, U., Chervitz, S., Bernhart, D., Sherlock, G., Ball, C., Lepage, M. et al.
(2002). Design and implementation of microarray gene expression markup language (MAGE-ML). Genome Biol 3, RESEARCH0046. Stoll, M., Cowley, A.W, Jr., Tonellato, P.J., Greene, A.S., Kaldunski, M.L., Roman, R.J., Dumas, P., Schork, N.J., Wang, Z. and Jacob, H.J. (2001). A genomic-systems biology map for cardiovascular function. Science 294, 1723–1726. Storey, J.D. and Tibshirani, R. (2003). Statistical significance for genomewide studies. Proc Natl Acad Sci USA 100, 9440–9445. Su, A.I., Welsh, J.B., Sapinoso, L.M., Kern, S.G., Dimitrov, P., Lapp, H., Schultz, P.G., Powell, S.M., Moskaluk, C.A., Frierson, H.F, Jr. et al. (2001). Molecular classification of human carcinomas by use of gene expression signatures. Cancer Res 61, 7388–7393. Szallasi, Z. and Liang, S. (1998). Modeling the normal and neoplastic cell cycle with “realistic Boolean genetic networks”: Their application for understanding carcinogenesis and assessing therapeutic strategies. Pac Symp Biocomput, 66–76. Tamayo, P., Slonim, D., Mesirov, J., Zhu, Q., Kitareewan, S., Dmitrovsky, E., Lander, E.S. and Golub, T.R. (1999). Interpreting patterns of gene expression with self-organizing maps: Methods and application to hematopoietic differentiation. Proc Natl Acad Sci USA 96, 2907–2912. Toronen, P., Kolehmainen, M., Wong, G. and Castren, E. (1999). Analysis of gene expression data using self-organizing maps. FEBS Lett 451, 142–146. Tseng, Y.H., Butte, A.J., Kokkotou, E., Yechoor, V.K., Taniguchi, C.M., Kriauciunas, K.M., Cypess, A.M., Niinobe, M., Yoshikawa, K., Patti, M.E. et al. (2005). Prediction of preadipocyte differentiation by gene expression reveals role of insulin receptor substrates and necdin. Nat Cell Biol 7, 601–611. Tusher, V.G., Tibshirani, R. and Chu, G. (2001). Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA 98, 5116–5121. van de Rijn, M. and Gilks, C.B. (2004). Applications of microarrays to histopathology. Histopathology 44, 97–108. Wang, L., Wu, Q., Qiu, P., Mirza, A., McGuirk, M., Kirschmeier, P., Greene, J.R., Wang, Y., Pickett, C.B. and Liu, S. (2001). Analyses of p53 target genes in the human genome by bioinformatic and microarray approaches. J Biol Chem 276, 43604–43610. Weinmann, A.S., Yan, P.S., Oberley, M.J., Huang, T.H. and Farnham, P.J. (2002). Isolating human transcription factor targets by coupling chromatin immunoprecipitation and CpG island microarray analysis. Genes Dev 16, 235–244. Weinstein, J.N., Myers, T.G., O’Connor, P.M., Friend, S.H., Fornace, A.J, Jr., Kohn, K.W., Fojo,T., Bates, S.E., Rubinstein, L.V.,Anderson, N.L. et al. (1997). An information-intensive approach to the molecular pharmacology of cancer. Science 275, 343–349. Welsh, J.B., Sapinoso, L.M., Kern, S.G., Brown, D.A., Liu, T., Bauskin, A.R.,Ward, R.L., Hawkins, N.J., Quinn, D.I., Russell, P.J. et al. (2003). Large-scale delineation of secreted protein biomarkers overexpressed in cancer tissue and serum. Proc Natl Acad Sci USA 100, 3410–3415. Wen, X., Fuhrman, S., Michaels, G.S., Carr, D.B., Smith, S., Barker, J.L. and Somogyi, R. (1998). Large-scale temporal gene expression mapping of central nervous system development. Proc Natl Acad Sci USA 95, 334–339. Wheeler, D.L., Chappey, C., Lash, A.E., Leipe, D.D., Madden, T.L., Schuler, G.D., Tatusova, T.A. and Rapp, B.A. (2000). Database resources of the National Center for Biotechnology Information. Nucleic Acids Res 28, 10–14.
Recommended Resources
Wheeler, D.L., Church, D.M., Edgar, R., Federhen, S., Helmberg, W., Madden, T.L., Pontius, J.U., Schuler, G.D., Schriml, L.M., Sequeira, E. et al. (2004). Database resources of the National Center for Biotechnology Information: update. Nucleic Acids Res 32(Database issue), D35–D40. Wheeler, D.L., Barrett, T., Benson, D.A., Bryant, S.H., Canese, K., Chetvernin, V., Church, D.M., DiCuccio, M., Edgar, R.,
Websites http://www.ebi.ac.uk/arrayexpress/ http://www.bioconductor.org/ http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?dbgene http://www.ncbi.nlm.nih.gov/geo/ http://www.broad.mit.edu/cancer/software/genepattern/ http://www.genmapp.org/ http://www.who.int/classifications/icd/en/ http://www.genome.jp/kegg/ http://www.tm4.org/mev.html http://obo.sourceforge.net/ http://www.peptideatlas.org/ http://www.ebi.ac.uk/pride/ http://www.r-project.org/ http://genome-www5.stanford.edu/ http://umlsks.nlm.nih.gov/
Books Brown, T.A. (2006). Genomes 3, 3rd edition. Garland Science Publisher, New York.
Review articles Butte, A. (2002). The use and analysis of microarray data. Nat Rev Drug Discov 1, 951–960. Quackenbush, J. (2006). Microarray analysis and tumor classification. N Engl J Med 354, 2463–2472. Dudoit, S., Gentleman, R.C. and Quackenbush, J. (2003). Open source software for the analysis of microarray data. Biotechniques Suppl, 45–51.
225
Federhen, S. et al. (2006). Database resources of the National Center for Biotechnology Information. Nucleic Acids Res 34, D173–D180. World Health Organization (2005). International Statistical Classification of Diseases and Health Related Problems.Tenth Revision/Ed, Geneva. Wuensche, A. (1998). Genomic regulation modeled as a network with basins of attraction. Pac Symp Biocomput, 89–102.
RECOMMENDED RESOURCES ArrayExpress Bioconductor Entrez Gene Gene Expression Omnibus GenePattern GenMAPP International Classification of Diseases Kyoto Encyclopedia of Genes and Genomes MultiExperiment Viewer Open Biomedical Ontologies PeptideAtlas Proteomics Identifications Database R Stanford Microarray Database Unified Medical Language System
■
CHAPTER
18 Fundamentals and History of Informatics for Genomic and Personalized Medicine A. Jamie Cuticchia
INTRODUCTION Bioinformatics is the collection, classification, storage, and analysis of biochemical and biological information using computers especially as applied to molecular genetics and genomics. Bioinformatics arose out of necessity as part of the infrastructure and analytic needs of genome research (see below). Medical informatics refers to computer applications in medicine and health care and historically has focused on clinical data and patient records and on medical terminology (Shortliffe et al., 2003). This information is increasingly being stored and manipulated through electronic health records (see Chapter 19). Genomic and personalized medicine is leading to a unification of these disciplines, and an intersection of the two is the main topic of the current chapter. The success of the Human Genome Project (1990–2003) rested largely on the ability to collect, manage and share gene mapping and sequencing information (Collins and Galas, 1993; Watson, 1990; Watson and Cook-Deegan, 1991). As early as 1988, researchers recognized the need for the collection and sharing of molecular data. In that year, the US Congress created the National Center for Biotechnology Information (NCBI) as a division of the National Library of Medicine (NLM) (Benson et al., 1990;Wheeler et al., 2007). NCBI’s mission includes “creating automated systems for storing and analyzing knowledge Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 226
about molecular biology, biochemistry, and genetics.” The history of modern biological databases began with the creation of GenBank in 1985 (Burks et al., 1985). The goal of GenBank was to collect and distribute nucleotide sequence information and provide accession numbers, which are unique identifiers for each sequence record. Prior to GenBank, journals published articles describing nucleotide sequence data, as well as the data itself in printed form, but the latter became infeasible as sequencing strategies and technologies advanced and the data began to grow exponentially (www.ncbi.nlm.nih.gov/Genbank/index.html). Similarly, physical and genetic mapping data were described and reported through printed journal articles. Periodically, compendia of such data were published as a product of scientific conferences where geneticists gathered together to integrate the data and build increasingly comprehensive genetic maps. The publications that resulted from the Human Gene Mapping Conferences lead to the development of an online resource for genetic maps, and online access began to be provided by the Genome Database (GDB) (Pearson, 1991) in 1990. Genetic and phenotypic data for genes involved in human disease were also originally compiled in printed form in a series of books by Victor McKusick entitled Mendelian Inheritance in Man (MIM) spanning 12 editions, the final of which was published a decade ago (McKusick, 1998). MIM was also provided Copyright © 2009, Elsevier Inc. All rights reserved.
Databases for Genomic Medicine
(and still is) as Online Mendelian Inheritance in Man or OMIM (www.ncbi.nlm.nih.gov/sites/entrez?db=OMIM). The development and evolution of various physical markers for genes is beyond the scope of this chapter. Single nucleotide polymorphisms (SNPs) are variations that occur in DNA sequences from different members of a population (see Chapter 7). Most of these variants are “silent”, that is, have no known phenotypic effects but are useful as signposts to regions of the genome in which they occur. Currently, there are more than 31,000,000 SNPs in the human population with the information stored in the dbSNP repository at NCBI (Wheeler et al., 2006). The database records in dbSNP are relatively simple: they contain the SNP itself in the context of its flanking sequence as well as chromosome localization, frequency, and association with genes. The large number of SNPs and the ease of detection make it feasible to survey human populations for genetic variation (Germer et al., 2000; Lavebratt and Sengul, 2006; Norton et al., 2002; Tahira et al., 2003). Thus, the HapMap project was created whereby large geographically diverse populations were surveyed across millions of SNPs (International HapMap Consortium, 2003; Couzin, 2002; Thorisson et al., 2005). Sets of nearby SNPs on the same chromosome that are inherited as a unit are called haplotypes, and a haplotype map or HapMap of the human genome now contains in excess of 3.3 million genotypes derived from 270 individuals and serves as an indispensable tool for researchers (Thorisson et al., 2005). One goal of the HapMap project was to provide a blueprint of human variation with associated genetic markers in order to facilitate efficient genotyping of patient populations, ultimately for the use in gene-disease association studies. By providing data that can be used for study design, researchers can chose to use commercially available genotyping chips or employ technologies that focus on specific regions of the genome. Two major companies providing assays are Affymetrix and Illumina. Gene chips containing 500,000 SNP markers are commercially available from Affymetrix, while Illumina allows researchers to chose which markers are to be used in a study from their database of over 1,000,000 validated markers (see Chapter 8). The discovery of millions of SNPs, a survey of human variation across those SNPs, and the low cost of SNP detection in patient populations is a major step forward for personalized medicine. For example, the HapMap is used to discover and study the relationships between different genotypes and the way patients respond to particular drugs, a field of study known as pharmacogenomics (Deloukas and Bentley, 2004) (see Chapter 27). Thus, the foundation has been laid for examining patient genetic and phenotypic variation as a first step toward personalized medicine (Andrawiss, 2005).
DATABASES FOR GENOMIC MEDICINE In the past few years there has been increasing interest in the scientific community for the creation of personalized medicine databases. The US National Institutes of Health (NIH) established
■
227
ClinicalTrials.gov as a resource providing information on federally and privately supported clinical trials (Gillen et al., 2004; Zarin and Keselman, 2007). This is the first major effort to attempt to provide a central resource cataloging ongoing clinical research. The National Institute of Diabetes, Digestive, and Kidney Diseases (NIDDK) established the Central Repositories Program in 2003 with the goal of providing a catalog for data available from completed clinical studies, as well as reporting the availability of DNA and biospecimens from both completed and ongoing research. The NIDDK Data Repository provides web access to general information on studies and acts as a catalog of data (Cuticchia et al., 2006). Since the data in this repository include patient data, it is not possible to provide such data freely on the web. Rather, researchers interested in using data from a single study (or multiple studies for a meta-analysis) work with the NIDDK Data Repository staff to determine the availability and suitability of the data. If the researchers wish to use the data, they must submit a formal request along with the approval of their institutional review board (IRB) for the handling of patient data. The documents are then reviewed by the NIDDK, and, if data access is approved, the data are provided on CD-ROM. In 2001, PharmGKB: the Pharmacogenetics Knowledge Base was publicly released.The PharmGKB mission includes the development, implementation, dissemination of a public genotype– phenotype resource focused on pharmacogenomics (Hewett et al., 2002). PharmGKB is publicly available and supported by the NIH and is part of the NIH Pharmacogenetics Research Network (PGRN). Although it has been supported for nearly 6 years, the amount of data available through the resource is small. As of September 2007, PharmGKB has information on 608 variant genes, 523 drugs, and 123 phenotypes. In order to create an extensive resource, the issues of appropriate levels of funding and incentives to submit data must be addressed. It has been estimated that an initial investment of $50–100 million over 5–10 years would be needed to meet this goal (Gurwitz et al., 2006). The Genome Medicine Database of Japan (GeMDBJ) represents an initial attempt to warehouse data for the Millennium Genome Project (MGP) (Tsujimoto, 2001). In particular, the database contains polymorphism information, expression profiles, and limited clinical and demographic information for Alzheimer’s disease, diabetes, hypertension, and gastric cancer. While rudimentary in its search capabilities and presenting little more than a collection of files, it is nevertheless a result of a growing interest for the development of molecular medicine databases. Moreover, the Marshfield Clinic has undertaken database development as part of its Personalized Medicine Research Project (McCarty et al., 2005). They have created a database of genotypic and phenotypic information of over 18,000 participants and have produced analytical tools to support the project. This database is an example of the population-based study databases that must be created under the NIH Genome-Wide Association Studies (GWAS) Project. NIH has designated NCBI as the group for the development of databases to support GWAS. The Database of Genotype
228
CHAPTER 18
■
Fundamentals and History of Informatics for Genomic and Personalized Medicine
and Phenotype (dbGAP), already has data from studies on macular degeneration and Parkinson’s diseases. The database collects and disseminates data from medical sequencing, and molecular diagnostic assays in addition to genotype–phenotype correlations. Database funding, unfortunately, is one of the areas most easily decreased or eliminated during times of limited funding. However, even if funding could be readily and steadily obtained, the issue of incentives to submit data is a much more difficult challenge to meet. Even though the success of the HGP was largely based on the ability of numerous research groups to work together in the sharing of data, the ability to get groups to submit data publicly during the initial stages was difficult (Cuticchia et al., 1993). This difficulty arose because the laboratories participating in the HGP were hesitant to divert resources away from the research process in order to perform data submission. Even when HGP funding required proof of data release, some research groups interpreted that as being met by merely placing files (sometimes with no documentation) on the Internet. It was only when “peer-pressure,” combined with a recognition of the overall value to all research groups, that data submission became commonplace. For pharmacogenomics data, a great deal of which is obtained through privately funded research, the incentives for public database submission are not immediately apparent. Some have argued that financial incentives such as tax exemptions for the cost of depositing data or increasing the US patent lives for drugs when their clinical trial data are publicly accessible are needed (Gurwitz et al., 2006). It remains to be seen the manner in which public databases receive both data and funding; however, it is likely that such a resource will eventually be developed. Knowledge and Data Management As an example relevant to genomic and personalized medicine, the key to pharmacogenomic research is the ability to accurately collect and interrelate data. While there are numerous components used to determine the genotype at a particular locus, the data that would be used for a gene association study at its simplest form is shown in Table 18.1. The data can be broken down into just four columns corresponding to: Run_ID, Patient_ID, Marker_ID, and Genotype (A, B, or AB) The fields Patient_ID and Marker_ID are references to other datasets, which might include patient records or associated genes or SNPs. From a database perspective, those fields are “foreign keys,” meaning that the information contained in those fields can be used to directly link to other data. The data could be stored together along with additional data in the same database or could be used as links to external databases. In the case of linkage to other databases, diligence must be given to ensure that the identifier used to link the data remains static. As an example, if the Marker_ID field were used to link to dbSNP, then the field must be populated with RefSNP identifier (i.e., rs#), as in Table 18.1, of each marker in order to unambiguously link the two data records. Nearly all publicly accessible biological databases provide
TABLE 18.1
Sample patient genotypes
Run ID
Patient ID
Marker ID
Genotypea
101
1
rs28771969
A
102
2
rs28771969
A
103
3
rs28771969
B
104
4
rs28771969
AB
105
5
rs28771969
A
201
6
rs28771969
AB
202
7
rs28771969
B
203
8
rs28771969
A
204
9
rs28771969
B
205
10
rs28771969
AB
a
These labels represent the presence of alleles at the particular locus. The patient may be homozygous for either allele A or B, or may be heterozygous having one copy of each (AB).
unchanging (and usually purposely ambiguous) identifiers, accession numbers, for their records. If it is unclear which field the database is using for its accession number, then it is important to contact the database host. The field Patient_ID would similarly be linked to data records relevant to that patient. However, the implementation of this link would depend on the manner in which clinical data are made available. Most hospitals and many private medical practices have continued to migrate from the collection of medical information on paper toward its collection electronically (Ford et al., 2006; Geiger et al., 1995; Miller and West, 2007; Welch et al., 2007). Even where paper is still used for the initial recording of medical information, it is likely that the data will later be input into a computer. For clinical trials, the data that are to be collected are predetermined before the start of the trial. Case report forms are created that are used by the health care providers for the collection of necessary data. In many cases, due to the environment in which the data is to be collected, the data collection form is completed on paper. However, with the increase in computer access within clinics, it is becoming easier for the health care provider to complete electronic versions of the form using laptops, tablet computers, or PDA’s. Regardless of the mode of collection, ultimately the medical information needed for the clinical trial is stored electronically. It is incumbent upon both the database creator and user to make certain that the data collected is stored in a manner compatible with IRB and HIPAA standards (Erlen, 2005; Fain, 2003; Harrelson and Falletta, 2007). In some cases, the data may need to be imported directly from electronic medical records. In those cases it will be necessary to work with the administrators of the medical records in order to determine the manner of access. It is quite unlikely, nor is it recommended, that links will be made
Databases for Genomic Medicine
229
ensures that fields on different patient report forms that collect the same data type are related, but also facilitates the linking of clinical trial data across multiple trials (Murphy et al., 2003). There are two problems that must be avoided in relating data: (1) linking data that should not be linked and (2) failing to link identical data. As an example:
Electronic patient records EPR Honest broker EPR_ID RD_ID
■
Research database RD
Figure 18.1 Honest broker association between two databases. The “honest broker” is an independent database with the sole purpose of storing the association of identifiers for linked data stored in two or more databases. Thus, related data from two or more databases can only be retrieved when sufficient access privilege is granted.
directly to the electronic health records. Rather, it is preferable to have an extract of the data prepared by the medical records administrator, which may be directly loaded into the research database or stored as its own database. Those data are then linked by Patient_ID to the research database. The advantage to storing this information separately is that the medical records administrators can create their own database using database tools and protocols compatible to the electronic medical records. They also would not be burdened with the need to understand the research database. Moreover, if the data are stored in a separate database, changes can be made to the structure (or even a full replacement) of the research database without interfering with the medical records. Accessing medical records through an intermediate database also allows for the use of an “honest broker” approach to the data management (Boyd et al., 2007). In some cases patient data that are stored in the intermediate database may need to be accessed by users with different data privileges (Figure 18.1). Additionally, there may be a requirement that no medical records can be directly linked to a research database. By using an honest broker, the patient record identifier from both databases are linked by another database. Thus, there is no way (other than through the honest broker) for someone to link the patient information with the genotype data. Data Modeling While the storage and modeling of genotype data are simple, the modeling of patient data is not! For example, in a recent investigation of a very well thought out clinical trial, there were no less than 10 different places, names, and values in which the data value referred to patient death. The problem of relating identical fields across studies becomes even more difficult if a common naming scheme is not employed (see Chapter 17). The use of a common naming convention, a “controlled vocabulary” in the creation of clinical trial databases, not only
Data Record A: IV
Data Record B: 4
If the records referred to Drug Number in A, and Drug Number in B, then the records should be linked since Record A is the Roman Numeral representation of Record B. Additionally, if the records referred to manner of drug administration in A (intravenous), and manner of drug administration (where 4 intravenous) they should be linked. However, if Record A referred to manner of drug administration (intravenous) and Record B referred to Drug Number (Drug 4) then the records should not be linked. Thus, two pieces of data, which to a computer would be a character data type and a numeric data type, can in fact represent the same values. In databases that are properly designed with meaningful field names, it should be relatively simple to make the appropriate connections. However, enough ambiguity will exist in field names that it is always recommended to go back to that patient report forms and their explanation in order to make associations. A controlled vocabulary can be defined as a collection of consistent terms used to refer to identical objects, that is to say that they have been enumerated explicitly. There are numerous examples such as the Systemized Nomenclature of Medicine (SNOMED) (Kudla and Rallins, 1998) and the Clinical Data Interchange Standards Consortium (CDISC) (Kuchinke et al., 2006) for clinical data, and the Gene Ontology (also known as the GO Ontology, 2006; Camon et al., 2003) and NCI thesaurus (Sioutos et al., 2007) for more basic biological data. Both SNOMED and CDISC curate a list of medical terminology in such a way as to unambiguously connect terms which refer to the same “entity.” As an example, the entity is defined as “a person who is participating in a clinical study.” In one instance “Patient” is used to refer to that entity, and in another instance “Subject” is used. The GO Ontology is an attempt to create a hierarchical system for describing biology by breaking genes into functions and biological process (as an example). Together, these tools provide different means and terminology to map common concepts to a unique identifier. To a computer: ● ● ● ● ●
Myocardial infarction MI Infarction of heart Myocardial infarct Heart attack
all appear as different objects. By using a tool such as those available for SNOMED, one could either use a single definitive term replacing all synonyms or create a linking table that would
230
CHAPTER 18
TABLE 18.2 Patient ID 1
■
Fundamentals and History of Informatics for Genomic and Personalized Medicine
Simplified sample clinical report forms Patient name (Last)a Smith
2
Jones
3
Treatment
Response
External Link to Patient Records Run_ID
X
Yes
X
Yes
Williams
Placebo
No
4
Douglas
X
Yes
5
Baker
Placebo
No
6
Clark
X
No
7
Carey
Placebo
No
8
Sanders
Placebo
No
9
Hawkins
X
Yes
10
Thomas
Placebo
No
Patient_ID
Patient_ID
Patient_Name Table Link
Genotype
Treatment
Marker_ID
Response
External Link to dbSNP
a
Figure 18.2 Data representation. Run_ID and Patient_ID are the primary keys for each of the two tables. Here Patient_ID is also a foreign key for the first table which is used to link those records with matching records in the second table. Patient_ID and Marker_ID also serve as foreign keys to external databases.
Fictional
associate all the terms with the same object. By doing the latter, one could create a query such as “Show me all patients who had a MI since 2001” and the query would retrieve all records with the synonyms matched to MI. The use of controlled vocabulary can also be related to ontologies. Ontologies are ‘specifications of a relational vocabulary’ (Bard, 2007; Bittner and Donnelly, 2007; Pinciroli and Pisanelli, 2006; Soldatova and King, 2005). In other words, once one has created a set of common terms, the terms can be networked together. The use of ontologies would allow researchers to perform queries such as “Show me all patients having adverse events involving the heart since 2001.” Through the ontological network, this would retrieve MI and other adverse events (e.g., cardiac arrest). Software Development As an example, a database to support a very simple clinical trial is outlined. The trial will test the effect of Drug X on headaches. Five patients with headaches are given Drug X, while five patients receive a placebo. Additionally the patients are genotyped at a locus rs28771969 to determine whether variation in the gene has an effect on response. The information received from the Genotyping Center is shown the Table 18.1, while the data collected from the clinical report forms are listed in Table 18.2. The database would look something like the representation shown in Figure 18.2. Points to note are: 1. The Marker_ID field is used as an external identifier to link to other genetic databases. 2. The Patient_ID field is an internal identifier set by the research study and can be used to link to medical records. This information would be covered as Private Health Information (PHI) and will require appropriate IRB and HIPAA treatment.
3. X and Placebo are both terms that refer to same class of treatment. While this in an overly simplistic example, it demonstrates the key points regarding database construction for genomic and personalized medicine: 1. It provides for unambiguous linking of genotypes and patients. 2. It provides for linking to electronic medical records without the need to extract information directly from the medical records database. 3. It uses the term “Treatment” to refer to drug given; where treatment is a general term that can be mapped to other studies. 4. It uses keys to link together database tables and external databases.
CONCLUSION As the field of genomic and personalized medicine evolves, the need for the effective collection and dissemination of data will become increasingly important. While data initially are being collected for genotype–phenotype association studies, ultimately such research will lead to the use of genomic information in the health care process. For this reason, good database standards must be created in the research phase and used as models for the establishment of databases used for diagnostics. No new information technologies need to be invented in order to meet the needs of genomic medicine. Rather, researchers and informaticians need to concentrate on the use of “best-of-breed” information technology in the building of databases. By using good information technology practices, the needed databases should easily be constructed and once completed, issues regarding data collection and the incentives needed to populate the databases can then be addressed by the community.
References
■
231
REFERENCES Andrawiss, M. (2005). First phase of HapMap project already helping drug discovery. Nat Rev Drug Discov 4, 947. Bard, J. (2007). Systems developmental biology: The use of ontologies in annotating models and in identifying gene function within and across species. Mamm Genome. Benson, D., Boguski, M., Lipman, D. and Ostell, J. (1990). The National Center for Biotechnology Information. Genomics 6, 389–391. Bittner, T. and Donnelly, M. (2007). Logical properties of foundational relations in bio-ontologies. Artif Intell Med 39, 197–216. Boyd, A.D., Hosner, C., Hunscher, D.A., Athey, B.D., Clauw, D.J. and Green, L.A. (2007). An ‘Honest Broker’ mechanism to maintain privacy for patient care and academic medical research. Int J Med Inform 76, 407–411. Burks, C., Fickett, J.W., Goad, W.B., Kanehisa, M., Lewitter, F.I., Rindone, W.P., Swindell, C.D., Tung, C.S. and Bilofsky, H.S. (1985). The GenBank nucleic acid sequence database. Comput Appl Biosci 1, 225–233. Camon, E., Magrane, M., Barrell, D., Binns, D., Fleischmann,W.,Kersey, P., Mulder, N., Oinn, T., Maslen, J., Cox, A. et al. (2003). The Gene Ontology Annotation (GOA) project: Implementation of GO in SWISS-PROT, TrEMBL, and InterPro. Genome Res 13, 662–672. Collins, F. and Galas, D. (1993). A new five-year plan for the US Human Genome Project. Science 262, 43–46. Couzin, J. (2002). Human genome. HapMap launched with pledges of $100 million. Science 298, 941–942. Cuticchia, A.J., Chipperfield, M.A., Porter, C.J., Kearns, W. and Pearson, P.L. (1993). Managing all those bytes: The Human Genome Project. Science 262, 47–48. Cuticchia, A.J., Cooley, P.C., Hall, R.D. and Qin, Y. (2006). NIDDK data repository: A central collection of clinical trial data. BMC Med Inform Decis Mak 6, 19. Deloukas, P. and Bentley, D. (2004). The HapMap project and its application to genetic studies of drug response. Pharmacogenomics J 4, 88–90. Erlen, J.A. (2005). HIPAA–Implications for research. Orthop Nurs 24, 139–142. Fain, J.A. (2003). Research and the new HIPAA regulations. Diabetes Educ 29, 348–350. Ford, E.W., Menachemi, N. and Phillips, M.T. (2006). Predicting the adoption of electronic health records by physicians: When will health care be paperless?. J Am Med Inform Assoc 13, 106–112. Geiger, G., Merrilees, K., Walo, R., Gordon, D. and Kunov, H. (1995). An analysis of the paper-based health record: Information content and its implications for electronic patient records. Medinfo 8 (Pt 1), 295. The Gene Ontology (GO) project in (2006). . Nucleic Acids Res 34, D322–D326. Germer, S., Holland, M.J. and Higuchi, R. (2000). High-throughput SNP allele-frequency determination in pooled DNA samples by kinetic PCR. Genome Res 10, 258–266. Gillen, J.E., Tse, T., Ide, N.C. and McCray, A.T. (2004). Design, implementation and management of a web-based data entry system for ClinicalTrials.gov. Medinfo 11, 1466–1470. Gurwitz, D., Lunshof, J.E. and Altman, R.B. (2006). A call for the creation of personalized medicine databases. Nat Rev Drug Discov 5, 23–26. Harrelson, J.M. and Falletta, J.M. (2007). The privacy rule (HIPAA) as it relates to clinical research. Cancer Treat Res 132, 199–207.
Hewett, M., Oliver, D.E., Rubin, D.L., Easton, K.L., Stuart, J.M., Altman, R.B. and Klein, T.E. (2002). PharmGKB: The Pharmacogenetics Knowledge Base. Nucleic Acids Res 30, 163–165. International HapMap Consortium (2003). The International HapMap Project. Nature 426, 789–796. Kuchinke, W., Wiegelmann, S., Verplancke, P. and Ohmann, C. (2006). Extended cooperation in clinical studies through exchange of CDISC metadata between different study software solutions. Methods Inf Med 45, 441–446. Kudla, K.M. and Rallins, M.C. (1998). SNOMED: A controlled vocabulary for computer-based patient records. J AHIMA 69, 40–44. quiz 45–46. Lavebratt, C. and Sengul, S. (2006). Single nucleotide polymorphism (SNP) allele frequency estimation in DNA pools using Pyrosequencing. Nat Protoc 1, 2573–2582. McCarty, C., Wilke, R., Giampietro, P., Wesbrook, S. and Caldwell, M. (2005). Marshfield Clinic Personalized Medicine Research Project (PMRP): Design, methods and recruitment for a large populationbased biobank. Personalized Medicine 2, 49–79. McKusick,V.A. (1998). Mendelian Inheritance in Man; a Catalog of Human Genes and Genetic Disorders. The Johns Hopkins University Press, Baltimore, MD. ISBN 0-8018-5742-2 Miller, R.H. and West, C.E. (2007). The value of electronic health records in community health centers: Policy implications. Health Aff (Millwood) 26, 206–214. Murphy, L.S., Reinsch, S., Najm, W.I., Dickerson, V.M., Seffinger, M.A., Adams, A. and Mishra, S.I. (2003). Searching biomedical databases on complementary medicine: The use of controlled vocabulary among authors, indexers and investigators. BMC Complement Altern Med 3, 3. Norton, N., Williams, N.M., Williams, H.J., Spurlock, G., Kirov, G., Morris, D.W., Hoogendoorn, B., Owen, M.J. and O’Donovan, M.C. (2002). Universal, robust, highly quantitative SNP allele frequency measurement in DNA pools. Hum Genet 110, 471–478. Pearson, P.L. (1991). The genome data base (GDB) – a human gene mapping repository. Nucleic Acids Res 19(Suppl), 2237–2239. Pinciroli, F. and Pisanelli, D.M. (2006). The unexpected high practical value of medical ontologies. Comput Biol Med 36, 669–673. Shortliffe, E.H., Perreault, L.E., Wiederhold, G. and Fagan, L.M. (2003). Medical Informatics: Computer Applications in Health Care and Biomedicine. Springer, New York. Sioutos, N., de Coronado, S., Haber, M.W., Hartel, F.W., Shaiu, W.L. and Wright, L.W. (2007). NCI Thesaurus: A semantic model integrating cancer-related clinical and molecular information. J Biomed Inform 40, 30–43. Soldatova, L.N. and King, R.D. (2005). Are the current ontologies in biology good ontologies?. Nat Biotechnol 23, 1095–1098. Tahira, T., Suzuki, A., Kukita, Y. and Hayashi, K. (2003). SNP detection and allele frequency determination by SSCP. Methods Mol Biol 212, 37–46. Thorisson, G.A., Smith, A.V., Krishnan, L. and Stein, L.D. (2005). The International HapMap Project Web site. Genome Res 15, 1592–1593. Tsujimoto, G. (2001). [“Millennium Project” of MHLW]. Nippon Rinsho 59, 1884–1888. Watson, J.D. (1990). The human genome project: Past, present, and future. Science 248, 44–49.
232
CHAPTER 18
■
Fundamentals and History of Informatics for Genomic and Personalized Medicine
Watson, J.D. and Cook-Deegan, R.M. (1991). Origins of the Human Genome Project. FASEB J 5, 8–11. Welch, W.P., Bazarko, D., Ritten, K., Burgess, Y., Harmon, R. and Sandy, L.G. (2007). Electronic health records in four community physician practices: Impact on quality and cost of care. J Am Med Inform Assoc 14, 320–328. Wheeler, D.L., Barrett, T., Benson, D.A., Bryant, S.H., Canese, K., Chetvernin, V., Church, D.M., DiCuccio, M., Edgar, R., Federhen, S. et al. (2006). Database resources of the National
Center for Biotechnology Information. Nucleic Acids Res 34, D173–D180. Wheeler, D.L., Barrett, T., Benson, D.A., Bryant, S.H., Canese, K., Chetvernin, V., Church, D.M., DiCuccio, M., Edgar, R., Federhen, S. et al. (2007). Database resources of the National Center for Biotechnology Information. Nucleic Acids Res 35, D5–D12. Zarin, D.A. and Keselman, A. (2007). Registering a clinical trial in ClinicalTrials.gov. Chest 131, 909–912.
RECOMMENDED RESOURCES Human Genome Project www.genome.gov NCBI www.ncbi.nlm.nih.gov GenBank www.ncbi.nlm.nih.gov/Genbank/ Human Genome Organisation (HUGO) www.hugo-international.org GDB Human Genome Database www.gdb.org Online Mendelian Inheritance in Man www.ncbi.nlm.nih.gov/omim/ dbSNP www.ncbi.nlm.nih.gov/projects/SNP/
HAPMAP Project www.hapmap.org Clinical Trials Registry www.clinicaltrials.gov NIDDK Repository www.niddkrepository.org PharmGKB www.pharmgkb.org Genomic Medicine Database of Japan gemdbj.nibio.go.jp Marshfield Clinic Personalized Medicine Research Project www.marshfieldclinic.org/chg/pages/default.aspx?page=chg_pers_med_res_prj
CHAPTER
19 Electronic Medical Records in Genomic Medicine Practice and Research Glenn S. Gerhard, Robert D. Langer, David J. Carey and Walter F. Stewart
INTRODUCTION The implementation of electronic medical records (EMRs) is expected to substantially improve the quality and efficiency of health care and provide an important vehicle to advance patientcentered personalized care, the sine quo non of genomic medicine. In the United States, the use of EMRs in care delivery is expanding rapidly, especially among large integrated health delivery systems. The amount of clinically relevant genomic data and the number of resources devoted to research on genomic medicine are increasing in parallel. However, relatively few publications have yet addressed the use of EMRs in genomic medicine, although the need for such integration has been clearly established (Martin-Sanchez et al., 2004). In this chapter, the role of EMRs in the practice of genomic medicine and their use for genomic research will be considered. The need for digital approaches to storing, processing, and using information is driven by the growing density of genomic data, whether derived from gene expression profiling, single nucleotide polymorphism (SNP) genotyping, or DNA sequencing, and from the pace of discovery. Moreover, the complexity of clinical decision-making will change commensurately as genetic data relevant to diagnosis and therapeutic interventions increases. Historically, the disciplines of medical (or clinical) informatics and biological informatics (or bioinformatics) have both evolved to address the needs of genomic medicine
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
practice and research (Maojo and Kulikowski, 2003). Differences between the two communities have resulted in an attempt to integrate the approaches (Altman, 2000), with some proposing that health care organizations should be more proactive in collaborating with other public and private sector organizations in the development of an EMR containing both clinical and genomic data (Groen et al., 2005). These are laudable goals, but most of the work published thus far has come from those who have focused on the manipulation and analysis of genomic data for the clinical setting. However, the major factors driving the development of EMRs, such as quality measures, efficiency, and reimbursement, are not directly related to genomic medicine. This chapter is written from the perspective of those involved in the use of EMRs in large integrated delivery systems that make use of data and populations for research. While the US health system is fragmented, the large health systems that are adopting EMRs are becoming increasingly integrated, especially in adopting and implementing practice standards (Orlova et al., 2005). A significant window of opportunity thus exists to incorporate various aspects of genomic medicine into the development of these emerging systems. The data relevant to genomic medicine are highly diverse and applicable to essentially all medical specialties. As a result, multiple approaches and solutions will likely need to be developed for the myriad of needs in genomic medicine research and practice (Stein, 2003; Sujansky, 2001).
Copyright © 2009, Elsevier Inc. All rights reserved. 233
234
CHAPTER 19
■
Electronic Medical Records in Genomic Medicine Practice and Research
EMRs AND GENOMIC MEDICINE CLINICAL PRACTICE The EMR is a longitudinal electronic record of patient health information generated by encounters in any care delivery setting, encompassing patient demographics, progress notes, problem lists, medications, vital signs, past medical history, immunizations, laboratory data, and radiology reports. The EMR substitutes for traditional paper-based “charts”. The EMR has the ability to generate a complete record of a clinical patient encounter, as well as supporting other care-related activities directly or indirectly, such as evidence-based decision support, quality management, and outcomes reporting. It may also be part of a solution to addressing a decades old problem of decreasing the long lag time that exists before evidence-based medical knowledge is routinely used in clinical care (Stewart et al., 2007). This may be of tremendous importance for the rapid progress being made in genomic medicine. In this section, the integration and use of genomics data in EMRs for the clinical practice of genomic medicine is discussed. Example of an EMR The Geisinger Clinic, a large integrated health system in northeastern and central Pennsylvania, initiated installation of an EMR (i.e., EpicCare) in 1996, which was completed (i.e., completely paperless operations) in all community practice sites and specialty clinics by 2001. The EMR contains patient information from a variety of sources that is routinely integrated into a common database and includes age, sex, height, weight, and other demographics; lifestyle data (e.g., smoking, alcohol, etc.); clinical measures (e.g., BP, pulmonary function, cardiac information); digital imaging (MRI, CT, X-ray); all orders (i.e., labs, prescriptions, imaging, procedures) that require at least one clinical indication (i.e., ICD-9 code); clinical notes, which are increasingly created using smart-sets, or structured protocols; and laboratory measures (including results of genetic and molecular tests). Patients have on-line access to their EMR in the EpiCare system through the My Geisinger Internet portal. With My Geisinger patients can securely access their EMR, schedule visits, and email their doctor. Such web-based communications represent future opportunities for genomic medicine research. To date, the Geisinger clinic EMR database contains information on more than 2.2 million patients. Integration of Genomic Medicine Data into the EMR EMRs make use of relational database structures and utilities to access and display data to facilitate medical care and clinical decision-making. Although most observers and users consider the EMR to be a single application, in reality there are multiple data feeds from disparate sources that comprise the EMR. Some data sources, for example laboratory information systems, are usually integrated with the main EMR application, while others, such as radiology, histopathology, and molecular data are incompletely integrated or are resident in entirely separate databases. Genomic data currently used for clinical practice are
often generated by hospital, reference, or specialty laboratories that produce a wide variety of reporting formats, data types, and nomenclature (Ogino et al., 2007). Integration of these data into the EMR thus requires the ability to assemble the different data sources into an accessible format. Customized systems are sometimes developed in-house using on-site information technology expertise that are motivated by institution- and/or laboratoryspecific needs and allow for flexibility and control over how the software evolves. However, the tendency is to invest significant resources into these in-house solutions, which sometimes underestimate the effort needed and fail to meet the intended goals. The choice of integration architecture and format for data representation are thus significant decisions and have a number of important issues associated with them (Haas et al., 2001). The disparate sources that generate genomic data often report text-based data that are distributed through laboratory information systems into the EMR. These text results are commonly discrete, that is genotype A or genotype B, rather than continuous, although quantitative results are also produced, such as the number of trinucleotide repeats in the gene (FMR1) responsible for many cases of fragile X syndrome (Murray et al., 1997). Genomics results may also be reported in unique formats developed by diagnostic companies in niche markets. The disparate nature of the data and data sources further complicates the growing trend of integrating genomics data with other laboratory and/or clinically relevant data to generate a comprehensive clinical report. For example, cytogenetic and molecular remissions are recommended in reporting standards for acute myelogenous leukemia (Cheson et al., 2003). With the tendency of health care organizations towards the delivery of care in a service line structure and/or disease-specific care model (e.g., heart hospitals, neuroscience centers, etc.), more comprehensive and integrated reporting of genomic medicine results will likely be needed. This will apply to both genomic and other types of data-intensive results, such as proteomic and metabalomic data (Table 19.1). Currently most genomic data generated for clinical care are in the form of a discrete result. However, as the cost of full DNA sequencing continues to decline (Bennett et al., 2005), it is likely that discrete and highly relevant-specific results will be provided along with a large amount of DNA sequence data that has no known clinical significance (Mitchell and Mitchell, 2007). Ultimately, complete genome sequencing may become the most cost effective means of identifying sequence variants (Hutchison, 2007). It would not be unexpected in the future if DNA sequence determination became the standard of care for every patient, even though most of the sequence data may not be relevant to clinical decisions. Prior to that, many smaller iterations of DNA sequence data will be determined, for example, comprehensive mutation detection for BRCA1 and BRCA2 (Myriad Genetics, Salt Lake City, Utah, “BRAC Analysis for BRCA1 and BRCA2, Breast and Ovarian Cancer, Comprehensive Analysis”). In this analysis, full sequence determination in both forward and reverse directions of approximately 5400 base pairs comprising 22 coding exons and approximately 750 adjacent base pairs in the non-coding intervening sequences (introns) of the BRCA1
EMRs and Genomic Medicine Clinical Practice
T A B L E 1 9 . 1 Examples of genomic and other related data types potentially relevant for inclusion in the EMR ● ● ● ● ● ● ● ●
DNA sequence data Single nucleotide polymorphism (i.e., SNP) genotypes Multiple nucleotide polymorphism (e.g., insertion, deletion, repeat) genotypes Cell and/or tissue-specific microarray gene expression profiles Proteomic (e.g., mass spectroscopy) profiles Glycomic (i.e., carbohydrate) structural data Lipomic (i.e., lipid) profiles Metabolic phenotypes
gene is made. For BRCA2, full sequence determination in both forward and reverse directions of approximately 10,200 base pairs comprising 26 coding exons and approximately 900 adjacent base pairs in the non-coding intervening sequence (intron) is performed. Similar sequence analysis is offered for several other gene or genes that are associated with specific inherited cancer syndromes. Such testing generates an enormous amount of extra information that is not currently clinically relevant. However, as new discoveries are eventually translated into practice, ready access to relevant DNA data on patients may become necessary. How to maintain such dense sequencing data is thus an important decision (Mitchell and Mitchell, 2007). For example, placing every base sequenced into the patient’s medical record may clutter the record with predominantly clinically insignificant data. Alternatively, only bases that differ from reference data may need to be tracked. Reference data may change, which may alter the interpretation of patient data. The American College of Medical Genetics (ACMG) has previously recommended a minimum amount of information for inclusion in DNA sequencing reports (ACMG, 2000). Important considerations include reporting of all of the bases sequenced, as well as linked sequence reference data. Such recommendations are likely to evolve as more clinical experience is acquired with sequence data. Family History Data An integral component of genomic medicine is the application of family history to clinical care. Interest in collecting family history data as a routine part of care delivery is growing, as knowledge advances in linking family history of disease to patient risk. Many common diseases, such as coronary heart disease, diabetes, several types of cancer, osteoporosis, and asthma, are known to occur more frequently in some families than in others; thus the identification of at-risk individuals is important for high quality medical care. Family history data are also necessary for research aimed at identifying environmental and/or genetic factors that may underlie these common diseases (Hopper et al., (2005)). However, the acquisition of such data by most physicians is inadequate, in part because tools do not exist for efficient, reliable, and valid collection and analysis of such data (Guttmacher et al., 2004). Structured family history data are almost universally absent in current EMRs (see Chapter 42).
■
235
The need for the development of better family history tools has been highlighted by projects at the Centers for Disease Control (CDC) and by the US Surgeon General’s Family History Initiative (Wolpert, 2005). However, these efforts have not directly addressed the integration of tools into the real-world scenario of busy physicians and a multiplicity of health record systems, and do not provide an adequate breadth of data capture necessary for research. Various paper family history tools are available, as are various computer-based tools for drawing pedigrees (Trager et al., 2007). Several EMRs incorporate free text documentation of family history, while others allow for the development of family history templates and support checklists for taking and documenting family history. However, currently available tools are inadequate for many situations, especially primary care (Rich et al., 2004). The need for new tools is therefore apparent, and several characteristics of the ideal family history tool have been suggested, including patient-completed (e.g., paper, desktop, telephone, or web input), adapted to patient age, gender, ethnicity, common conditions, elicits-patient specific concerns, brief, understandable, easy to use, compatible with multiple clinical applications (e.g., paper, EMR, personal digital assistant), contains clinical decision support, and prioritizes based on clinical significance (Rich et al., 2004). No such electronic family history tools have yet been developed, despite the availability of suitable technologies. EMRs, Genomic Medicine, and Clinical Decision Support Genomic medicine is accelerating the growth of medical knowledge. In contrast, the means by which knowledge is translated into clinical practice have not evolved to keep up with the accelerating growth. Translation of knowledge into practice relies on a century-old tradition, the direct education of clinicians, an approach that was useful when knowledge was relatively limited. Today, it is not possible for most physicians to stay abreast of medical knowledge, especially genomics, and provide state-of-the-art care. The nuances in clinical decision-making in genomic medicine already render many care scenarios complex. Access to an EMR and the ongoing codification of medical knowledge (i.e., Clinical Practice Guidelines or CPGs) will be essential to addressing this growing translation gap. CPGs greatly facilitate, but are not sufficient for translating knowledge to practice. Almost 2000 active CPGs exist in the US National Guideline Clearinghouse (Kozma, 2006) and an individual CPG may encompass dozens to hundreds of clinical recommendations. These recommendations will rapidly expand in the era of genomic medicine, and thus codification of knowledge will be essential to increasing its access. EMRs offer a platform to translate codified knowledge into real-time actionable processes. Genomic data will need to be accessed with other patient data located in disparate locations within the EMR and evaluated in relation to a rule set. Real-time actionable recommendations will need to be created and be supported by an integrated and intuitive visual display of information, such as a set of orders or recommendations for the physician or other care provider.
236
CHAPTER 19
■
Electronic Medical Records in Genomic Medicine Practice and Research
An example of clinical decision-making using an EMR potentially relevant to genomic medicine is a process with a specific focus on cardiovascular risk management in primary care being developed at the Geisinger Health System. The process involves a decision support rules “engine”, external to the EMR, in which patient data such as labs, questionnaire data for phenotyping, and other clinical measures (Figure 19.1) are extracted in real time and evaluated in relation to clinical rules. The process can be used to generate a span of outputs including orders, clinical notes, and medication lists, etc. (Figure 19.2). The current model
involves rules for smoking cessation, control of LDL, control of blood pressure, and management of other cardiovascular risk factors. Together, the combination of rules for different risk factors translates into more than 100,000 possible ordersets. The process itself demonstrates that it is technically feasible to develop and implement a comprehensive set of rules for state-of-the-art management of a clinical condition. Genetic variants reported to predispose to coronary heart disease (Samani et al., 2007) may impact this process and could greatly increase the number of ordersets.
EMRs AND GENOMIC MEDICINE RESEARCH
Figure 19.1 Screenshot of EMR (Epic Systems, Corporation, Verona, WI). The software uses a tab-based design for access to medical record information including encounters, clinic notes, lab and imaging results, orders, medications, and procedures. These data can be extracted and used to automatically generate orders or recommendations, such as laboratory tests or imaging procedures for health maintenance or disease management.
Figure 19.2 Screenshot of EMR (Epic Systems, Corporation, Verona, WI). Examples of output from problem list, immunizations, allergies, medications, and health maintenance fields.
The EMR offers an exciting opportunity for genomic medicine research, both as a tool and as a data resource. It also represents an economically efficient means of obtaining phenotypic data and biosamples for generating genotypic data. Initial large-scale efforts to use genomic and clinical information for research, for example, deCode Genetics in Iceland, required the conversion of medical information from paper records into an electronic format (Hoffman, 2007). Even current large-scale efforts, such as the UK BioBank (Ollier et al., 2005), are based upon a mix of paper and electronic sources. The EMR thus represents a potentially large increase in efficiency for obtaining phenotypic data, and as described below, can also be an extremely efficient tool for patient recruitment and biosample acquisition. EMRs and Genome-Wide Association Studies The continued acceleration in genotyping technologies has enabled a dramatic increase in the use of genome-wide association studies (GWAS) to identify genetic variants underlying a variety of clinical phenotypes. However, the very large sample sizes and the logistics of collecting phenotype data and biological samples often require multi-center studies with coordinating centers and are very costly and time consuming to complete. Moreover, multiple replication studies are essential to verify initial GWAS findings. The long time period and substantial costs in initiating population-based studies limits the scope of such approaches. As an alternative, biorepositories are being developed to provide the resources necessary for GWAS. In Iceland (Hakonarson et al., 2003), the United Kingdom (Ollier et al., 2005), and elsewhere (Austin et al., 2003a), efforts are being made to link biobanking activities to national health records. In the United States, however, there are challenges with devising a national biobanking strategy (Austin et al., 2003b; Hsieh, 2004; McCarty et al., 2007; Sanner and Frazier, 2007). The US population is very heterogeneous, highly mobile, and health care is highly fragmented. Information on patients is highly dispersed and not in an easily accessible form. Hence, GWAS and related validation studies are largely conducted in conventional single project formats (i.e., de novo data collection), where there is a long lead-time between data acquisition and analysis.
EMRs and Genomic Medicine Research
In the United States, integrated delivery systems that use EMRs may be critical to addressing logistical challenges to conducting large-scale GWAS studies. While health care in the United States is highly fragmented, approximately 10% of care is provided by integrated delivery systems. Nationwide, these systems are undergoing a quiet revolution through the introduction of EMRs. Efforts to standardize data elements for research purposes and data sharing are being pursued by the larger systems (Hornbrook et al., 2005). There may be substantial opportunities to accelerate genomics research in the United States by leveraging the growing number of systems that are both adopting EMRs and creating biorepositories. However, there are many unanswered questions regarding whether such resources can be effectively used for genomics research. At the individual system level, there are questions regarding the quality, specificity, and completeness of data for phenotyping. In addition, little is known about system-level challenges that influence the quality and completeness of data captured during clinical encounters (e.g., impacts upon workflow, technical, and business priorities). Moreover, in some systems, even if quality data are available, there may be substantial technical, ethical, or policy barriers that limit access or use. Finally, GWAS and follow-up replication studies will require collaboration of multiple systems to achieve sample sizes that are sufficiently large, but there are enormous data quality, technical, and policy challenges to pooling data across institutions. Using the EMR to Access Populations for GWAS Case-control studies and family based studies have been the two primary designs for GWAS, but the economic efficiencies of the case-control format conducted within health systems are significant. Patient populations and inherent selection bias also impact GWAS, which may vary substantially depending on the source (e.g., primary care, specialty care, outpatient, inpatient). Primary care populations tend to be similar to general population samples with the exception that younger individuals (especially males) and individuals from lower middle-income strata (i.e., not eligible for Medicaid and no access to health care benefits through work) are under-represented. Selection bias inherent to specialty care samples depends on the extent to which there are competing providers in the same market and whether a provider has local, regional, national, or international reach. Selection bias increases directly in relation to market reach. These potential biases will be reflected in clinical data available in EMRs, depending upon where and when the system was installed. Genetic heterogeneity in a population may also arise due to population structure and recent admixture and may confound the results of genetic association studies in unrelated individuals, leading to a potential excess of both false positive and false negative results (Ziv and Burchard, 2003). The degree of confounding is dependent in part on how genetically different the subgroups are within the population. For example, among many contemporary populations, recent admixture between two or more genetically distinct subgroups has added to the genetic complexity. In the United States, this complexity is particularly
■
237
pertinent for admixed populations such as Latino- and AfricanAmericans (Parra et al., 1998). Within indigenous continental population groups, European subpopulations appear to be the most similar to each other (Bowcock et al., 1994; Mountain and Cavalli-Sforza, 1997). The ability of an EMR to distinguish among ethnicities within its patient base may impact the usability of the data for GWAS. The Geisinger MyCode Biorepository The creation of biorepositories that are closely linked with EMRs may be an extremely efficient approach to genomic medicine research, especially for GWAS that require DNA samples from large populations with corresponding robust phenotypic data. The challenges to establishing such biorepositories involve complex institutional, legal, and social issues that also must be addressed. However, the EMR opens new opportunities to address these concerns, such as patients’ safety, rights, informed consent, and privacy. The Geisinger Clinic MyCode biorepository project is perhaps the first large-scale biobanking initiative built around an EMR. In this project, the EMR is used as a tool for identifying and recruiting patients, obtaining blood samples, and retrieving phenotype data. The primary goal of the MyCode project is to bank blood/serum/DNA samples from patients on a large-scale coupled with access to their EMR data. MyCode was initiated in 2007 to leverage the resources of the Geisinger Health System, an integrated delivery system offering healthcare services to residents of 31 of Pennsylvania’s 67 counties, with a significant presence in central and northeastern Pennsylvania between Pittsburgh and Philadelphia. Census data indicate that in most counties, the out-migration rate is less than 1% per year, thus providing an extremely attractive population for longitudinal studies. The population within the Geisinger Clinic catchment area is relatively homogeneous. Based on available historical, demographic, and self-reported ethnicity data, the Geisinger population is almost entirely Caucasian of mixed European ancestry, with only small contributions from other continental population groups, such as African, Asian, or Latino. The relatively lower degree of genetic variation among European subpopulations, and the length of time for intermingling of European subgroups in Pennsylvania, also serves to minimize population stratification. Such a population straddles genetically homogeneous populations such as Iceland or Finland with strong founder effects, and admixed populations, now present in many urban areas. The MyCode patient recruitment process (Figure 19.3) begins with a search of the EMR for patients scheduled to visit an outpatient or designated specialty clinic in which a consenter will be stationed. Eligible patients are identified through a daily EMR report, while consent status (e.g., consent/did not consent/ not interested) is verified in the patient’s EMR to prevent unnecessary requests for participation at future appointments. Patients are invited to participate in MyCode through direct interaction with a project consenter. After consent, a standing order for the additional MyCode blood draw is placed in
238
CHAPTER 19
■
Electronic Medical Records in Genomic Medicine Practice and Research
Use of EMR in MyCode Biorepository Project
Obtain Clinic Schedule from EMR
Generate Eligibility List from EMR
Patient Consent
Questionnaire Data into EMR
Blood Draw Order Placed in EMR
Frozen Storage
Blood Samples Processed
Blood Drawn
EMR Blood Draw Order Read by Phlebotomy
Patient to Phlebotomy
Figure 19.3 Use of the EMR in the MyCode Biorepository Project. The EMR functions as a tool at multiple points in the patient recruitment and consenting processes, and for acquisition of blood samples. The EMR also serves as the repository for phenotype data.
the EMR such that it is appended to the new order for blood work ordered by a physician at a future visit. With this strategy, blood is not specifically drawn for research which has resulted in high patient participation rates, as has been found in other studies (Sanner and Frazier, 2007). The sample collection process is highly efficient because it leverages the EMR (i.e., maintaining a standing order with an automatic protocol to activate the order), the laboratory infrastructure (e.g., standard protocols for phlebotomy, data systems, transportation of blood samples to a central location, etc.), and a passive low cost means of obtaining samples. Blood samples are transported from clinics across the Geisinger coverage region using the existing Geisinger Lab courier service to a central research laboratory that processes the samples for the MyCode biorepository. The EMR serves as the electronic enabler of the entire MyCode process. The approach to creating the MyCode biorepository mitigates several methodological challenges with GWAS, including possible genotype errors, phenotype misclassification, and selection bias. While genotyping platforms are robust, DNA quality can affect genotyping results. Frozen blood and/or DNA are collected using the same protocol and processed using the same methodology. Errors in labeling patient samples are minimal, since an integrated clinical lab quality protocol is used for tracking samples from the point of phlebotomy to freezer storage. Phenotype misclassification is minimized through the use of careful case and control definition. Selection bias is minimized through the selection of cases and controls from the same population base. Example of EMR Data Extraction The EMR is a powerful and convenient source of phenotypic data for GWAS and genomic medicine research. A reference
dataset has been created with a record for each patient in the Geisinger EMR population that includes demographics, physical measurements, personal medical history, health behaviors, laboratory tests, surgical and medical procedures, and medications prescribed. The dataset was developed to reflect the top 80% of elements most likely to be used for standard research purposes. Core data were extracted from the EMR using imbedded routines (known as Clarity), provided by the software vendor. Clarity tables can be manipulated using standard queries in SQL applications. The Clarity tables containing the specific elements in the data dictionary for a particular domain (e.g., physical measurements, or endocrine-related lab) are identified, and those fields are extracted into a text file. Typically, all instances of an element are pulled (e.g., the dataset contains all blood pressure measurements, or all glucose lab for a patient), not just the most recent or the first. The text file is read into SAS/STAT software (SAS Institute Inc., Cary, NC) and mapped to pre-defined fields. The resulting files are documented, cleaned and stored on a dedicated directory controlled by the biostatistics team. Analysis files are created by merging the appropriate data and applying any algorithms necessary for derived values. As an example, this extraction of core data from the EMR can be used to obtain body mass index (BMI) data on over 270,000 adults (Table 19.2). Analyses were limited to adult patients 20 years of age and older. The mean and median BMIs were 28.4 and 27.2, respectively. Slightly over 30% of patients had a BMI 25 while 7.8% had a BMI of 40+. Patients with a BMI of 40+ were three times more likely to be female and more likely to be between 40 and 59 years of age (i.e., with 46.2% versus 29.6%) compared to those with a BMI 25. The relatively smaller proportion above age 60 and the gender bias may suggest a survival effect for this subgroup.
EMRs and Genomic Medicine Research
TABLE 19.2
■
239
Percent of adult patients by BMI and specific phenotypes
Phenotype
BMI Category ≤25 n ⴝ 80,861
>25, <30 n ⴝ 85,626
>30, <40 n ⴝ 83,298
40ⴙ n ⴝ 21,187
Male
31.2
46.2
42.2
27.9
Female
68.8
53.8
57.8
72.1
Age 40
45.4
28.2
25.1
30.2
Age 40–59
29.6
36.5
40.7
46.2
Age 60+
25.0
35.3
34.2
23.6
Diagnosis of HTN
39.5
59.9
72.2
78.5
Diagnosis of LoHDL
18.8
28.2
39.8
47.8
7.8
12.8
21.1
32.1
13.3
23.4
34.4
42.5
Diagnosis of HTN and T2D
6.2
11.1
19.0
29.0
Diagnosis of LoHDL and T2D
5.3
9.0
15.3
23.6
Diagnosis of HTN, LoHDL, and T2D
4.9
8.3
14.4
22.2
35.4
20.2
11.0
6.1
Diagnosis of T2D Diagnosis of HTN and LoHDL
No Diagnosis of HTN, LoHDL, or T2D
The prevalence of Type II diabetes (T2D), hypertension (HTN), and low HDL levels (LoHDL), and combinations of these, for each BMI subgroup was also determined. A patient was defined as having T2D or HTN if the condition was on the EMR Problem List (i.e., ICD-9 linked searchable fields) or if they had two encounters within 24 months where the condition(s) were listed as a reason for visit. HTN was also designated if there was any record of diastolic blood pressure of 90 mm Hg or more, or a systolic blood pressure of 140 mm Hg or more, or if the patient was taking an antihypertensive medication. T2D was designated if any of the following conditions were true: a random blood sugar of 200 or more, a hemoglobin A1C of 7% or higher, or if there was a prescription for insulin or an oral hypoglycemic medication. LoHDL was designated if the patient was male and had an HDL 35 or was female with an HDL 45 mg/dl. In general, prevalence of HTN, LoHDL, and T2D increase substantially in relation to BMI category. A diagnosis of all three conditions was found for 22.2% of those with a BMI of 40+, more than four times higher than the prevalence among those with normal BMI. In contrast, fully 35.4% of individuals with a BMI 25 had one of these conditions (as opposed to only 6.1% of patients with a BMI of 40+). Data Warehouses In addition to EMRs, data warehouses, which refer to the assemblage of diverse data sources into a single point of reference accessible using a generalized global architecture (Louie et al., 2007), are an instrument for genomic medicine research. Data warehouses allow for optimizing performance and maintaining local control and privacy aspects of data handling, all of which
are important aspects for clinical genetics databases (Birch and Friedman, 2004), biobanks, or clinical trials databases. Data warehouses are built to be reliable and relatively robust, and to have fast response times for queries. Performance is a key issue with such databases (Haas et al., 2001), since the volume of data to be built into a data warehouse can provide some formidable challenges for initial construction as well as continued maintenance. Another significant challenge is to develop global data structures that allow for data diversity. Nuances of the specific data sources of data may be lost or homogenized, or analysis structures may become quite complex. At the Geisinger Clinic, a data warehouse project in collaboration with IBM was initiated in 2006, the Clinical Decision Information System (CDIS) project. The goal of this project is to convert all data relevant to clinical practice into an enterprise data warehouse (EDW), to be completed in late 2008. In its first iteration, the EDW will incorporate clinical, financial, administrative, operational, claims, and patient survey data. A total of 40 different source systems including laboratory, pathology, pharmacy, scheduling, billing, cost accounting, claims (inpatient, specialty, pharmacy, and ambulatory), questionnaires, operating room, radiology, and patient safety and care management will feed data to the CDIS, complemented by hundreds of look up tables for input databases (e.g., ICD-9, CPT, etc.). In anticipation of its growing importance, a research module for genomics data is also in development. Data Federations and Multi-System Studies Unlike a data warehouse, a database federation does not incorporate data into a single source but leaves it housed locally. Constituent databases therefore remain under local control but
240
CHAPTER 19
■
Electronic Medical Records in Genomic Medicine Practice and Research
are distributed across a network. Such federations are easy to update since each constituent is responsible for its own updates. However, performance can be limited by the query load capacities of individual constituent members of the federation. A data federation has been initiated by the HMO Research Network (HMORN). The network is a consortium of 15 research centers, each affiliated with a non-profit integrated health care delivery system, all of which have or are developing ambulatory care EMR systems. The HMORN is managed by a board of directors, comprised of directors of the respective research centers. In addition to the development of best practices for research administration for multi-site collaborations, the HMORN has initiated efforts to establish a Virtual Data Warehouse (VDW), to simplify data sharing among network participants. Although formally a data federation, the VDW is “virtual” because data are only stored locally within each research center, not in a central warehouse. Linkages across data systems are established through a set of files created to map relevant data elements within each health systems databases to standardized variable definitions, names, and codes. These common VDW standards allow investigators and data analysts at one site to develop SAS programming code to be used by participating
sites and run against a local health system database with minimal site-specific effort. Such VDWs may ultimately play a role in the conduct of GWAS, since phenotypic information from very large sample populations may be obtained. Corresponding biorepository efforts, such as the Geisinger MyCode project, will facilitate the use of VDWs for genomic medicine research.
CONCLUSION The use of the EMR for either genomic medicine practice or research is in its early stages. To fully exploit the power and potential of EMRs for genomic and personalized medicine, efforts such as those described in this chapter will need to be expanded. The continued implementation of EMRs throughout the United States, especially in large health systems, sets the stage for these efforts.
ACKNOWLEDGEMENTS The authors wish to thank the Geisinger Clinic and the Pennsylvania State Department of Community and Economic Development for support of the MyCode project.
REFERENCES ACMB Laboratory Practice Committee Working Group (2000). ACMG recommendations for Standards for interpretation of Sequence Variations. Genet Med 2(5), 302–303. Altman, R.B. (2000). The interactions between clinical informatics and bioinformatics: A case study. J Am Med Informat Assoc 7(5), 439–443. Austin, M.A., Harding, S. et al. (2003a). Genebanks: A comparison of eight proposed international genetic databases. Community Genet 6(1), 37–45. Austin, M.A., Harding, S.E. et al. (2003b). Monitoring ethical, legal, and social issues in developing population genetic databases. Genet Med 5(6), 451–457. Bennett, S.T., Barnes, C. et al. (2005). Toward the 1,000 dollars human genome. Pharmacogenomics 6(4), 373–382. Birch, P. and Friedman, J.M. (2004). Utility and limitations of genetic disease databases in clinical genetics research: A neurofibromatosis 1 database example. Am J Med Genet 125(1), 42–49. Bowcock, A.M., Ruiz-Linares, A. et al. (1994). High resolution of human evolutionary trees with polymorphic microsatellites. Nature 368(6470), 455–457. Cheson, B.D., Bennett, J.M. et al. (2003). Revised Recommendations of the International Working Group for Diagnosis, Standardization of Response Criteria, Treatment Outcomes, and Reporting Standards for Therapeutic Trials in Acute Myeloid Leukemia. J Clin Oncol 21(24), 4642–4649. Groen, P., Wine, M. et al. (2005). Genomic Information Systems and Electronic Health Records (EHR). Virtual Medical Worlds Monthly. (October). Guttmacher, A.E., Collins, F.S. et al. (2004). The family history––more important than ever. N Engl J Med 351(22), 2333–2336. Haas, L.M., Schwarz, P.M. et al. (2001). DiscoveryLink: A system for integrated access to life sciences data sources. IBM Syst J 40(2). Hakonarson, H., Gulcher, J.R. et al. (2003). deCODE genetics, Inc. Pharmacogenomics 4(2), 209–215.
Hoffman, M.A. (2007). The genome-enabled electronic medical record. J Biomed Informat 40(1), 44–46. Hopper, J.L., Bishop, D.T. et al. (2005). Population-based family studies in genetic epidemiology. Lancet 366(9494), 1397–1406. Hornbrook, M.C., Hart, G. et al. (2005). Building a virtual cancer research organization. J Natl Cancer Inst (35), 12–25. Hsieh, A. (2004). A nation’s genes for a cure to cancer: Evolving ethical, social and legal issues regarding population genetic databases. Columbia J Law Soc Probl 37(3), 359–411. Hutchison, C.A., III (2007). DNA sequencing: Bench to bedside and beyond. Nucleic Acids Res 35: 6227–6237. Kozma, C.M. (2006). The National Guideline Clearinghouse. Manag Care Interface 19(5). 43, 51. Louie, B., Mork, P. et al. (2007). Data integration and genomic medicine. J Biomed informat 40(1), 5–16. Maojo, V. and Kulikowski, C.A. (2003). Bioinformatics and medical informatics: Collaborations on the road to genomic medicine?. J Am Med Informat Assoc 10(6), 515–522. Martin-Sanchez, F., Iakovidis, I. et al. (2004). Synergy between medical informatics and bioinformatics: facilitating genomic medicine for future health care. J Biomedl Informat 37(1), 30–42. McCarty, C.A., Nair, A. et al. (2007). Informed consent and subject motivation to participate in a large, population-based genomics study: The Marshfield Clinic Personalized Medicine Research Project. Community Genet 10(1), 2–9. Mitchell, D.R. and Mitchell, J.A. (2007). Status of clinical gene sequencing data reporting and associated risks for information loss. J Biomed Informat 40(1), 47–54. Mountain, J.L. and Cavalli-Sforza, L.L. (1997). Multilocus genotypes, a tree of individuals, and human evolutionary history. Am J Human Genet 61(3), 705–718. Murray, J., Cuckle, H. et al. (1997). Screening for fragile X syndrome. Health Technol Assess 1(4), 1–71, i-iv.
Recommended Resources
Ogino, S., Gulley, M.L. et al. (2007). Standard mutation nomenclature in molecular diagnostics: Practical and educational challenges. J Mol Diagn 9(1), 1–6. Ollier, W., Sprosen, T. et al. (2005). UK Biobank: From concept to reality. Pharmacogenomics 6(6), 639–646. Orlova, A.O., Dunnagan, M. et al. (2005). Electronic health recordpublic health (EHR-PH) system prototype for interoperability in 21st century healthcare systems. AMIA Annu Symp Proc, 575–579. Parra, E.J., Marcini, A. et al. (1998). Estimating African American admixture proportions by use of population-specific alleles. Am J Hum Genet 63(6), 1839–1851. Rich, E.C., Burke, W. et al. (2004). Reconsidering the family history in primary care. J Gen Intern Med 19(3), 273–280. Samani, N.J., Erdmann, J. et al. (2007). Genomewide association analysis of coronary artery disease. N Engl J Med 357(5), 443–453.
RECOMMENDED RESOURCES James M. Walker (Editor), Eric J. Bieber (Editor), Frank Richards (Editor), Implementing an Electronic Health Record System (Health Informatics) Publisher: Springer 2004. Implementing an Electronic Health Record System addresses the range of issues and opportunities that implementing an electronic health records system (EHR) poses for any size of medical organizationfrom the small one-man operation to a large healthcare system. The book is divided into sections on preparation, support, implementation and a summary and prospects section, enabling the clinician to define the framework necessary to implement and evaluate a clinically effective EHR system. With the increasing involvement of clinicians in the day-to-day running of the practice, interest is now focused on EHR as a key area for improving clinical efficiency. This book uniquely provides the guidance a clinical team needs to plan and execute an effective EHR system within any clinical setting. URL: http://www.hmoresearchnetwork.org/
■
241
Sanner, J.E. and Frazier, L. (2007). Factors that influence characteristics of genetic biobanks. J Nurs Scholarsh 39(1), 25–29. Stein, L.D. (2003). Integrating biological databases. Nat Rev Genet 4(5), 337–345. Stewart, W.F., Shah, N.R. et al. (2007). Bridging the inferential gap: The electronic health record and clinical evidence. Health Aff (Millwood) 26(2), w181–w191. Sujansky, W. (2001). Heterogeneous database integration in biomedicine. J Biomed Informat 34(4), 285–298. Trager, E.H., Khanna, R. et al. (2007). Madeline 2.0 PDE: A new program for local and web-based pedigree drawing. Bioinformatics 23(14), 1854–1856. Wolpert, C.M. (2005). Surgeon General launches new public health campaign: The family history initiative. JAAPA 18(1), 20–22. Ziv, E. and Burchard, E.G. (2003). Human population structure and genetic association studies. Pharmacogenomics 4(4), 431–441.
CHAPTER
20 Clinical Decision Support in Genomic and Personalized Medicine Kensaku Kawamoto and David F. Lobach
INTRODUCTION Genomic medicine, in which individuals’ health is optimized through the use of genomic patient data (Willard et al., 2005), is a critical component of personalized medicine, in which individuals’ health is optimized through the use of all available clinical, molecular, and genetic patient data (Abrahams et al., 2005). Current genomic medicine interventions include the use of a monoclonal antibody to the HER2 receptor to treat HER2-positive breast cancers (Piccart-Gebhart et al., 2005); the use of a gene expression assay to guide breast cancer therapy (Paik et al., 2006); and the genotyping of individuals’ HIV infections to guide antiretroviral therapy (Blum et al., 2005). The potential benefits of genomic medicine include disease prevention, earlier disease detection, more effective disease treatment, and reduced overall health care costs (The Personalized Medicine Coalition, 2006). Given this enormous promise, the Secretary of the US Department of Health and Human Services (HHS) in 2007 identified personalized medicine as one of the top priorities of the Department (United States Department of Health and Human Services Press Office, 2007). This chapter has been written with a particular focus on the US health care system. However, many of the issues and challenges discussed in this chapter are also applicable to the health care systems of other industrialized nations. Despite the tremendous promise of personalized medicine, past experience would predict that new genomic interventions, like any new medical intervention, will remain significantly Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 242
under-utilized for some time without the concurrent introduction of supportive technologies. Patients in the United States only receive about half of recommended care relative to established care standards (McGlynn et al., 2003), and many medical interventions remain significantly under-utilized decades after their efficacy was established in landmark clinical trials (National Committee for Quality Assurance, 2006). Moreover, genomic interventions may face even greater barriers to clinical adoption compared to more traditional medical interventions, due to such factors as limited clinician familiarity with genomics and the volume and complexity of the data and patient management algorithms that need to be considered. In recognizing this challenge, the Secretary of HHS announced in March 2007 that supporting the achievement of personalized medicine through health information technology (IT) is a top priority of his department (United States Department of Health and Human Services Press Office, 2007). As noted by the HHS Secretary, “Personalized health care will combine the basic scientific breakthroughs of the human genome with computer-age ability to exchange and manage data…. Increasingly it will give us the ability to deliver the right treatment to the right patient at the right time – every time” (United States Department of Health and Human Services Press Office, 2007). An important component of HHS’s vision for the role of IT in personalized health care is the use of computer systems to provide clinical decision support (CDS), which entails the delivery of pertinent knowledge and/or person-specific Copyright © 2009, Elsevier Inc. All rights reserved.
CDS Background: History, Examples, Evidence of Effectiveness, and Desirable Attributes
information to clinicians, patients and other health care stakeholders to enhance health and health care (Osheroff et al., 2006; United States Department of Health and Human Services, 2007b). In its draft March 2007 report entitled Realizing the Promise of Pharmacogenomics: Opportunities and Challenges, the HHS Secretary’s Advisory Committee on Genetics, Health, and Society identifies the need for CDS tools as an important challenge to achieving the promise of personalized medicine with regard to pharmacotherapy (United States Department of Health and Human Services Secretary’s Advisory Committee on Genetics, 2007). Given the need for CDS to fulfill the full potential of genomic and personalized medicine, this chapter explores the role that CDS could play in ensuring the consistent and equitable use of genomic technologies to personalize individuals’ health care. In beginning this exploration of the role of CDS in personalized medicine, relevant background information is provided on CDS in general.
CDS BACKGROUND: HISTORY, EXAMPLES, EVIDENCE OF EFFECTIVENESS, AND DESIRABLE ATTRIBUTES Brief History of CDS Research on the use of computers to support clinical decisionmaking began soon after computers became more widely available with the advent of minicomputers in the 1960s and microcomputers in the 1970s. These early CDS systems included a system for providing advice on the management of acid–base disorders (Bleich, 1969); a system for guiding antimicrobial therapy (Shortliffe et al., 1975); and a system that assisted with diagnosing a patient’s medical problem given the patient’s clinical findings (Pople et al., 1975). The first randomized controlled trial demonstrating the ability of a CDS system to improve patient care was reported in the mid-1970s (McDonald, 1976), and since then, numerous randomized controlled trials have demonstrated the ability of CDS systems to significantly improve clinical care for both acute and chronic medical conditions (Kawamoto et al., 2005). Despite mounting evidence of the ability of CDS systems to improve care quality and enhance patient safety, most health care decisions in the United States and elsewhere continue to be made without the assistance of CDS (Osheroff et al., 2006). The only major exception to the limited use of CDS has been with regard to the automated screening of drug–drug interactions and drug-allergy contraindications during the medication ordering process (Osheroff et al., 2006). Many factors have contributed to the limited use of CDS systems in the United States, including a reimbursement model that has traditionally failed to reward the provision of higher quality care (Petersen et al., 2006) and a frequent belief among clinicians that the use of decision aids would reduce the art of medicine to a “cookbook” approach to patient care (Cabana et al., 1999). Another factor restricting the use of CDS systems has been the limited adoption of core health IT
■
243
systems that facilitate the provision of CDS, such as electronic health record (EHR) systems that support the secure storage and efficient use of electronic health information regarding individuals (Jha et al., 2006; Institute of Medicine, 2003). Furthermore, a particularly important barrier to the widespread adoption of CDS systems has been the limited portability of most existing CDS systems and their underlying knowledge bases (Osheroff et al., 2006). This critical barrier to widespread CDS adoption will be the focus of later sections of this chapter. CDS Examples To gain insights into how CDS could support genomic and personalized medicine in the future, it is important to understand how CDS has been applied in the past. Table 20.1 provides an overview of traditional ways in which CDS has been leveraged to support clinical decision-making. In addition, Figures 20.1 and 20.2 provide screenshots from a clinical information system at Duke University Health System that provides a prototypical CDS function – the provision of reminders to a clinician regarding a patient’s overdue health maintenance needs and disease management needs (Lobach et al., 2007b). Also, Figure 20.3 provides a sample patienttailored care reminder letter that is sent to Medicaid beneficiaries residing in North Carolina regarding overdue disease management services. As discussed later in this chapter, these types of CDS interventions could be modified so that the care recommendations provided are optimized using relevant data regarding a patient’s genome and its downstream products. Evidence of CDS Effectiveness A large number of clinical trials have been conducted in both inpatient and outpatient settings to assess the impact of CDS interventions on health care delivery. Approximately twothirds of such trials have resulted in significant improvements in clinical practice (Kawamoto et al., 2005; Garg et al., 2005). Moreover, when CDS interventions possessing four critical attributes were evaluated in randomized controlled trials, 94% of the trials reported significant improvements in clinical practice (see next section for a description of the four critical attributes) (Kawamoto et al., 2005). Thus, CDS represents one of the most promising approaches available for improving care quality and ensuring patient safety. The positive impacts of CDS can include improved disease management, more appropriate pharmacotherapy, and improved adherence to recommended care standards (Garg et al., 2005). For example, when patient-specific care recommendations based on National Cholesterol Education Program (NCEP) treatment guidelines were placed on the charts of patients attending an academic cardiology clinic, patients in the intervention group achieved 93.8% overall compliance with the NCEP guidelines, as compared to 35.2% overall compliance among the control patients (p 0.001) (Stamos et al., 2001). As another example of CDS effectiveness, a time series study conducted at Brigham and Women’s Hospital in Boston identified significant benefits from the provision of CDS within a computerized provider order entry
244
CHAPTER 20
TABLE 20.1
■
Clinical Decision Support in Genomic Medicine
Traditional CDS types and examples
CDS type
Example
Care reminders to clinicians
When a clinician sees a patient in a clinic, he is provided with a printed encounter form that notes the patient appears to be overdue for a Pap test, a pneumococcal vaccine, and several laboratory tests for monitoring the patient’s diabetes.
Care reminders to patients
A clinic uses its clinical and/or billing systems to identify patients who appear to be overdue for mammograms and sends out care reminder letters.
Alerts
The clinician of a hospitalized patient is paged regarding a critical electrolyte imbalance detected from a recently processed serum chemistry panel. When ordering a cephalosporin antibiotic using a computerized provider order entry (CPOE) system, a clinician is alerted that the patient’s allergy to penicillin may result in an allergic reaction to the cephalosporin.
Order sets
When placing initial admission orders in a CPOE system for a patient with an acute ischemic stroke, the clinician is able to order the hospital’s standardized interventions for stroke patients by selecting the “acute ischemic stroke” order set.
Interactive consultation
A clinician enters a patient’s clinical findings into a web-based tool, which in turn returns a differential diagnosis given the patient’s findings. A clinician uses a web-based tool to obtain guidance on the optimal dose of a drug with a narrow therapeutic window (e.g., warfarin).
Figure 20.1 Sample care reminder module for health maintenance within a clinical information system.
(CPOE) system, which allows a clinician to directly enter medical orders using a computer (Ash et al., 2004). In this study, a CDSenabled CPOE system reduced the incidence of serious medication errors by 86%, with increasing benefits resulting from the incorporation of more advanced CDS capabilities into the system (p 0.0003) (Bates et al., 1999). Desirable CDS Attributes As with any complex intervention, CDS must be provided in an appropriate manner in order to have the desired impact on clinical care. In order to help determine the attributes of CDS systems that are most important to system effectiveness, we recently
Figure 20.2 Sample care reminder module for chronic disease management within a clinical information system.
conducted a systematic review of the CDS literature to (a) identify CDS attributes that were considered to be potentially important to CDS effectiveness by at least three sources and to (b) conduct a meta-analysis to determine whether the presence or absence of these attributes could explain the results of randomized controlled trials of CDS interventions (Kawamoto et al., 2005). The meta-analysis included 70 randomized controlled trials of computer-generated and manually-generated CDS
Potential uses of CDS to Support Genomic and Personalized Medicine
Durham Community Health Network Lincoln Community Health Center – Duke University Medical Center – Durham Country Department of Social Services Durham Country Health Department – Durham Pediatrics – Regional Pediatric Associates – Central Family Practice August 9, 2005 To the parents of We are sending you this letter to address your child’s health care needs. Based on our records, it appears your child may be due for the following services:
■
245
hand, a disease management module that is integrated with a specific electronic health record (EHR) system may have desirable attributes with regard to system effectiveness but may be much more difficult to deploy in a setting that uses a different EHR system or no EHR system at all. To a great extent, the central challenge in designing, developing, and deploying CDS interventions for genomic and personalized medicine will lie in creating systems that are effective yet still widely deployable.
Diabetes services that may be due: Hemoglobin A1c test: This test is recommended every 6 months for patients with diabetes. Cholesterol test: This test is recommended every 12 months for patients with diabetes. Urine protein test: This test is recommended every 12 months for patients with diabetes. Please call our office at (919) 477-2202 to schedule an appointment, so that the doctor can check to see if your child is in need of these services. Also, please bring this letter with you to the appointment and show it to the doctor. We look forward to seeing you soon! Sincerely,
Your Care Team Your Care Team Regional Pediatric Associates A Member of the Durham Community Health Network
Figure 20.3
Sample patient care reminder letter.
interventions. This analysis revealed that four features of CDS systems were significantly associated with these systems’ ability to significantly improve clinical practice: 1. automatic provision of CDS as part of clinician workflow, 2. provision of recommendations rather than just assessments, 3. provision of CDS at the time and location of decisionmaking, and 4. computer-based generation of the CDS. In particular, the automatic provision of CDS as a part of the routine clinician workflow was tightly correlated with study outcomes (p 0.00001). Additionally, CDS systems lacking one or more of these four features failed to improve clinical practice in over half of the studies (n 39), whereas CDS systems with all four features significantly improved clinical practice in 94% of the randomized controlled trials (n 32). Thus, CDS systems should be designed to incorporate these features whenever feasible and appropriate. Beyond features desirable for CDS effectiveness, there are attributes that are desirable for ease of CDS maintenance and deployment. These attributes include centralized knowledge management; the ability to re-use knowledge created for one CDS application for a different CDS application; adherence to relevant health IT standards; low cost of initial deployment and ongoing maintenance; and the ability to easily re-deploy the CDS capability in various clinical settings with heterogeneous IT infrastructures (Kawamoto, 2006). An important insight is that a trade-off often arises between making a CDS system effective and making a CDS system easy to maintain and deploy. For example, a standalone Web application or a personal digital assistant (PDA) tool may be easy to deploy widely, but such systems require clinicians to proactively access the tool and therefore cannot provide CDS automatically as a part of routine clinician workflow. On the other
POTENTIAL USES OF CDS TO SUPPORT GENOMIC AND PERSONALIZED MEDICINE The vast size and enormous complexity that can be associated with genomic data will undoubtedly pose new challenges to existing approaches to delivering CDS. At the same time, the purpose of genomic and personalized medicine – the use of available data to optimize an individual’s health – is fully aligned with the way in which CDS has been leveraged for many decades. Thus, the use of CDS for genomic and personalized medicine should be seen as a natural extension of how CDS has been traditionally used, rather than as a completely novel development. Indeed, from the perspective of clinicians and patients, CDS for “traditional medicine” and “genomic medicine” should be delivered seamlessly and in concert. For example, consider the case of a patient with hypertension for whom pharmacotherapy is about to be initiated. In practice, CDS can facilitate the optimal selection of antihypertensive therapy whether a particular drug class is selected for the patient because the patient has a genetic profile predicting a favorable response to the drug class or whether a particular drug class is selected because the patient has a clinical condition favoring a particular drug class (e.g., type I or type II diabetes mellitus, which is considered a compelling indication for the use of an angiotensin converting enzyme inhibitor or an angiotensin II receptor blocker as the antihypertensive therapy of choice). Regardless of whether the rationale for a clinical recommendation is due to genomic data, non-genomic data, or a combination of the two, a clinician would expect to receive the CDS in a similar manner. Thus, CDS for genomic and personalized medicine should be provided alongside the CDS provided using traditional medical data. CDS supporting genomic and personalized medicine could be delivered in a variety of ways and in a variety of settings. Essentially, CDS on genomic medicine could be provided in any situation in which a clinician, patient, or other health care stakeholder needs to make a clinical decision that could be optimized through the use of genomic data. Table 20.2 provides examples of how traditional CDS mechanisms could be adapted to support genomic medicine. Given the current prominence of personalized cancer management and pharmacogenomics in clinical genomics, the examples in the table place particular emphasis on these aspects of genomic and personalized medicine.
246
CHAPTER 20
TABLE 20.2
■
Clinical Decision Support in Genomic Medicine
Traditional CDS types and potential applications to genomic and personalized medicine
CDS type
Example application to genomic and personalized medicine
Care reminders to clinicians
When a clinician sees a patient in a clinic, she is provided with a printed encounter form that notes the patient appears to be overdue for a mammogram and a pneumococcal vaccine. The Pap test is recommended more frequently for this patient compared to the general population because the patient has a family history of cervical cancer.
Care reminders to patients
A clinic identifies patients who appear to be overdue for mammograms and sends out care reminder letters. A patient’s family history, BRCA1/2 status, and other genetic predictors are considered when determining the screening frequency for the patient.
Alerts
When ordering a thiazide diuretic for a patient with hypertension using a CPOE system, a clinician is alerted that the patient’s genetic profile is associated with significant adverse reactions to the drug class. An angiotensin converting enzyme inhibitor is recommended as the optimal therapy given the patient’s genetic profile. When an oncologist orders a chemotherapy regimen for a patient with lung cancer using a CPOE system, the system alerts the clinician that the patient should be placed on a more aggressive regimen given the poor prognosis associated with the gene expression profile of the patient’s tumor.
Order sets
When placing initial admission orders in a CPOE system for a patient with an acute ischemic stroke, the clinician is able to use the hospital’s standardized order set for patients with the condition. The CPOE system automatically adjusts the recommended orders based on the patient’s genetic profile, which was on file from a previous visit. For example, drug doses are adjusted based on the patient’s age, gender, serum creatinine level, and genotype with respect to drug metabolizing enzymes.
Interactive consultation
An oncologist uses a web-based tool provided by the National Cancer Institute to identify whether any genomic tests should be ordered to determine the best therapeutic regimen for a patient with cancer.
LIMITED DEPLOYABILITY: THE POTENTIAL ACHILLES’ HEEL OF CDS SYSTEMS FOR GENOMIC MEDICINE Given the rapid advances in the scientific community’s understanding of how our health is impacted by our genes and their downstream products, it will likely only be a matter of time until there is a large body of knowledge on how the management of various conditions can be optimized through the use of genomic data. This knowledge will bring with it an increasing need for CDS to support the use of genomic data in clinical practice. Currently, there are only a small number of reports in the literature describing the use of a CDS tool to support decision-making based on genetic data, and these tools focus almost exclusively on the prediction of cancer risk based on family history (Bianchi et al., 2007; Emery et al., 2000; Wilson et al., 2006). In the near future, however, we anticipate that there will be an increasing number of reports regarding CDS tools for genomic medicine and the impact of these tools as determined through clinical trials. As has been the case with CDS systems in support of traditional medical interventions, we anticipate that high-quality clinical trials will soon demonstrate that significant clinical benefits
can be achieved through the use of CDS systems to support genomic medicine. For example, we anticipate clinical trials will demonstrate that the use of CDS systems can improve clinicians’ appropriate use and interpretation of various genomic tests, including tests that characterize patients’ drug metabolizing enzymes (see Chapter 27) and tests that measure gene expression levels in tumor specimens (see Chapter 13). Such trials will be an important step toward fulfilling the promise of genomic and personalized medicine, as they will provide valuable insights into how information regarding genomic interventions can be effectively delivered to clinicians and patients. However, a critical question that must be asked of such trials is whether the CDS intervention of interest can be widely replicated in other settings for operational use. As discussed earlier in this chapter, an important limitation of many effective CDS systems is that they are difficult to deploy in settings that differ from the original deployment setting. Unless this issue is explicitly considered and addressed, it is likely that many CDS interventions for genomic medicine will also have limited impact due to the difficulty of re-deploying the interventions in other settings. Thus, the remainder of this chapter will be devoted to exploring challenges to the widespread deployment of effective CDS systems and potential approaches to overcoming these challenges for genomic and personalized medicine.
Challenges to Widespread Deployment of Effective CDS Systems
CHALLENGES TO WIDESPREAD DEPLOYMENT OF EFFECTIVE CDS SYSTEMS The widespread deployment of effective CDS systems is hindered by several important challenges. This section describes four of the most important challenges that exist in the United States: 1. the limited adoption of EHR systems; 2. significant variability in how clinical and genomic data are represented in the various health IT systems that are used across the country; 3. the high cost of developing and maintaining medical knowledge that has been encoded in a machine-executable format suitable for CDS; and 4. the limited reusability and portability of the knowledge in many CDS systems. Challenge # 1: Limited adoption of EHR systems. EHR systems facilitate the delivery of CDS by capturing the coded patient data required for CDS and by providing a natural user interface for making CDS available to clinicians. Thus, the limited adoption of EHR systems in the United States hinders efforts at widely providing CDS. Currently, it is estimated that only about 25% of US physicians use EHR systems in the outpatient setting (Jha et al., 2006). Challenge # 2: Significant variability in how patient data are represented in health IT systems. In the United States, a large variety of commercial and locally developed health IT systems are in use. These systems can vary widely in how they represent both clinical and genomic patient data. For example, health IT systems oftentimes use idiosyncratic codes and terminologies to identify medical concepts (e.g., diagnoses, medications, and laboratory tests). Moreover, the manner in which relevant clinical information is modeled (i.e., represented) can differ in important ways between systems. For example, one EHR system may record the results of a Pap test as a narrative text, while another system may record the results using both a narrative text and structured data (e.g., whether the result was abnormal and the nature of the abnormality). As another example, different EHR systems may store the results of genetic tests as structured data or as narrative text (Scheuner, 2007). This variability makes it difficult to take a CDS module that was originally designed to work with one EHR system and re-leverage that module in a different EHR system. Challenge # 3: High cost of developing and maintaining machineexecutable medical knowledge. In order to support CDS, medical knowledge must be encoded in a machine-executable format that can be used to interpret structured patient data and generate patient-specific care recommendations. This knowledge encoding process can become quite expensive due to a number of factors, including the need for highly skilled and expensive personnel who are knowledgeable in both clinical medicine and information technology; the need to precisely define concepts that are oftentimes left ambiguous in care guidelines intended
■
247
for human consumption; and the need to rigorously test the knowledge to ensure the accuracy of the encoded algorithms. The cost of encoding the knowledge goes up with the complexity of the underlying clinical algorithm, and as such, the complex algorithms associated with genomic medicine will require significant resources to encode in a format suitable for use in CDS systems. Challenge # 4: Limited reusability and portability of knowledge in many CDS systems. Given the high cost of knowledge engineering, it is critical that machine-executable medical knowledge created for one CDS application be re-used across institutions and across different types of CDS applications. However, in many cases, machine-executable medical knowledge that is used to drive a CDS application at one institution cannot be easily re-used at a different institution or in a different type of CDS application. Common reasons for the difficulty of re-using executable medical knowledge include the embedding of the knowledge within a specific CDS system’s application code; a dependence of the knowledge on how patient data are represented and stored in a specific clinical information system; and the lack of a widely accepted standard formalism for representing machine-executable medical knowledge.This difficulty of re-using machine-executable medical knowledge is a major factor in the limited availability of CDS capabilities in clinical practice. Therefore, it is important to work toward CDS knowledge resources that can be decoupled from the applications that they support and, thus, can be ported to other environments. Strategies for Addressing Challenges to Widespread Deployment of CDS Systems Table 20.3 outlines strategies for addressing the challenges to the widespread deployment of CDS systems that were just described. Many of these strategies are part of a standards-based software architecture that we have proposed for enabling CDS on a national scale and for fulfilling the strategic objectives of the Roadmap for National Action on CDS commissioned by the US Department of Health and Human Services’ Office of the National Coordinator for Health IT (Kawamoto and Lobach, 2007). Strategies for Addressing Limited Adoption of EHR Systems With regard to the limited adoption of EHR systems, an important strategy for addressing this challenge is encouraging and facilitating the adoption of EHR systems. In the United States, both the government and the private sector have been taking steps to encourage the adoption of EHR systems, due in no small part to the anticipated benefits of CDS delivered in the context of EHR systems. For example, the federal government has eased regulations that had previously forbidden hospitals from financially supporting the acquisition of EHR systems by referring physicians (Egan, 2006), and health insurers have financially supported the adoption of EHR systems among physician practices (Havenstein, 2007). Despite these efforts, however, the
248
CHAPTER 20
■
Clinical Decision Support in Genomic Medicine
T A B L E 2 0 . 3 Strategies for addressing challenges to widespread deployment of effective CDS systems Challenge Limited adoption of EHR systems
Strategy for addressing challenge Increase adoption of EHR systems Design CDS systems that do not require EHR systems
Significant variability in how patient data are represented in health IT systems
Standardize the representation of clinical and genomic data in IT systems Allow relevant clinical and genomic data to be located and retrieved using a standard software interface Encourage adoption of relevant standards
High cost of developing and maintaining machineexecutable medical knowledge
Create and manage machine-executable medical knowledge in a centralized manner
Limited re-use and portability of CDS knowledge
Allow machine-executable medical knowledge to be accessed using a standard software interface
reality remains that a large majority of physicians in the United States continue to practice without an EHR system. Given this reality, another strategy to address the limited adoption of EHR systems is to design and deploy CDS systems that do not depend on the availability of an EHR system. In particular, all public and private health insurance plans in the United States require a standard set of data from health care providers for the purposes of reimbursement. These claims data include standardized data on a patient’s diagnoses, prescription medications that were dispensed, and the procedures that were conducted (including the identity of laboratory tests, but not test results). While claims data collected for the purposes of reimbursement are not as detailed and accurate as clinical data collected for the purposes of patient care in an EHR system, claims data can be used to drive a number of effective CDS interventions, even in clinical settings that make minimal or no use of health IT systems (Javitt et al., 2005). With regard to genomic medicine, claims data could be combined with genomic test results to deliver CDS on a large scale, as claims data are typically managed by a relatively small number of large health insurers, and as a significant portion of the laboratory test market is controlled by a small number of large commercial laboratories (Fuhrmann, 2006). For example, if a major insurer could establish a data sharing agreement with the major clinical laboratories servicing its covered individuals, then the insurer would be capable of identifying patients for whom genetic tests are potentially indicated or for whom the results of genetic tests
indicate that a certain course of therapy should be taken. These genomic care recommendations could then be communicated to clinicians and/or patients to guide their decision-making. Strategies for Addressing Variability in Patient Data Representation in Health IT Systems The CDS scenario just described, as well as any widely replicable model for CDS, relies on the ability to consistently access patient data in a standardized format. Thus, an important strategy for addressing the significant variability that exists with regard to patient data is to standardize how clinical and genomic data are represented and communicated across various health IT systems. Standardization of such data include (i) standardization of the information models used to represent concepts important to clinical care (e.g., problems, procedures, test results, medications, DNA sequences, and gene expression assay results); (ii) standardization of the terminologies used to identify concepts in the information models; and (iii) standardization of how such data are communicated between systems (e.g., between an EHR system and a CDS system). Relevant standards have been developed by standards development organizations (SDOs) for all of these aspects of standardization (Hammond, 2005). In particular, standards in all three categories have been developed by Health Level 7 (HL7), a not-for-profit, volunteer-driven SDO which is the most prominent SDO in the health care sector internationally. HL7 standards include standards for the representation of traditional clinical data as well as clinically relevant genomic data (Shabo, 2006). These standards make use of standard clinical terminologies, including the World Health Organization’s International Classification of Diseases (ICD) for medical conditions; the Systematized Nomenclature of Medicine, Clinical Terms (SNOMED CT) for a variety of clinical concepts; the Logical Observation Identifiers, Names, and Codes (LOINC) for laboratory tests; RxNorm for medications; the microarray and gene expression markup language (MAGE-ML) for gene expression data; and the bioinformatic sequence markup language (BSML) for sequencing data (Hammond, 2005; Shabo, 2006). Furthermore, an HL7 specification known as the HL7 Retrieve, Locate, and Update Service standard provides a common approach for locating and retrieving patient data – including genomic data – using a standard software interface (Health Level 7, 2006b). As noted above, many of the standards needed for interoperable, standards-based exchange of clinical and genomic data do exist. Thus, the problem with data heterogeneity in the current health care industry lies less with the lack of appropriate standards and more with the lack of widespread and effective use of the available standards (Hammond, 2005). In recognition of this problem, there are currently significant efforts underway to specify how existing standards should be used to meet important clinical needs (e.g., the communication of laboratory data) and to encourage the use of existing standards in the specified manner (e.g., through certification). An initiative prominent in this regard is Integrating the Healthcare Enterprise (IHE), which was initiated in 1998 as a joint effort between the Radiological Society of North America and the Healthcare Information and Management Systems Society (Siegel and Channin, 2001).
References
Another prominent organization in this space is the Healthcare Information Technology Standards Panel (HITSP), which is a public-private effort initiated in 2005 by the US Department of Health and Human Services to identify and harmonize data and technical standards for health care (United States Department of Health and Human Services, 2007a). Strategies for Addressing High Cost of Developing and Maintaining Machine-Executable Medical Knowledge This challenge can be addressed by centralizing the knowledge authoring and management process so that the translation of medical knowledge into a machine-executable format suitable for CDS occurs once rather than being duplicated unnecessarily across a large number of institutions. For example, leading academic institutions, professional medical associations, or governmental agencies could create and maintain machine-executable medical knowledge on behalf of a large group of health care organizations. Then, as the second step, these centrally managed knowledge resources could be made accessible to health care organizations and vendors of health IT systems through a standard mechanism. Strategies for such standardized sharing of CDS knowledge are discussed next. Strategies for Addressing the Limited Reusability and Portability of Knowledge in CDS Systems One approach for CDS knowledge sharing that has been attempted in the past involves standardizing how machineexecutable medical knowledge is represented. While a formalism for knowledge representation known as the Arden Syntax (Pryor and Hripcsak, 1993) has been adopted as an HL7 standard, the use of this approach has been limited by the fact that it has been difficult to obtain industry-wide consensus on a single best way to represent machine-executable medical knowledge. An alternate approach to accommodating the reuse and portability of machine-executable medical knowledge is to expose the decision-making capabilities of the knowledge through a standard software interface. This approach does not place a restriction on how the knowledge is represented, as what is shared is not the knowledge itself but the decisionmaking capabilities of that knowledge. In this approach, knowledge resources are hosted by a CDS software service which takes patient data as the input and returns patient-specific inferences as the output. For example, a CDS service may take in as input a patient’s cytochrome P450 genotype and drug name and return the patient’s P450 metabolism status for the specified drug. Using this service-based approach to CDS, our research group has leveraged a centrally managed knowledge base to implement a number of operational CDS applications, including an enterprise-wide disease management system within the
■
249
Duke University Health System (Figures 20.1 and 20.2) (Lobach et al., 2007b) and a population health management system for North Carolina Medicaid beneficiaries that provides alerts to care providers, feedback reports to clinics, and care reminder letters to patients (Figure 20.3) (Lobach et al., 2007a). The Web-based CDS service at the core of these systems is known as SEBASTIAN (Kawamoto and Lobach, 2005), and it is the basis of an HL7 standard for CDS services known as the HL7 Decision Support Service standard (Health Level 7, 2006a). Together with a centralized approach to knowledge management, the HL7 Decision Support Service standard could allow genomic medicine algorithms to be widely leveraged by various CDS applications in a plug-and-play manner.
CONCLUSIONS Genomic and personalized medicine has the potential to transform the delivery of health care, making clinical medicine increasingly a science rather than an art. Effective CDS systems will be needed to make genomic and personalized medicine a routine aspect of clinical care, but a deliberate and concerted effort will be required to ensure that CDS interventions for genomic medicine can be deployed in a widespread manner. In particular, efforts at enabling the widespread practice of genomic medicine through CDS should (i) encourage the adoption of EHR systems while minimizing the IT infrastructure required for a CDS system to be used; (ii) be standards-based; (iii) support the centralized management of machine-executable genomic medicine algorithms; and (iv) enable these algorithms to be leveraged by a large number of health care organizations through a standard software interface. While significant challenges do exist in the road to a health care system in which genomic and personalized medicine is routinely practiced through the support of CDS, we believe the strategies outlined in this chapter will provide a solid foundation for tackling these challenges in the years to come.
DISCLOSURES The authors are seeking to establish a commercial implementation of the HL7 Decision Support Service described in this chapter using the SEBASTIAN CDS technology. The authors hold intellectual property rights to SEBASTIAN (patent pending). The authors and Duke University may benefit financially if products utilizing the SEBASTIAN technology are commercially successful. With regard to the recommendations made in this chapter, the SEBASTIAN system represents just one of potentially many approaches for instantiating an HL7 Decision Support Service.
REFERENCES Abrahams, E., Ginsburg, G.S. and Silver, M. (2005). The personalized medicine coalition: Goals and strategies. Am J Pharmacogenomics 5, 345–355.
Ash, J.S., Gorman, P.N., Seshadri, V. and Hersh, W.R. (2004). Computerized physician order entry in US hospitals: Results of a 2002 survey. J Am Med Inform Assoc 11, 95–99.
250
CHAPTER 20
■
Clinical Decision Support in Genomic Medicine
Bates, D.W., Teich, J.M., Lee, J., Seger, D., Kuperman, G.J., Ma’Luf, N., Boyle, D. and Leape, L. (1999). The impact of computerized physician order entry on medication error prevention. J Am Med Inform Assoc 6, 313–321. Bianchi, F., Galizia, E., Bracci, R., Belvederesi, L., Catalani, R., Loretelli, C., Giorgetti, G., Ferretti, C., Bearzi, I., Porfiri, E. et al. (2007). Effectiveness of the CRCAPRO program in identifying patients suspected for HNPCC. Clin Genet 71, 158–164. Bleich, H.L. (1969). Computer evaluation of acid-base disorders. J Clin Invest 48, 1689–1696. Blum, R.A., Wylie, N., England, T. and French, C. (2005). HIV resistance testing in the USA – a model for the application of pharmacogenomics in the clinical setting. Pharmacogenomics 6, 169–179. Cabana, M.D., Rand, C.S., Powe, N.R., Wu, A.W., Wilson, M.H., Abboud, P.A. and Rubin, H.R. (1999). Why don’t physicians follow clinical practice guidelines? A framework for improvement. JAMA 282, 1458–1465. Egan, C. (2006). HHS announces regulations to support IT adoption. http://www.ihealthbeat.org/articles/2006/8/2/HHS-AnnouncesRegulations-To-Support-IT-Adoption.aspx?a1. Emery, J., Walton, R., Murphy, M., Austoker, J., Yudkin, P., Chapman, C., Coulson, A., Glasspool, D. and Fox, J. (2000). Computer support for interpreting family histories of breast and ovarian cancer in primary care: comparative study with simulated cases. BMJ 321, 28–32. Fuhrmann, R. (2006). The quest for market share. http://www.fool. com/investing/general/2006/07/24/the-quest-for-market-share. aspx. Garg, A.X., Adhikari, N.K., McDonald, H., Rosas-Arellano, M.P., Devereaux, P.J., Beyene, J., Sam, J. and Haynes, R.B. (2005). Effects of computerized clinical decision support systems on practitioner performance and patient outcomes: A systematic review. JAMA 293, 1223–1238. Hammond, W.E. (2005). The making and adoption of health data standards. Health Affairs 24, 1205–1213. Havenstein, H. (2007). Health insurer offering free electronic health records to doctors in four states. http://www.computerworld. com/action/article.do?commandviewArticleBasic&articleId9 017520&intsrchm_list. Health Level 7. (2006a). HL7 Decision Support Service specification (Draft Standard for Trial Use). http://www.hl7.org/v3ballot/html/ infrastructure/dss/dss.htm. Health Level 7. (2006b). HL7 Retrieve, Locate, and Update Service specification (Draft Standard for Trial Use). http://www.hl7.org/ v3ballot/html/infrastructure/rlus/rlus.htm. Institute of Medicine (2003). Key Capabilities of an Electronic Health Record System: Letter report. National Academy Press, Washington DC. Javitt, J.C., Steinberg, G., Locke, T., Couch, J.B., Jacques, J., Juster, I. and Reisman, L. (2005). Using a claims data-based sentinel system to improve compliance with clinical guidelines: Results of a randomized prospective study. Am J Manag Care 11, 93–102. Jha, A.K., Ferris, T.G., Donelan, K., DesRoches, C., Shields, A., Rosenbaum, S. and Blumenthal, D. (2006). How common are electronic health records in the United States? A summary of the evidence. Health Affairs 25, w496–w507. Kawamoto, K. (2006). Integration of knowledge resources into applications to enable clinical decision support: Architectural considerations. In Clinical Decision Support: The Road Ahead (R.A. Greenes, ed.), Elsevier, Inc, U.S.A., pp. 503–538.
Kawamoto, K. and Lobach, D.F. (2005). Design, implementation, use, and preliminary evaluation of SEBASTIAN, a standardsbased Web service for clinical decision support. Proc AMIA Symp, 380–384. Kawamoto, K. and Lobach, D.F. (2007). Proposal for fulfilling strategic objectives of the US roadmap for national on decision support through a service-oriented architecture leveraging HL7 services. J Am Med Inform Assoc 14, 146–155. Kawamoto, K., Houlihan, C.A., Balas, E.A. and Lobach, D.F. (2005). Improving clinical practice using clinical decision support systems: A systematic review of trials to identify features critical to success. BMJ 330, 765–768. Lobach, D.F., Kawamoto, K., Anstrom, K.J., Kooy, K.R., Eisenstein, E.L., Silvey, G.M., Willis, J.M., Johnson, F. and Simo, J. (2007a). Proactive population health management in the context of a regional health information exchange using standards-based decision support. Proc AMIA Symp 473–477. Lobach, D.F., Kawamoto, K., Anstrom, K.J., Russel, M.L., Woods, P. and Smith, D. (2007b). Development, deployment and usability of a point-of-care decision support system for chronic disease management using the recently-approved HL7 Decision Support Service standard. Medinfo 861–865. McDonald, C.J. (1976). Protocol-based computer reminders, the quality of care and the non-perfectability of man. N Engl J Med 295, 1351–1355. McGlynn, E.A., Asch, S.M., Adams, J., Keesey, J., Hicks, J., DeCristofaro, A. and Kerr, E.A. (2003). The quality of health care delivered to adults in the United States. N Engl J Med 348, 2635–2645. National Committee for Quality Assurance. (2006). The state of health care quality 2006. http://www.ncqa.org/tabid/136/Default.aspx. Osheroff, J.A., Teich, J.M., Middleton, B., Steen, E.B., Wright, A., and Detmer, D.E. (2006). A roadmap for national action on clinical decision support. American Medical Informatics Association, Bethesda, MD. http://www.amia.org/inside/initiatives/cds/ cdsroadmap.pdf. Paik, S., Tang, G., Shak, S., Kim, C., Baker, J., Kim, W., Cronin, M., Baehner, F.L., Watson, D., Bryant, J. et al. (2006). Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer. J Clin Oncol 24, 3726–3734. Petersen, L.A., Woodard, L.D., Urech, T., Daw, C. and Sookanan, S. (2006). Does pay-for-performance improve the quality of health care?. Ann Intern Med 145, 265–272. Piccart-Gebhart, M.J., Procter, M., Leyland-Jones, B., Goldhirsch, A., Untch, M., Smith, I., Gianni, L., Baselga, J., Bell, R., Jackisch, C. et al. (2005). Trastuzumab after adjuvant chemotherapy in HER2positive breast cancer. N Engl J Med 353, 1659–1672. Pople, H.E., Myers, J.D. and Miller, R.A. (1975). DIALOG: A model of diagnostic logic for internal medicine. Proceedings of the Fourth International Joint Conference on Artificial Intelligence., 848–855. Pryor, T.A. and Hripcsak, G. (1993). The Arden syntax for medical logic modules. Int J Clin Monit Comput 10, 215–224. Scheuner, M.T. (2007). Family history and genetic tests in electronic health records. http://www.hhs.gov/healthit/ahic/materials/03_ 07/phc/scheuner.ppt. Shabo, A. (2006). Clinical genomics data standards for pharmacogenetics and pharmacogenomics. Pharmacogenomics 7, 247–253. Shortliffe, E.H., Davis, R., Axline, S.G., Buchanan, B.G., Green, C.C. and Cohen, S.N. (1975). Computer-based consultations in clinical therapeutics: Explanation and rule acquisition capabilities of the MYCIN system. Comput Biomed Res 8, 303–320.
Recommended Resources
Siegel, E.L. and Channin, D.S. (2001). Integrating the Healthcare Enterprise: A primer. Part 1. Introduction. Radiographics 21, 1339–1341. Stamos, T.D., Shaltoni, H., Girard, S.A., Parrillo, J.E. and Calvin, J.E. (2001). Effectiveness of chart prompts to improve physician compliance with the National Cholesterol Education Program guidelines. Am J Cardiol 88, 1420–1423. The Personalized Medicine Coalition. (2006). The case for personalized medicine. http://www.personalizedmedicinecoalition.org/communications/TheCaseforPersonalizedMedicine_11_13.pdf. United States Department of Health and Human Services. (2007a). Data and technical standards: Health Information Technology Standards Panel (HITSP). http://www.hhs.gov/healthit/standards/activities/. United States Department of Health and Human Services. (2007b). Goals of personalized healthcare initiative. http://www.hhs.gov/ myhealthcare/goals/index.html. United States Department of Health and Human Services Press Office. (2007). HHS Secretary Leavitt announces steps toward a future
■
251
of “personalized health care”. http://www.hhs.gov/news/press/ 2007pres/20070323a.html. United States Department of Health and Human Services Secretary’s Advisory Committee on Genetics, Health and Society. (2007). Realizing the promise of pharmacogenomics: opportunities and challenges (draft report). http://www4.od.nih.gov/oba/ SACGHS/SACGHS_PGx_PCdraft.pdf. Willard, H.F., Angrist, M. and Ginsburg, G.S. (2005). Genomic medicine: Genetic variation and its impact on the future of health care. Philos Trans R Soc Lond B Biol Sci 360, 1543–1550. Wilson, B.J., Torrance, N., Mollison, J., Watson, M.S., Douglas, A., Miedzybrodzka, Z., Gordon, R., Wordsworth, S., Campbell, M., Haites, N. et al. (2006). Cluster randomized trial of a multifaceted primary care decision-support intervention for inherited breast cancer risk. Fam Pract 23, 537–544.
RECOMMENDED RESOURCES Health Level 7. (2007). Technical committees and special interest groups. http://www.hl7.org/Special/committees. This Web page provides links to committees within the HL7 standards development organization, which is playing an active role in standardization efforts related to genomic medicine and the use of CDS to support this approach to health care. Committees of particular interest may include the HL7 Clinical Genomics Special Interest Group and the HL7 CDS Technical Committee. Each committee’s Web site includes a link to register for the committee’s list service. Kawamoto, K. and Lobach, D.F. (2007). Proposal for fulfilling strategic objectives of the US roadmap for national action on decision support through a service-oriented architecture leveraging
HL7 services. J Am Med Inform Assoc 14, 146–155. This article outlines a standards-based approach to CDS that could form the basis of a national strategy for enabling the widespread deployment of effective CDS systems. The recommendations made in this chapter draw from the approach proposed in this article. Greenes, R.A. (2007). Clinical Decision Support: the Road Ahead. Elsevier, Inc., USA.This book provides a comprehensive overview of CDS, including lessons learned from past implementations and state-ofthe-art approaches to managing CDS systems and their underlying knowledge bases. This book is recommended for the reader who wishes to obtain an in-depth understanding of CDS from a single authoritative source.
CHAPTER
21 Online Health Information Retrieval by Consumers and the Challenge of Personal Genomics Mark S. Boguski
INTRODUCTION In an age of personalized medicine, nothing represents the zeitgeist more than individual consumers using the Internet and World Wide Web to seek medical and health information. According to surveys by the Jupiter organization and Harris Interactive, 71% of people who use the Internet also used it to seek health information in 2007, and this percentage, which represents an estimated 160 million people in the United States, had increased by 37% since 2005 (Anonymous, 2007b; Levy, 2007). Consumers perform health information search (HIS) and retrieval (HIR) for themselves as well as for friends and family. Studies have shown that most of these consumers do not later discuss the information with a health-care provider and that, for many people, the Internet may be the primary or even sole source of health information. (Fox and Fallows, 2003). Seventy percent of people who obtain health information online say that it has influenced a decision about their treatment. Clearly, it is important for health care professionals to understand how their patients find health information and the pitfalls associated with this activity. Indeed, given the challenges that consumers face in obtaining quality health care information from Internet sources, health-care providers will increasingly be in a position where they have to act as reviewers of information Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 252
and as educators of patients who bring this information to their attention.
CHARACTERISTICS OF CONSUMER SEARCHES FOR HEALTH INFORMATION A 2007 study by the Pew Internet & American Life Project (Fox, 2006) showed that the top 10 reasons American adults search online for health information are: ● ● ● ● ● ● ● ● ● ●
a specific disease or medical condition; a particular medical treatment or procedure; diet, nutrition and supplements; exercise or fitness; prescription and non-prescription drugs; a particular doctor or hospital; alternative medicine or treatments; health insurance; mental health issues and environmental health hazards.
Zeng and Tse (Zeng and Tse, 2006) reported that, in performing searches, consumers use query terms that consist of Copyright © 2009, Mark S. Boguski All rights reserved.
What and Where are Consumers Searching?
“every-day language, technical terms (with or without knowledge of the underlying concepts) and various explanatory models, all influenced by psychosocial and cultural variations...” According to Lorence and Spink (Lorence and Spink, 2004), lay terminology is only partly effective in retrieving useful health information and often produces irrelevant or misleading information because, as reported by Zeng et al. (Zeng et al., 2006), “… the terms and concepts used by consumers often do not accurately reflect their information needs and therefore do not constitute effective queries.” Problems with such consumer health vocabularies may occur at various levels (Zeng et al., 2002): ● ●
●
Lexical (form-level) mismatches (e.g., misspellings) Semantic misunderstanding (e.g., incomplete or misinterpretation of abbreviations or acronyms, over-generalization, redundancy) Misleading mental models, conceptual models that consumers employ either incorrectly and/or in a manner different than health care professionals.
Form-level (lexical) mismatches need no further explanation, as examples are easy to imagine. In the case of semantic mismatches, searches with general search engines may yield a large amount of irrelevant information. Consider the abbreviation “MS”, which a consumer might use to mean multiple sclerosis but could also be interpreted to mean Microsoft, mass spectrometry or the state of Mississippi (Romacker et al., 2006). In another example, “CHF” would be interpreted as congestive heart failure by a physician whereas an international banker would interpret CHF as a unit of currency, the Swiss franc. Other groups (Tse and Soergel, 2003) have similarly characterized problems with consumer health vocabularies on multiple levels, namely: ●
●
●
Shared forms/different concepts (e.g., the phrase “the results of the [diagnostic] test were negative” could mean “unfavorable” in the mind of a patient but is usually a positive step forward in the mind of a physician.) Different forms/shared concepts (e.g., “blood cancer” is a lay term that a physician might refer to as a “hematologic malignancy.” Likewise when a physician refers to “metastatic disease”, the patient would say “the cancer has spread.”) Different forms/different concepts (e.g., “Miracle cure” rather than a “statistically unlikely” or “idiopathic” remission of a disease, emergence from coma, etc.).
The most challenging HIS/HIR problems of all undoubtedly arise at the conceptual level because it is intimately tied to an individual’s level of health literacy (McCray, 2005). Conceptlevel mismatches might include notions outside the framework of mainstream medicine such as homeopathy, acupuncture and other alternative medicine concepts. Another example is when a patient may focus on a body part (anatomical location) or symptom when a clinician might view the same problem in terms of pathophysiology or from a sub-specialty perspective. Even mainstream medical concepts are subject to significant cultural influences (Payer, 1996).
■
253
There have been a number of attempts and exploratory studies to map lay terminology to controlled vocabularies in a number of formal coding systems and comprehensive collections of medical terms such as CTP, SNOMED, MeSH, ICD-9 and the UMLS Metathesaurus (Shortliffe et al., 2003). Furthermore, ontologies have been employed to perform semantic expansion on queries that may be too general or simplistic (see (Spasic et al., 2005) and below). The idea is to take queries consisting of consumer health terms and to translate or reformulate them into professional terms and qualifiers with the aim of improving query meaning, precision and recall. So far such experiments have met with only minor or mixed success (Plovnick and Zeng, 2004; Zeng et al., 2006). There is also the challenge of “back translating” from professional to consumer language the results of a search using a reverse medical dictionary. (See also Chapter 17.) Two additional reasons for suboptimal performance of consumer HIS/HIR is short query length and low complexity. Consumer HIS/HIR queries tend to be very short and too general to be effective. Indeed one study found that 63% of consumer queries consisted of only a single word and only 10% contained more than two words (Zeng et al., 2002). Also, consumers are generally unaware of stop words that add little or nothing to search performance. Thus, limited knowledge of both medical vocabulary and query string search principles contributes to the construction of simplistic, ineffective queries. Another important aspect is the inability or unwillingness to construct complex queries (such as Boolean combinations of terms) even when such “advanced search” options are provided via simple check boxes. Users seem to prefer (or at least have become accustomed to) search and retrieval as an iterative process during which their goals are refined, focused or revised again and again (Fredin and Prabu, 1998) until either a satisfactory answer is obtained or they abandon the effort altogether due to frustration or fatigue. A precise, unambiguous and well-formed query would ideally produce the answer(s) one is looking for in the first iteration. This ideal assumes that the information sources containing the information exist and are accessible.
WHAT AND WHERE ARE CONSUMERS SEARCHING? Equally important to how searches are formulated by consumers are the sources of the information that are being searched. As an editorial in the British Medical Journal put it: “…the internet has vastly increased the availability of information, but often what it offers is untailored, incomplete, irrelevant, and plain wrong” (Jones, 2003) although actual cases of harm have been difficult to document (Crocco et al., 2002). Online health information is problematic as it encompasses everything from evidencebased, peer-reviewed content in professional journals to advertiser-sponsored content and personal testimonials (Table 21.1). A consumer’s online HIS/HIR experience may even be greatly influenced by non-substantive characteristics of a site such as an attractive and professional-looking appearance even though the quality of the underlying information is unknown.
254
CHAPTER 21
TABLE 21.1
■
Online Health Information Retrieval
Comparisons of Selected Online Health Information Sites Libraries, Publishers
e-Magazines
Search portals
Social Networks
Content
Curated, Aggregated
Commissioned, Aggregated
Distributed
User-generated
Audience
Healthcare Professionals
General public
General public
General public
Stage
Pre-Internet
Dot-com era
Beta
Pre-alpha or alpha
Revenue
Gov’t-sponsored, Subscription, Advertising
Gov’t-sponsored, Advertising, Services portal
Advertising
Advertising Services?
Examples
National Library of Medicine http://www.nlm.nih.gov/ Reed-Elsevier http://www.elsevier.com/wps/ find/homepage.cws_home Macmillan/Nature http://us.macmillan.com/ splash/publishers/nature.html Thompson http://www.thompson. com/public/library. jsp?catHEALTHCARE
National Institutes of Health http://health.nih.gov/ Healthline http://www.healthline.com/ WebMD http://www.webmd.com/ default.htm RevolutionHealth http://www. revolutionhealth.com/
Healia http://www. healia.com/healia/ Mamma http://www. mammahealth.com/ Medstory http://www. medstory.com/ Kosmix http://www. kosmix.com/health
TauMed http://www. taumed.com/ Trusera http://www.trusera. com/corp
Sixty-six percent of consumer HIS/HIR begin with a general search engine such as Google or Yahoo! (Fox, 2006), and the rest will start with one of a number of emerging or established specialty sites that are surveyed in Table 21.1. These sites fall into four general categories (online libraries, e-magazines, search and social networking) that largely reflect their historical origins and/ or stages of development. Although a complete description of the origins and development of these sites is beyond the scope of this chapter, one pioneering organization deserves special mention. Online medical libraries date back to mid-1960s, when the US National Library of Medicine (NLM) experimented with teletypewriter machines and satellite communications (Schoolman and Lindberg, 1988). Searches at this time were not interactive but rather conducted by “batch processing.” In the early 1970s, interactive searches of NLM’s MEDLINE became possible through dedicated remote terminals in 10 regional and 14 large academic medical libraries (Atlas, 2000). During this era, users of these systems were trained librarians who learned specific query commands and retrieval protocols. It was not until after the advent of personal computers that online search of the medical literature became accessible to consumers. In 1986 a computer application named “Grateful Med,” that ran on IBM-compatible DOS computers was produced by NLM, was distributed at a cost of $9.95 through the National Technical Information Service of the US Department of Commerce (Schoolman and Lindberg, 1988). Using Grateful Med and a 300 bits-per-second modem, users could use dial-up telephone lines to NLM computers and receive literature citations. The current descendents of these pioneering efforts are PubMed (Wheeler et al., 2007) for the professional literature and MedlinePlus (Miller et al., 2004) for consumer information. The Dot-com era from 1995 to 2001 (2007) was characterized by experimentation with new sites and services, including
commercial entities supported by sponsored advertising. Compared with online “libraries,” which maintained content and user interfaces more oriented toward health care professions than consumers, Dot-com era sites introduced magazine-type articles and layouts focused on non-professional users, enabled by the development of web browsers such as NCSA Mosaic (later named Netscape Navigator (2007)). Somewhat prior to this, sites providing comprehensive (“horizontal”) Internet search were mainly being used by academic researchers and computer professionals (e.g., (Boguski and Ouellette, 1995)). But gradually, at first, search engines of the time (e.g., AltaVista (2007)) began to index content that expanded into the consumer realm. The ascendance of horizontal Internet search by companies like Yahoo! (2007) and Google (2007) has recently led to the emergence of several startup companies applying a similar approach to deep vertical (specialized) health care content on the web (Table 21.1). The latest experimentation with consumer-oriented, medically-oriented websites involves social-networking, following the example of more horizontal (although demographically differentiated) socialnetworking communities and sites (e.g., MySpace (2007) and Facebook (2007)). In July 14, 2007 article in the Wall Street Journal, Borzo reported running a variety of searches on several vertical health sites and noted that many features were not self-evident and required repeated experimentation to uncover. Borzo’s overall conclusion from this limited, non-scientific study was that conducting a useful search requires the consumer to run multiple queries on several sites and to then compare the results and reach their own consensus on the adequacy and usefulness of the retrieved information. These observations are consistent with more formal research (Fox, 2006). Borzo did not examine, however, the underlying sourcing and quality of the information.When one examines
Personalized Genomics for Consumers
TABLE 21.2 Search Sites
■
255
Survey* of Content of Selected “Vertical” Health Information
Site
Primary content
Healia www.healia.com/healia/
www.pubmed.gov/ www.clinicaltrials.gov/
Mamma www.mammahealth.com/
www.emedicinehealth.com/ www.healthatoz.com/ www.mayoclinic.com/ www.medem.com/ www.medicinenet.net/ medlineplus.gov/ www.nhsdirect.nhs.uk/ www.webmd.com/
Medstory www.medstory.com/
http://www.breastcancer.org/ http://online.wsj.com/public/us
Kosmix www.kosmix.com/health
www.nlm.nih.gov www.americanheart.org www.cdc.gov www.mayoclinic.com health.nih.gov www.cancer.org www.revolutionhealth.com www.womentowomen.com www.4women.gov www.healthline.com
*
Survey date was November 16, 2007
this aspect, a perplexing assortment of both public and propriety information from both consumer-oriented sources as well as sources designed for medical professionals is revealed (Table 21.2). Only a tiny fraction of Internet health information sites publish any sourcing and date-stamped information or other information quality indicators (Anonymous, 2007). Published guidelines for quality assurance and quality control are rare and consumers are left to evaluate the quality of the information based on whether or not they consider the site to be a trusted “brand.” Given this situation, and the difficulty non-professionals have in constructing effective queries (see previous section), the best advice on Internet HIS/HIR that one can give to the consumer at the present time is to validate their findings with a health care professional. The US National Library of Medicine provides information for consumers to finding health information and includes a “Guide to Healthy Web Surfing” at their MedlinePlus site (Table 21.1).
PERSONALIZED GENOMICS FOR CONSUMERS Personalized genomics involving DNA polymorphism scans (e.g., see (Weber, 2006)) is currently in a tumultuous gestational
stage that will ultimately lead to a new way to teach and practice medicine (Childs et al., 2005) (see Chapter 1). Consider the following examples: ●
●
●
●
●
A father uses commodity DNA sequencing technologies and publicly-available medical databases to investigate the elusive cause of his daughter’s genetic illness (Maher, 2007). A controversial scientist publishes his autobiography which is partially based on the complete sequence of his own genome (Venter, 2007). A Harvard geneticist launches a Personal Genome Project (Table 21.3) to encourage medical altruism and self-knowledge, makes his own biological material available for study and encourages others to volunteer to do likewise (Church, 2005). Several start-up companies (one of which is backed by Google) form to provide genotyping services as direct-toconsumer businesses (Winslow, 2007) and Table 21.3. The cover story of WIRED magazine asserts that “a new $1000 DNA test can tell how you’ll live – and die” (Goetz, 2007).
The ethical, legal and social issues arising in this milieu certainly signify a brave new world well beyond the scope of this chapter. But what are the medical issues? Kohane and colleagues (Kohane et al., 2006) consider the situation “a threat to genomic
256
CHAPTER 21
TABLE 21.3
■
Online Health Information Retrieval
Selected Sites of Interest for Personal Genomics*
Site
Services Offered
Cost
“a forum for those searching for explanations and the help of the interested community of geneticists, patients, physicians, scientists and family members”
Not applicable
“focuses on the practical issues of recruiting and informing volunteers… a test bed for personalized medicine and new ways of interfacing with the research subjects”
Not applicable
“personal insight into ancestry, genealogy, and inherited traits”
Not available
Navigenics www.navigenics.com
“a personalized genetic analysis that, combined with relevant health and wellness information, enables a far more personalized health strategy for each individual”
$2500
deCODE genetics www.decode.com
“Subscribers … can take their genome and examine it in the context of the literature”
$985 (Introductory, promotional price)
Not-for-Profit www.MyDaughtersDNA.org
Personal Genome Project www.pgen.us Commercial Businesses 23andMe Personal Genome Service www.23andme.com
*Information current as of November 16, 2007.
medicine” because the volume and complexity of genotypes and their statistical, phenotypic associations will undoubtedly lead to a plethora of incidental, “abnormal” findings that will be pursued at great cost but little benefit by the patients and their physicians. A person’s genotype will become an “incidentalome” – analogous to the incidentalomas (Mirilas and Skandalakis, 2002) recognized by a previous generation of physicians but unbelievably more complex. Kohane and colleagues call for several key actions including the creation of information systems for “estimating and explaining the risks associated with various incidental genomic findings” (Kohane et al., 2006). They envisioned such systems as being used by medical professionals “…in the clinic and at the bedside.” However, it is likely that consumer versions will lead the way as patients “surf the web” trying to understand the implications of their genotypes and manage their personal health accordingly. Dr. Google has office hours 24/7.
SUMMARY AND CONCLUSIONS Major challenges exist in both educating and assisting consumers with HIS/HIR queries, assessing the sources and quality of the information they find, and helping them decide how to act on it. Much more work is needed on the theory and practice of health information search and retrieval by consumers as health care systems become more patient-centric and consumers are expected to make informed choices and exert more control over the management of their personal health. Health care professionals need to become familiar with the challenges consumers face. Health care systems will need to devise a mechanism for joint consumerprovider review of information, and decision support for the development of personalized health care. Personal genomes are an extreme example of the challenges we all face.
REFERENCES Anonymous. (2007a). CDC Wonder DATA2010: Focus Area 11-Health Communication, from http://www.wonder.cdc.gov/data2010/. Anonymous (2007b). Harris Poll Shows Number of “Cyberchondriacs” – Adults Who Have Ever Gone Online for Health Information – Increases to an Estimated 160 Million Nationwide. Harris Interactive. Atlas, M.C. (2000). The rise and fall of the medical mediated searcher. Bull Med Libr Assoc 88(1), 26–35. http://www.ncbi.nlm.nih.gov/ entrez/query.fcgi?cmdRetrieve&dbPubMed&doptCitation &list_uids10658961 Boguski, M.S. and Ouellette, B.F.S. (1995). Internet Basics for Biologists. Current Protocols in Molecular Biology. John Wiley & Sons, New York. Childs, B., Wiener, C. and Valle, D. (2005). A science of the individual: Implications for a medical school curriculum. Annu Rev Genomics
Hum Genet 6, 313–330. http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?cmdRetrieve&dbPubMed&doptCitation&list_ uids16124864 Church, G.M. (2005). The personal genome project. Mol Syst Biol 1, 0030. http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmdRet rieve&dbPubMed&doptCitation&list_uids16729065 Crocco, A.G., Villasis-Keever, M. and Jadad, A.R. (2002). Analysis of cases of harm associated with use of health information on the internet. JAMA 287(21): 2869–2871. http://www.ncbi.nlm.nih. gov/pubmed/12038937?ordinalpos1&itoolEntrezSystem2. PEntrez.Pubmed.Pubmed_ResultsPanel.Pubmed_RVDocSum Fox, S. (2006). Online Health Search 2006. Washington, DC, Pew Internet & American Life Project.
Wikipedia References
Fox, S. and Fallows, D. (2003). Internet Health Resources. Washington, DC, Pew Internet & American Life Project. Fredin, E.S. and Prabu, D. (1998). Browsing and the hypermedia interaction cycle a model of self-efficacy and goal dynamics. J Mass Comm Quarterly 75(1), 35–54. http://www.eric.ed.gov/ERICWebPortal/ custom/portlets/recordDetails/detailmini.jsp?_nfpb true &_ &ERICExtSearch_SearchValue_0EJ569971&ERICExtSearch_ SearchType_0no&accnoEJ569971 Goetz, T. (2007).Your life decoded. Wired 283, 256–265. Jones, G. (2003). Prescribing and taking medicines. BMJ 327(7419), 819. http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmdRetrieve& dbPubMed&doptCitation&list_uids14551062 Kohane, I.S., Masys, D.R. and Altman, R.B. (2006). The incidentalome a threat to genomic medicine. JAMA 296(2), 212–215. http://www. ncbi.nlm.nih.gov/entrez/query.fcgi?cmdRetrieve&dbPubMe d&doptCitation&list_uids16835427 Levy, M. (2007). Assessing consumers demand for health vertical search engines. Jupiter Research Industry Focus Health. Lorence, D.P. and Spink, A. (2004). Semantics and the medical web a review of barriers and breakthroughs in effective healthcare query. Health Info Libr J 21(2), 109–116. http://www.ncbi.nlm.nih.gov/ entrez/query.fcgi?cmdRetrieve&dbPubMed&doptCitation &list_uids15191602 Maher, B. (2007). His Daughter’s DNA. Nature 449(7164), 772–776. McCray, A.T. (2005). Promoting health literacy. J Am Med Inform Assoc 12(2), 152–163. http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?cmdRetrieve&dbPubMed&doptCitation&list_ uids15561782 Miller, N., Tyler, R.J. and Backus, J.E.B. (2004). MedlinePlus® the National Library of Medicine® brings quality information to health consumers. Library Trends 53(2), 375–388. http://hdl.handle. net/2142/1735 Mirilas, P. and Skandalakis, J.E. (2002). Benign anatomical mistakes incidentaloma. Am Surg 68(11), 1026–1028. http://www.ncbi.nlm. nih.gov/entrez/query.fcgi?cmdRetrieve&dbPubMed&dopt Citation&list_uids12455801 Payer, L. (1996). Medicine and Culture. Holt Paperbacks. Plovnick, R.M. and Zeng, Q.T. (2004). Reformulation of consumer health queries with professional terminology a pilot study. J Med Internet Res 6(3), e27. http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?cmdRetrieve&dbPubMed&doptCitation&list_ uids15471753 Romacker, M., Grandjean, N., Parisot, P., Kreim, O., Cronenberger, D., Vachon, T., Peitsch, M.C. (2006). The UltraLink an expert system for contextual hyperlinking in knowledge management. In
257
Computer Applications in Pharmaceutical Research and Development (S. Ekins, ed.), Wiley Interscience, 729–753. Schoolman, H.M. and Lindberg, D.A.B. (1988). The information age in concept and practice at the National Library of Medicine. Annals of the American Academy of Political and Social Science 495, 117–126. Shortliffe, E.H., Perreault, L.E., Wiederhold, G. and Fagan, L.M. (2003). Medical Informatics Computer Applications in Health Care and Biomedicine. Springer, New York. Spasic, I., Ananiadou, S., McNaught, J. and Kumar, A. (2005). Text mining and ontologies in biomedicine making sense of raw text. Brief Bioinform 6(3), 239–251. http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?cmdRetrieve&dbPubMed&doptCitation&list_ uids16212772 Tse, T. and Soergel, D. (2003). Exploring medical expressions used by consumers and the media an emerging view of consumer health vocabularies. AMIA Annu Symp Proc 674-8. http://www.ncbi.nlm. nih.gov/entrez/query.fcgi?cmdRetrieve&dbPubMed&dopt Citation&list_uids14728258 Venter, J.C. (2007). A Life Decoded: My Genome My Life.Viking, New York. Weber, J.L. (2006). Clinical applications of Genome Polymorphism Scans. Biol Direct 1, 16. http://www.ncbi.nlm.nih.gov/entrez/ query.fcgi?cmdRetrieve&dbPubMed&doptCitation&list_ uids16756678 Wheeler, D.L., Barrett, T., Benson, D.A., Bryant, S.H., Canese, K., Chetvernin, V., Church, D.M., DiCuccio, M., Edgar, R., Federhen, S. et al. (2007). Database resources of the National Center for Biotechnology Information. Nucleic Acids Res 35(Database issue), D5–D12. http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmd Retrieve&dbPubMed&doptCitation&list_uids17170002 Winslow, R. (2007). Is there a heart attack in your future. New York Times. Zeng, Q., Kogan, S., Ash, N., Greenes, R.A. and Boxwala, A.A. (2002). Characteristics of consumer terminology for health information retrieval. Methods Inf Med 41(4), 289–298. http://www.ncbi.nlm. nih.gov/entrez/query.fcgi?cmdRetrieve&dbPubMed&dopt Citation&list_uids12425240 Zeng, Q.T., Crowell, J., Plovnick, R.M., Kim, E., Ngo, L. and Dibble, E. (2006). Assisting consumer health information retrieval with query recommendations. J Am Med Inform Assoc 13(1), 80–90. http:// www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmdRetrieve&db PubMed&doptCitation&list_uids16221944 Zeng, Q.T. and Tse, T. (2006). Exploring and developing consumer health vocabularies. J Am Med Inform Assoc 13(1), 24–29. http:// www.ncbi.nlm.nih.gov/entrez/query.fcgi?cmdRetrieve&db PubMed&doptCitation&list_uids16221948
WIKIPEDIA REFERENCES (2007). AltaVista. Wikipedia. (2007). Dot-com bubble. Wikipedia. (2007). Facebook. Wikipedia. (2007). Google. Wikipedia.
■
(2007). Mosaic (web browser). Wikipedia. (2007). MySpace. Wikipedia. (2007).Yahoo! Wikipedia.
This page intentionally left blank
PART
Two
Translational Approaches in Genomic and Personalized Medicine
259
This page intentionally left blank
Section
Enabling Strategies in the Translation of Genomics into Medicine
22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32.
4
Translational Genomics: From Discovery to Clinical Practice Principles of Study Design Biobanking in the Post-Genome Era Application of Biomarkers in Human Population Studies Validation of Candidate Protein Biomarkers Pharmacogenetics and Pharmacogenomics The Role of Genomics and Genetics in Drug Discovery and Development Role of Pharmacogenomics in Drug Development Clinical Implementation of Translational Genomics Translating Innovation in Diagnostics: Challenges and Opportunities The Role of Genomics in Enabling Prospective Health Care
CHAPTER
22 Translational Genomics: From Discovery to Clinical Practice Geoffrey S. Ginsburg I am convinced that within five years every college-educated person in America is going to have a [whole genome] profile ... you cannot afford not having this. deCODE chief executive Kari Stefansson, PhD (Wohlsen, 2008)
INTRODUCTION The use of genomic technologies to improve medical decisionmaking so as to achieve the goal of personalized medicine is poised to transform health care (Ginsburg and McCarthy, 2001; Ginsburg et al., 2005; Snyderman and Williams, 2003; Willard et al., 2005). New knowledge from the study of genomes and their by-products now has permitted the development of predictors of disease predisposition (“who is at risk”), prognosis (“who to treat”) and therapeutic response (“how to treat”) in individual patients and thus provides an opportunity to employ these predictors in present day practice. Many of these opportunities are highlighted in Chapters 54–112, in which an astounding number of genome-based discoveries are highlighted with the potential to fundamentally alter how medicine is practiced in cardiovascular disease, oncology, metabolic disease, neuropsychiatric disorders, infectious disease and diseases with a basic inflammatory component. The rate of genome-based discovery is stunning. However, despite the vast amount of genomic information now in the Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 262
scientific literature, the translation of basic scientific findings from the genome and its derivative products, RNA and protein, to daily use in medical practice has been slow (Graham et al., 2006; Sung et al., 2003). It now takes an average of 17 years for 14% of scientific discoveries to enter day-to-day practice (Westfall et al., 2007). For genomic discoveries, the path is undoubtedly more complex. The research paradigm for translational genomics is inconsistent and nonsystematic; multiple, and in some cases non-linear, pathways to translation of genomics to clinical medicine exist. For example, diagnostic companies have traditionally “cherrypicked” a narrow repertoire of new genomics-based molecular diagnostics from academic laboratories and the literature. The pharmaceutical industry has developed “targeted therapies” such as trastuzamab, imatinib and erlotinib that use molecular diagnostic tests – some of which are based on genetic variation (Lynch et al., 2004) – to identify individual patients who are likely to benefit from the drug. The lack of a clear system of regulatory oversight of genetic and genomic tests has led to gaps and ambiguity in the pathways of translation, an area that Copyright © 2009, Elsevier Inc. All rights reserved.
Where Can Genomics Have Impact in the Continuum of Health and Disease?
the US Secretary’s Advisory Committee on Genetics, Health and Society has recently addressed (SACGHS, 2007; see also Chapters 31 and 33). Personal genomics and direct-to-consumer avenues are being opened to bring genomics directly to the public, thus circumventing the traditional health care delivery and regulatory channels (Feero et al., 2008; Hunter et al., 2008; Janssens et al., 2008; McGuire et al., 2007). These issues speak to the need for a comprehensive agenda for translational research to move human genome discoveries into health practice. This chapter will focus on paradigms for translational genomics to human health as well as the various opportunities, challenges, and strategies that will enable the effective translation of human genome information to clinically relevant actions and outcomes.
A ROADMAP FOR TRANSLATION Khoury and colleagues (2007) have developed a framework for the acceleration of human genome discoveries into health care and disease prevention. This is a logical first step in guiding genomic innovation to health care and more widespread use. The framework (Table 22.1) contemplates four phases of translational research from discovery to public health impact. T1 research seeks to move a basic genome-based discovery into a
T A B L E 2 2 . 1 The continuum of translation research in genomic medicine: Types of research Translation research phase
Notation
Types of research
T1
Discovery to candidate health application
Phases I and II clinical trials; observational studies
T2
Health application to evidence-based practice guidelines
Phase III clinical trials; observational studies; evidence synthesis and guidelines development
T3
Practice guidelines to health practice
Dissemination research; implementation research; diffusion research; Phase IV clinical trials
T4
Practice to population health impact
Outcomes research (included many disciplines); population monitoring of morbidity, mortality, benefits and risk
Adapted from Khoury et al. (2007).
■
263
candidate health application (e.g., the association of a gene mutation with a health outcome). This phase of research starts after gene discovery and provides the early evidence that genomic information might support predictive testing, screening, diagnostic, prognostic or pharmacogenomic testing. T1 research will usually take advantage of observational studies or clinical trials where the correlation of the genomic information with health characteristics or outcome can first be made. T2 research assesses the value of a genomic application for health practice, leading to the development of evidence-based guidelines (e.g., prospective or retrospective assessment of the predictive value of the mutation test for the health outcome). This phase of research usually begins after the analytical validity of the test has been established and progresses to establish clinical utility, or impact on clinical decision-making. T3 research examines the pathways for delivery, dissemination, and diffusion of genomic testing into practice (e.g., defining the barriers to the use of the test in relevant populations). Public–private partnerships will usually be required to facilitate T3 research, and this research phase will also involve policy makers, regulatory bodies, professional organizations and insurers. Finally, T4 research assesses the impact on human health (e.g., outcomes studies across a broad and diverse population). This area of research is about the results of the implementation, as opposed to the implementation research itself (T3). T1 research is illustrated in many of the clinical chapters in this book that highlight genomic discoveries with potential disease relevance. The molecular signatures described by Dave et al. (2006), Bullinger and Valk (2005) and Potti et al. (2006) to stratify and number of cancer phenotypes based on their molecular signature are illustrative. Examples of T2 research are found in Marcom et al. (2008) and Potti et al. (2006), in which a genomic-guided clinical trial design is described that uses a microarray-based molecular profile to assign patients to the various treatment arms in the trial. The clinical outcomes in the trial are dependent on the genomic signatures and the evidence for the genomic application to humans will be determined by these prospective studies. Examples of T3 research are described in Khoury et al. (2007) as surveys of physicians, internists and oncologists on the knowledge of genetic testing for susceptibility for breast and ovarian cancers. This information would provide support for the basis of strategies to enhance the dissemination of innovative genomic tools and information. T4 research is ongoing in areas such as BRCA1 testing for hereditary breast and ovarian cancers or for HNPCC testing in colon cancer (see Chapters 72 and 73). In these early cases, the value of such testing is being assessed on a population level in terms of survival, quality of life and economic value. We are clearly in the early days of T2–T4 research.
WHERE CAN GENOMICS HAVE IMPACT IN THE CONTINUUM OF HEALTH AND DISEASE? DNA-based approaches (SNP detection, copy number variation and sequencing) (see Chapters 7–9) can now be used to
264
CHAPTER 22
■
Translational Genomics: From Discovery to Clinical Practice
Disease burden
quantify one’s predisposition, susceptibility, and risk for complex diseases. These measures can be made during health and even at birth or at any time point over a person’s lifetime, as these “stable” genomic measures do not change from the time of conception. Individual SNPs or multi-SNP panels are emerging from genome-wide approaches that might be used as part of health risk assessment (see e.g.: Bare et al., 2007; Zheng et al., 2008). DNA variation may also provide information about the possibility of being relatively protected from disease development (Cox et al., 2007), as well as one’s sensitivity or resistance to certain medications (Eichelbaum et al., 2006) and ability to metabolize nutrients in our diets (Ordovas and Corella, 2004). All of this can and probably should be done early in life, such that a strategy to maintain health can be formulated well in advance of the development of potentially detrimental lifestyle habits and exposures. The “dynamic” components of the genome (gene-, protein- and metabolite expression; see Chapter 1) that are responsive to environmental stimuli, lifestyles, diets and pathogens are rapidly improving capabilities to predict and intervene at an individual level. Transcriptional profiles, protein expression and levels of metabolites combined with dynamic imaging modalities should provide more precise ways to screen individuals who are at high risk for developing a disease for its earliest molecular manifestations while the disease is subclinical (Seo and Ginsburg, 2005).This same type of information may also provide a definitive diagnosis and a molecular classification of a disease state that foretells prognosis. For example, today a her2/neu-positive breast cancer ascribes that patient to a more aggressive form of the disease and directs care to a much different course than a her2/neu negative cancer
Baseline risk Initiating events
(see Chapter 72). Similarly the selection of drugs can be guided both by the patients underlying genetic make up and by the molecular architecture of the disease in the individual (see Chapters 27, 28 and 30). The emerging picture is one of applied genomics throughout an individual’s lifetime, from health through death, that enables assessment of disease predisposition, screening, diagnosis, prognosis, therapeutic selection with precision that until now has been unachievable (Figure 22.1). “Strategic health planning” (see Chapter 32 and Snyderman and Yoediono, 2006) and disease prevention should be possible and will allow a shift in the current paradigm of care from the time when disease is manifest (rightward in Figure 22.1) to proactive personalized predictive care at a more cost-effective time in the disease life cycle (leftward in Figure 22.1).
THE GENOMICS “GOLD RUSH” While genetic testing for Mendelian disorders such as cystic fibrosis, Huntington’s disease, familial breast cancer and phenylketonuria among others have been widely available prior to the genomic era, the genetic basis for complex disease remains unclear. From 1980 through 2002, fewer than 10 genes were associated with complex diseases in contrast to more than 90 genes that were associated with Mendelian disorders (Glazier et al., 2002). Today we are witnessing a “genomics gold rush” afforded by genome-wide association studies using high-density genotyping technologies that allow for assaying 500,000–1,000,000 SNPs per individual at relatively low cost
Earliest molecular detection
Typical Earliest current clinical detection intervention
Time Baseline risk Sources of New Stable Genomics: Biomarkers: SNPs
Preclinical progression
Disease initiation and progression
Dynamic Genomics: Gene expression
Haplotypes
Proteomics
Gene sequencing
Metabolomics MolecularImaging
Figure 22.1 Use of genomic markers predict, prognose, diagnose, treat and monitor health and disease (adapted from Snyderman and Yoediono, 2006).
The Genomics “Gold Rush”
(Topol et al., 2007) (see Chapter 8). The results in the past year have been no less than astounding (Topol et al., 2007), with genetic loci being identified for many complex diseases on almost a weekly basis (Table 22.2). T A B L E 2 2 . 2 Selected genome-wide association study findings for complex disease susceptibility: 2007–2008 Disease
Gene or loci (references)
Cancer Prostate cancer
TCF2 (Gudmundsson et al., 2007a)
Prostate cancer
8q24 (Gudmundsson et al., 2007b)
Multiple solid tumors
CASP8 (Bethke et al., 2008; Cox et al., 2007)
Colon
11q23, 8q24, 18q21 ( Tenesa et al., 2008)
Breast cancer
FGFR2, TNCR9, MPA3K1, LSP (Easton et al., 2007; Hunter et al., 2007)
Metabolic disease Obesity
FTO (Frayling et al., 2007)
Diabetes, Type I
IL2RA (Lowe et al., 2007)
Diabetes, Type II
TCF7L2 (Saxena et al., 2007; Sladek et al., 2007)
Diabetes, Type II
WFS1 (Sandhu et al., 2007)
Diabetes, Type II
CDKAL1 (Steinthorsdottir et al., 2007)
Cardiovascular disease Myocardial infarction, CAD
9p21 (Helgadotti et al., 2007; McPherson et al., 2007)
Atrial fibrillation
4q25 (Gudbjartsson et al., 2007)
Coronary artery disease
6q25, 2q36 (Samani et al., 2007)
Inflammation/infectious disease Rheumatoid arthritis
TRAF1-C5, STAT-4 (Plenge et al., 2007; Remmers et al., 2007)
Multiple sclerosis
IL7Ra; IL2Ra (Gregory et al., 2007)
Crohn’s disease
IRGM (Parkes et al., 2007)
HIV host control
HLA-B*5701 (Fellay et al., 2007)
Celiac disease
IL-2, IL-21 ( Van Heel et al., 2007)
Asthma (childhood)
ORMDL3 (Moffatt et al., 2007)
CNS disease Bipolar disorder
16p12 (Sklar et al., 2007)
Glaucoma
LOXL1 (Ma et al., 2007)
Age-related macular degeneration
C3 ( Yates et al., 2007)
■
265
Thus far, notwithstanding significant challenges (see Chapter 8), genome-wide association studies have produced robust and reproducible findings. For example, three genomewide studies for coronary artery disease have been published using three different populations and have all identified a locus at 9p21 (Helgadotti et al., 2007; McPherson et al., 2007). Type 2 diabetes has been extensively studied and independent genomewide scans have identified several loci for diabetes susceptibility: CDKN2A/CDKN2B, CDKAL1 and IGSF2BP2 as well as confirming TCF7L2, PPARG and KCNJ11 that had been previously identified via other methods (Saxena et al., 2007; Scott et al., 2007; Sladek et al., 2007; Steinthorsdottir et al., 2007; Zeggini et al., 2007, 2008). A susceptibility gene for obesity, FTO, was identified by the same groups studying type 2 diabetes (Zeggini et al., 2007). Genes for Crohn’s disease, rheumatoid arthritis, adult macular degeneration and prostate cancer have all been identified in the past year using genome-wide approaches (Gudmundsson et al., 2007a, b; Klein et al., 2005; Plenge et al., 2007; Rioux et al., 2007). Perhaps the most important and powerful result of these studies has been their unbiased approach and their resulting insights into novel disease genes and molecular pathways that contribute mechanistically to these complex diseases, their etiologies and the basis for their progression and clinical manifestations. As a result of the above types of studies, large numbers of additional susceptibility markers are sure to emerge in the coming years. Whether these markers will be applied to clinical practice remains uncertain; their clinical utility must first be demonstrated – that is, the impact on health outcomes of the clinical use of these genetic markers. These findings raise a series of questions that must be evaluated; for example, do these genetic markers add additional estimates of disease risk compared to current clinical models? If combined, do multiple SNPs, each with only a small predictive value, used in concert with one another provide enough power to be clinically significant? What actionable options are there for patients with these results and how do these change outcomes compared to the current standard of care? The recent study by Zheng and colleagues (2008), who examined the association between prostate cancer and five SNPs that map to the three 8q24 loci, to 17q12, and to 17q24.3, is emblematic of these issues. When four or five high-risk genotypes were present, they were associated with a composite risk ratio for prostate cancer of 4.47. When family history was added to the model, the risk ratio was an astounding 9.46. So what are these results telling us? They certainly point to the multifactorial underpinnings of prostate cancer. But as was pointed out in the accompanying editorial to this report (Gelmann, 2008): “What we cannot yet derive from these studies is an easily applicable test that will assist in the identification of men who are at high risk for prostate cancer. Most men with a family history of prostate cancer are aware that screening is recommended by some groups as early as the age of 40 years. Genetic tests are unlikely to alter the screening behavior of men who are already under surveillance for prostate cancer by the monitoring of prostatespecific antigen.” So even in cases of very high risk for common
266
CHAPTER 22
■
Translational Genomics: From Discovery to Clinical Practice
complex disease on the basis of genetic prediction, the clinical applications are at best ambiguous.
THE PERSONAL GENOME: PRECIOUS CODE OR FOOL’S GOLD? The same advances in SNP technology that are providing insightful gene association data also are allowing a glimpse variation in our personal genomes and at low cost. Direct-to-consumer initiatives to make variation in one’s genome available in the form of SNPs at specific loci for $1000–$2500 have been recently announced by three companies (Navigenics, 23andMe and deCODEme). These initiatives use the 500K–1000K SNP chip or similar technologies (see Chapter 8). They report information back to their customers on 20–30 areas of the genome that have been identified as disease risk susceptibility loci and some offer additional information such as one’s ancestry (see Chapter 1). This may be a “disruptive technology” in health care delivery with the provision of health and disease risk information to consumers without physician intervention and guidance. It will not be long before a patient will bring a report of a whole genome to a physician’s office and ask for guidance. What will the physicians of today, armed with a paucity of genomic training, tell them? Whether our genomes are “SNP’d” or sequenced, our genomic information will be available in a short period of time. There is a clear and important research agenda that needs to be developed in concert with these technological breakthroughs that allows health providers and the public to understand the information and more so to believe that is accurate, informative and actionable (Feero et al., 2008; Hunter et al., 2008; McGuire et al., 2007). With the vast amount of information contained in the human genome sequence, the stakes are high for patients, physicians and the public to ensure the proper reading, interpretation and communication of the information are carried out.
“GRAND CHALLENGES” IN TRANSLATION OF GENOMICS TO HUMAN HEALTH Despite the clear advances in technology to bring genomic information closer to physicians, patients and the public, looming ever closer are issues that are outside of the sphere of the scientists that have been involved in the discovery and early translational activities. In the United States, the Institute of Medicine has convened a Roundtable on Translating Genomics to Health (IOM, 2007), the Centers for Disease Control has independently developed a pilot project on the Evaluation of Genomic Applications in Practice and Prevention (EGAPP, 2004) and the Genomic and Personalized Medicine Act of 2007 (Personalized Medicine, 2007) was introduced in Congress to also address similar issues.
Recently Scheuner and colleagues (2008) carried out an extensive meta-analysis of studies using genomics toward clinical application in chronic disease. This study aimed to understand the current state of translation focusing on the following questions: “what are the outcomes of genomic medicine? What is the current level of consumer understanding about genomic medicine and what information do consumers need before they seek services? How is genomic medicine best delivered? What are the challenges and barriers to integrating genomic medicine into clinical practice?” Using a total of 68 articles in their analysis, the authors synthesized information on the delivery of genomic medicine for common adult-onset conditions (Scheuner et al., 2008). The major findings of the study were not surprising in terms of the genome policy issues that have been previously summarized (Haga and Willard, 2006): ●
●
●
●
Education – The primary care workforce feels woefully unprepared to integrate genomics into regular practice. Consumers are also unclear about genetics and genetic testing for common diseases. Privacy – Consumers are worried about the possible adverse consequences of genetic testing, particularly the privacy issues and discrimination against receiving employment and health insurance. Evidence – There needs to be outcome data for genetic testing and chronic disease to assess whether patients who receive the test do better clinically. Cost – Cost uncertainty (both in terms of delivery and reimbursement) is an important issue to many of the stakeholders of genomic issues.
Health Professional and Public Education Education of health professionals and the public will be essential to advancing the use of genomics into health care (see Chapter 34 and Frueh et al., 2005). With all of the rapid advancements in genomics research and technologies, it will be challenging to keep health professionals informed about the benefits, risks and limitations of new tools as they become available. In addition, the public and health care workforce will need to understand the appropriate applications of genomic tools – including their benefits, risks and limitations, how they may improve clinical management, inherited versus acquired genomic variations (e.g., implications for family members) and privacy and confidentiality. Although several surveys have documented the below average physician knowledge of genetics (Metcalfe et al., 2002; Wilkins-Haug et al., 2000), none has assessed knowledge of the newer field of genomics. But several papers have been published recognizing the importance of pharmacogenetics (Frueh and Gurwitz, 2004; Gurwitz et al., 2003), and steps are underway to develop more educational materials in this area. Privacy Fears There has been ongoing debate about the uniqueness of genetic information and whether it warrants special protections beyond
Translational Genomics: Enabling Competencies
those in place for standard medical information (see Chapter 33 and Haga and Willard, 2006). In the United States, fear of discrimination by employers and health insurers is the main concern, whereas in the United Kingdom, use of genetic information by life insurers is the major concern (Apse et al., 2004; Hall et al., 2005). Despite the outcome of these debates, the attention paid to genetics by the popular press and public has raised genetic information to a different level compared to other medical information. The potential for genetic discrimination has been a major concern for researchers, health professionals, patients and the public. In order for genomic biomarkers to be integrated into routine clinical practice, associated fears with this type of testing must be put to rest. While the majority of states in the United States have enacted legislation to protect against genetic discrimination by employers and health insurers, national protections are only now being addressed at the federal level. Building the Evidence for Clinical Utility Perhaps the most important factor hindering the appropriate integration of genomics into clinical practice is the lack of evidence for its clinical utility (i.e., evidence that use of a genomic technology improves health outcomes) (see Chapters 30 and 31). Evidence generation needs to be more practicable and practical. There is also a need for greater collaboration among stakeholder groups and for innovation in both study design and analysis methods. Clinical outcome studies are needed that demonstrate the clinical utility of genomic interventions that are linked to specific, actionable clinical recommendations or practice guidelines. Public–private partnerships are likely to be required to generate the evidence base for genomic medicine. These collaborations are desirable because no single stakeholder group is likely to have sufficient resources or expertise to conduct the necessary studies. Cost Issues As with any new innovation, genomic testing must be demonstrated to be clinically useful and cost-effective and of value (see Chapter 36). But because genomic technologies inherently involve diagnostic or prognostic testing, in addition to the complexities of incomplete gene penetrance and multiple gene and environmental interactions, their assessment can be more challenging. In addition, perhaps more than in any other area of medicine, questions have arisen in regard to the economic incentives to develop these technologies. Formal health economics frameworks can be used to gain insights into these issues and provide guidance for research and development investment, technology appraisal and policy development (Carlson et al., 2005; Flowers and Veenstra, 2004; Stallings et al., 2006). It is important to examine the drivers of cost-effectiveness of genomic technologies and to consider approaches that include value-based reimbursement for genomic testing technologies. A particularly challenging area is pharmacogenomics, in which the economic incentives for developing diagnostics linked to therapeutics in the pharmaceutical industry are unclear. An integrated business model is needed that will be favorable for the effective delivery of genomic information to patients and clinicians.
■
267
TRANSLATIONAL GENOMICS: ENABLING COMPETENCIES The pharmaceutical industry has a clearly defined and decisiongated series of steps to move a potential therapeutic from discovery to the clinic. Translational genomics may also benefit from defining the processes, the key competencies, and capabilities that support the directional movement of genomics from discovery to human health applications (Khoury et al., 2007). Figure 22.2 is a potential (and admittedly simplified) schema for moving genomic discoveries from bench to bedside from the perspective of needed and enabling capabilities and competencies. Academic centers, industry and government research groups right benefit from developing organizing principles to move a potential genomic discovery into practice through a variety of stages similar to what the pharmaceutical industry has done for drug development candidates. Below and in the ensuing chapters, we will consider these capabilities in greater detail. Centralized Biobanking and Biorepositories Biobanks – centralized, institutional repositories for biological samples – are among the most important enabling resources in support of genomic medicine (Ginsburg et al., 2008). The study of complex diseases has been greatly facilitated by the availability of well-annotated DNA, blood and tissue specimens for high-throughput genomic analyses. Scale and scope characterize biobanks today from previous tissue repositories in Departments of Pathology or held by individual investigators. Today biobanks manage hundreds of thousands to millions of samples and require robotics and advanced informatics tools. Key capabilities for efficient and effective biobanking include centralized sample and data management systems and clearly define standards for handling of samples and their associated data. Equally important is the ability of biobanks to provide support for longitudinal studies enabling sample and data capturing in the entire disease cycle from pre-disease onset, to disease diagnosis, and clinical outcomes. In Chapter 24, Chow and Liu explores several existing and developing biobanking models and highlight developing trends in biobanking organization and management. The Singapore experience is particularly instructive in that the establishment of a national biobank was coupled with simultaneous development of the entire biomedical research infrastructure at an accelerated pace. Key considerations involve professional acceptance, custodianship, security and access. Data privacy for subjects is of paramount importance and applications of informatics technologies may provide the requisite security. Once properly implemented, biobanks may not only accelerate medical research but also contribute to public health. Genomic “Profiling” There are many sources of genomic data (see Chapters 7–16) that can now be applied to an integrated view of genomic epidemiology. A competency in genomic profiling technologies is clearly
268
CHAPTER 22
■
Translational Genomics: From Discovery to Clinical Practice
Surgical specimens Animal models
Biomarkers
Drug development and pharmacogenomics
Novel targets
Registries
Biorepository
– Sample management – Data management – Standards – Centralization
Profiling
– SNPs/genotyping – Transcript profiling – Proteomics – Metabolomics
Biomarker discovery and development
Informatics
– Data acquisition and storage – Computational biology – Data standards – Novel methods – Statistical modeling
– Model systems – Human samples – Animal physiology – Clinical discovery – Phase 0 studies
Genomebased clinical trials
– Test platform development – Public–Private Partnerships – Translational diagnostics
Population studies Outcomes research
Implementation in health care systems
– Cost-effectiveness – Clinical decision support – Outcomes measures – Diagnostic capabilities
Technology infrastructure and development
Figure 22.2 Enabling capabilities and competencies to translation genomic-based discoveries into health care applications. Based on Willard et al., 2005.
needed to optimize the value of clinical samples and data to develop novel predictive models of disease states and outcomes. The information provided by today’s clinical and biochemical markers of disease falls well short of an adequately describing disease complexity and heterogeneity (West et al., 2006). The advent of the full complement of technologies spawned by the Human Genome Project capable of interrogating complex diseases such as cancer, cardiovascular disease, obesity and diabetes mellitus provides opportunities for acquiring quantitative data that can match the complexity of the disease. Genomic technologies can now potentially capture data that do provide this complexity, identifying discrete subsets of disease that have not been previously recognized. Profiles and patterns that identify new subclasses of tumors, such as the distinction between acute myeloid leukemia and acute myeloid leukemia (Bullinger and Valk, 2005), or Burkitt’s lymphoma from diffuse B cell lymphomas (Dave et al., 2006), without prior knowledge of the classes. More recently several genomic signatures that go beyond disease classification have been discovered and validated that predict prognosis and response to therapy for many solid tumors and hematologic malignancies (Potti et al., 2006; Staudt, 2003). Not far behind, although lacking the throughput of RNA expression analyses, are whole-proteome analyses of tissue homogenates or serum using surface-enhanced laser desorption ionization–timeof-flight or liquid chromatograph/mass spectrometry/mass spectrometer technologies (Caprioli, 2005; see Chapter 14). The capability to discern structure in this data – in the form of patterns of gene, protein, or metabolite expression that provide snapshots of gene activity in a cell or tissue sample at a given instant of time that can then be used to describe a phenotype – is
transforming biology from an observational molecular science to a data-intensive quantitative genomic science. The dimension and complexity of such data provide opportunities for uncovering patterns and trends that can distinguish subtle phenotypes in ways that traditional methods cannot (West et al., 2006). In an integrated view of genomic epidemiology, all of these sources of data will be used to develop signatures and models that classify disease and predict outcomes (Figure 22.3). To fully realize the clinical potential of genome-scale information requires a dramatic shift in the way complex, large-scale data are captures, viewed, analyzed and used. Data Acquisition, Standards and Bioinformatics All of the exciting medical, biological and technological advances described in this book are limited by requirements for intensive data acquisition, storage and computational capacity. Individuals with skills in this arena are in critically short supply. Informatics is a fundamental competency for the era of genomic medicine. Unless the nomenclature of the current “Tower of Babel” is improved, we will miss the opportunity for applying our newfound knowledge, because these measurements will not be compatible when attempts are made to aggregate them. Currently, most data collection efforts are relatively isolated and many different groups are working on controlled vocabularies and data standards (see Chapters 17 and 18 and Hanauer et al., 2007). In addition to nomenclature standards that allow observations to be compared across observers, the methods for exchanging data must be standardized (see Chapter 19 and Klein, 2002). Health Level Seven has essentially solved this problem for common laboratory measurements that can be measured as a
Translational Genomics: Enabling Competencies
Gene expression profiles Clinical data Treatments Family history Demographics Environmental
Signatures Models
Genomic data SNPs Genome-scale sequence Metabolomic data Proteomic data
Imaging
■
269
applied in medical research. In Chapter 25, Bonassi and colleagues address the basic issues concerning the use of genomic biomarkers in human population studies, and how they allow a better insight into early stages of the disease. In other chapters the use of biomarkers in clinical trials or as enabling diagnostics is also covered.
???
Predictions: Risk Individualized prognosis and diagnosis Drug response Environmental response
Figure 22.3 Integration of clinical, genomic, imaging, and other data to develop signatures and models to predict disease and response to drugs and environmental stimuli.
number or coded output. Unfortunately, such standards are only now emerging for clinical data, and considerable variation exists in the purposes, stated or perceived, for this effort, including better individual patient care, quality measurement, research and improved administrative efficiency and billing. An effective and unifying cross-platform system will be needed that will allow collaborations among industry, academia, the NIH and other partners (see below). There are fundamental “bioinformatics blocks” in translation of genomics that must be overcome (see Chapters 17–21). First, there is an enormous amount of data at the level of fundamental biological measurements. Second, at the level of the patient in the intact health care system, a major challenge exists in integrating disparate electronic health records. Third, there are further challenges in integrating research, clinical and demographic/societal data into a common format. And lastly, there must be ways to link the clinical and molecular information, so that analyses that yield both cross-sectional and longitudinal patterns relevant to disease sub-classification can be carried out in a reproducible fashion. Biomarker Discovery and Development The discovery and development of physiologic and disease biomarkers has provided a valuable set of tools to assess disease outcomes (see Chapters 25 and 26 and Downing, 2000; Goodsaid et al., 2008). Biomarker research has also greatly improved our knowledge and understanding of mechanisms of etiology and disease pathogenesis. This will require access to robust model systems, human samples and exquisitely curated phenotypic data. Indeed, a genomic signature that characterizes a disease subset also provides a wealth of biological information through the members of genes, gene transcripts or proteins that constitute such a signature, either as markers of disease or as target for potential therapeutics. This approach has been particularly well developed in cancer research, although the driving concept of the gene–environment interaction has been widely
Genome-Based Clinical Trials Fundamental to defining the clinical validity and clinical utility of genomic biomarkers is the genomics-based clinical trial (see e.g., Marcom et al., 2008). In this setting, genetic and genomic biomarkers are fundamental to the trial design and the selection of patients for the trial (inclusion criteria) or for assignment to a specific experimental arm of the study. The basic principles of experimental design can also be applied to genomic studies. The task is to design a study in as economical a way as possible and simultaneously to optimize the information content of the experimental data. Clinical studies that serve to enrich a trial with a particular genomic biomarker or to test that a particular genomic subset of patients respond to therapy are required to advance translational genomics. Often these studies will both “validate” the biomarker – showing that it indeed identifies the relevant group of patients – and at the same time show its clinical utility – that the patient group selected by the biomarker indeed responds differently to the intervention in the trial. Key to enabling genome-based clinical trials is the interface between discovery genomics with the platforms for developing a genomic assay that is ready for clinical use. In many cases, this will require key public–private partnerships to facilitate the transition from discovery research to human applications. As discussed in Chapter 23 by Grass, there are several open issues in the design of genomic studies that need further investigation, such as selection bias at the patient recruitment level, tissue specimen sampling procedures, assays, and time point selection; in addition lack of calibration, poor reproducibility of molecular signatures across different studies and technical variation between different chip technologies also are in need of greater study as these novel study designs are developed. Integration of Genomics into Drug Discovery and Development Currently, the pharmaceutical industry is facing challenges in terms of the low number of new chemical entities (NCEs) that are approved each year, despite soaring research and development (R&D) costs. A main cause for low pharmaceutical productivity is pipeline attrition due to failure of NCEs to demonstrate efficacy or the emergence of safety issues during or after development. In parallel to genomic-guided clinical trials, pharmaceutical studies that use biomarkers for trial enrichment or for pharmacogenomics will be another pathway for genomics to reach the clinic in the form or diagnostic/therapeutic combinations. In Chapters 28 and 29, the practice of genomics and related technologies in transforming drug discovery and development is discussed in detail. Genomics has provided sophisticated and novel new approaches for target identification, target
270
CHAPTER 22
■
Translational Genomics: From Discovery to Clinical Practice
validation, the drug screening process and the understanding of the mechanism of action of targets and drugs (Stoughton and Friend, 2005). In drug development, the process by which new drugs are tested in model organisms and in humans, genomics has provided a means to understand heterogeneous disease populations at the gene level. This has enabled the selection of subjects for clinical trials with more homogeneous disease characteristics, thereby enabling a much better assessment of drug activity. In addition, the understanding of drug metabolism on a genetic level has offered a means to provide drugs with a better safety profile for certain patient populations. Pharmacogenetics and Pharmacogenomics The impact of the genome on our ability to predict drug response is one of the most promising and fertile areas of genomic and personalized medicine (see Chapters 27 and 29). Pharmacogenetics is the study of genetic variation that ultimately gives rise to the variable responses in individuals to any given drug treatment. More recently, pharmacogenetics has provided an explanation as to why certain individuals do not respond, or respond differently to a given drug treatment. Pharmacogenomics utilizes genomic technology to understand the effects of all relevant genes on the behavior of a drug or conversely the effect of a drug on gene expression. Pharmacogenomics, like pharmacogenetics, has rapidly embraced genomic technologies to identify molecular patterns of response, drug disposition and drug targets, yielding molecular biomarkers of drug response both of which have great potential to positively affect the area of medicine. Grossman and Goldstein (Chapter 27) outline the scope of study designs for pharmacogenomics studies, depending on specific features related to the study population, underlying disease, phenotypes of interest (i.e., drug efficacy, safety and dosing) and drug compound being tested. Considerations related to the realization of a valid genetic effect into an acceptable diagnostic kit are reviewed, providing analytical tools objectively evaluating the validity, utility and cost-effectiveness, as well as ethical properties, associated with a specific drug application. Translational Diagnostics Translational diagnostics is defined in Chapter 31 as “the subfield of translational medicine concerned with diagnostic methods and information.” Many of the translational issues affecting genomics as whole also affect this area of innovation. These include communication and education of researchers, institutional support for translational research activities, the regulatory environment, cost and reimbursement issues, the need for bioinformatics and complex data handling capabilities, intellectual property and licensing protection and the willingness to embrace new paradigms for diagnostic and therapeutic strategies (Horig and Pullman, 2004; Horig et al., 2005; Humes, 2005; Marincola, 2003; Sonntag, 2005). Integration of Genomic Testing into Clinical Practice A final hurdle that genome-based medicine faces is the integration of testing into day-to-day clinical practice. Despite the
optimism that genomic testing might bring to improving the practice of medicine, there are barriers that must be overcome for their seamless integration into clinical practice. The incorporation of genetics and genomics into patient management guidelines has largely failed to occur thus far, perhaps because researchers, diagnostic firms and the regulatory authorities are still seeking to establish methodologies by which to judge their effectiveness because practicing clinicians and guideline writers are still working to understand how such new tests fit into current models of care and risk assessment, and because payers are just beginning to foresee new pressures to cover the additional costs. Indeed, population studies and outcome studies will be critical for this final phase of clinical uptake, implementation into guidelines and into reimbursement strategies. To address some of these issues, a framework has been proposed to assist in genetic testing evaluation, consisting of six areas (Table 22.3): technical and operational excellence, diagnostic capability, impact on diagnostics and/or prognostics, impact on therapeutic strategy, cost-effectiveness and health outcomes (Douglas and Ginsburg, 2008). It is clear that the physician community faces many of the same issues the payer and diagnostics groups do as well. Therefore, it makes sense to consider developing the partnerships that bring the stakeholders together to jointly address these translational obstacles.
HOW ARE WE GOING TO DO THIS? DEVELOPING ENVIRONMENTS THAT FOSTER TRANSLATIONAL GENOMICS TO HEALTH APPLICATIONS It is clear that academicians and industry scientists are on convergent paths to personalized medicine using genomic technologies and information. Specialized centers for genomic and personalized medicine in academic health systems can be instrumental in integrating, facilitating and catalyzing the needs of both the academic and industry stakeholders by providing: ●
●
●
●
●
●
Access to patients, patient data, and molecular and biological data that drive the development and exploration of genomic information and its link to clinical outcomes. The scientific foundation for novel biomarker discovery for both disease and drug response based on mechanism. An environment for innovation that fuels the development of novel translational strategies. A vehicle for aligning the efficiency and quality metrics of patient care with the goals of personalized medicine. The infrastructure for the types of public–private partnerships required for executing genomic assay-guided clinical trials, and, finally, A place to engage in a dialog and research on the key issues challenging the translation of genomics into personalized health care (Haga and Willard, 2006): education, facilitating clinical trials, information systems and integration into practice.
How Are We Going to Do This? Developing Environments that Foster Translational Genomics to Health Applications
TABLE 22.3
■
271
Framework for evaluating genetic tests*
Technical capability and operational excellence
●
Diagnostic capability
●
●
●
Impact on diagnostics and/or prognostics
● ●
Impact on therapeutic strategy
● ● ●
Cost-effectiveness
● ●
Health outcomes
● ● ●
development of analytic standards computational capabilities to acquire, store and analyze large datasets define the clinical situation(s) that may be informed by genomic data validation of genomic associations or predictors in distinct and separately ascertained cohorts development of clinical decision support platforms and “just-in–time” solutions in the practice environment linkage to authoritative knowledge sources evaluate genomic testing in the context of physicians’ confidence in decision-making develop educational strategies with test implementation assess physician behavior assessment of system wide costs resultant health care utilization and outcomes assess patient, provider, payer and government perspectives incorporation into guidelines driven by the evidence base carry out prospective randomized trials of usual care versus genomics guided care development of an educational strategy aimed at the patient
*Modified from Douglas and Ginsburg (2008).
Public private partnerships
Diagnostic Companies
Specialized centers for genomic and personalized medicine
Personalized medicine
Dx tools Genome scientists Clinicians Clinical researchers Statisticians Informaticians Assay developers Policy analysts
Rx/Dx
Academic medical centers
Health systems Genomic discoveries
Clinical implementation
Drug Companies
Figure 22.4 A framework for an organization to translate genomics to personalized medicine through public–private partnership and academic health centers.
272
CHAPTER 22
■
Translational Genomics: From Discovery to Clinical Practice
To bring about the transformation in health care promised by the genome sequence and its derivatives will require assembling diverse stakeholders focused on the application and translation of genomics with a goal of improving the health of individuals and driving efficiency in health care. Specialized centers housing basic genome science laboratories, clinical researchers, informaticians, clinicians, health policy makers and in partnership with industry (pharmaceutical and diagnostic companies), and with health systems that will enable the scientific output of the genome to cross the chasm between bench and bedside. Just a few years ago, academic medical centers alone were the primary mechanism to foster translational research and facilitate collaboration between researchers and clinicians. In the United States, the key ingredients for this effort are in place through the NIH Roadmap and the Clinical and Translational Science Awards (NIH, 2007). This NIH initiative will eventually fund up to 60 academic health centers and their associated clinical practice partners, with significant infrastructure capability in most of the areas above. These large centers are being bolstered by multiple grants and contracts that are intended to provide
avenues for innovative approaches to scientific problems ranging from biomedical informatics to community-based research networks. Similar programs are under way in Canada, the United Kingdom, Singapore, Germany and other countries. In particular, it will be important for governmental and foundational funding agencies to anticipate the critical imperative for interdisciplinary work (Figure 22.4). Academic health systems will need to consciously steer faculty activities in this direction to reap the benefits of the enormous societal investment in the biomedical enterprise that has made this clinical scientific revolution possible. Industry will need to engage earlier in the discovery process and assist in guiding the research agenda. Regulatory agencies and payers will need to prepare for this fundamental shift in the basis for assessing diagnostic and therapeutic technologies and their reimbursement. Effective planning for this coming shift in the way human disease and its prevention and treatment are viewed through the genomic lens, particularly the systems aspects of creating interdisciplinary hubs for the development of translational genomics, will result not only in an advanced clinical scientific discipline, but improved health for people worldwide.
REFERENCES Apse, K., Biesecker, B., Giardiello, F., Fuller, B. and Bernhardt, B. (2004). Perceptions of genetic discrimination among at-risk relatives of colorectal cancer patients. Gene Med 6, 510–516. Bare, L., Morrison, A., Rowland, C., Shiffman, D., Luke, M., Iakoubova, O., Kane, J., Malloy, M., Ellis, S., Pankow, J. et al. (2007). Five common gene variants identify elevated genetic risk for coronary heart disease. Genet Med(10), 682–689. Bethke, L., Sullivan, K., Webb, E., Murray, A., Schoemaker, M., Auvinen, A., Kiuru,A., Salminen,T., Johansen, C., Christensen, H.C. et al. (2008). The common D302H variant of CASP8 is associated with risk of glioma. Cancer Epidemiol Biomarkers Prev 17(4), 987–989. Bullinger, L. and Valk, P. (2005). Gene expression profiling in acute myeloid leukemia. J Clin Oncol 23(26), 6296–6305. Caprioli, R. (2005). Deciphering protein molecular signatures in cancer tissues to aid in diagnosis, prognosis, and therapy. Cancer Res 65(23), 10642–10645. Carlson, J., Henrikson, N.,Veenstra, D. and Ramsey, S. (2005). Economic analyses of human genetics services: A systematic review. Genet Med 7(8), 519–523. Cox, A., Dunning, A., Garcia-Closas, M., Balasubramanian, S., Reed, M., Pooley, K., Scollen, S., Baynes, C., Ponder, B., Chanock, S. et al. (2007). A common coding variant in CASP8 is associated with breast cancer risk. Nat Genet 39(3), 352–358. Dave, S., Fu, K., Wright, G., Lam, L., Kluin, P., Boerma, E., Greiner, T., Weisenburger, D., Rosenwald, A., Ott, G. et al. (2006). Lymphoma/ leukemia molecular profiling project. Molecular diagnosis of Burkitt’s lymphoma. N Engl J Med 354(23), 2431–2442. Douglas, P., Ginsburg, G. (2008). Clinical genomic testing: Getting it right. J Cardiovasc Translational Res 1(1), 17–20. Downing, G.J. (2000). Biomarkers and Surrogate Endpoints: Clinical Research and Applications. Elsevier, New York. Easton, D.F., Pooley, K.A., Dunning, A.M., Pharoah, P.D., Thompson, D., Ballinger, D.G., Struewing, J.P., Morrison, J., Field, H., Luben, R.
et al. (2007). Genome-wide association study identifies novel breast cancer susceptibility loci. Nature 447(7148), 1087–1093. EGAPP (2004) http://www.egappreviews.org/about.htm Eichelbaum, M., Ingelman-Sundberg, M. and Evans, W. (2006). Pharmacogenomics and individualized drug therapy. Annu Rev Med 57, 119–137. Feero, W., Guttmacher, A. and Collins, F. (2008). The genome gets personal–almost. J Am Med Assoc 299(11), 1351–1352. Fellay, J., Shianna, K.V., Ge, D., Colombo, S., Ledergerber, B., Weale, M., Zhang, K., Gumbs, C., Castagna, A., Cossarizza, A. et al. (2007). A whole-genome association study of major determinants for host control of HIV-1. Science 317(5840), 944–947. Flowers, R. and Veenstra, D. (2004). The role of cost-effectiveness analysis in the era of pharmacogenomics. Pharmacoeconomics 22(8), 481–493. Frayling,T.M.,Timpson, N.J.,Weedon, M.N., Zeggini, E., Freathy, R.M., Lindgren, C.M., Perry, J.R., Elliott, K.S., Lango, H., Rayner, N.W. et al. (2007). A common variant in the FTO gene is associated with body mass index and predisposes to childhood and adult obesity. Science 316(5826), 889–894. Frueh, F., Goodsaid, F., Rudman, A., Huang, S. and Lesko, L. (2005). The need for education in pharmacogenomics: A regulatory perspective.. Pharmacogenomics Journal 5(4), 218–220. Frueh, F.W. and Gurwitz, D. (2004). From pharmacogenetics to personalized medicine: A vital need for educating health professionals and the community. Pharmacogenomics 5, 571–579. Gelmann, E.P. (2008). Complexities of prostate cancer risk. N Engl J Med 258(9), 961–963. Ginsburg, G., Burke,T. and Febbo, P. (2008). Centralized biorepositories for genetic and genomic research. J Am Med Assoc 299(11), 1359–1361. Ginsburg, G.S. and McCarthy, J.J. (2001). Personalized medicine: Revolutionizing drug discovery and patient care. Trends Biotechnol 19(12), 491–496.
References
Ginsburg, G.S., Konstance, R.P., Allsbrook, J.S. and Schulman, K.A. (2005). Implications of pharmacogenomics for drug development and clinical practice. Arch Intern Med 165(20), 2331–2336. Glazier, A., Nadeau, J. and Aitman, T. (2002). Finding genes that underlie complex traits. Science 298(5602), 2345–2349. Goodsaid, F., Frueh, F. and Mattes, W. (2008). Strategic paths for biomarker qualification. Toxicology 245(3), 219–223. Graham, I.D., Logan, J., Harrison, M.B., Straus, S.E.,Tetroe, J., Caswell,W. and Robinson, N. (2006). Lost in knowledge translation: Time for a map?. Journal of Continuing Education in the Health Professions 26, 13–24. Gregory, S.G., Schmidt, S., Seth, P., Oksenberg, J.R., Hart, J., Prokop, A., Caillier, S.J., Ban, M., Goris, A., Barcellos, L.F. et al. (2007). Interleukin 7 receptor alpha chain (IL7R) shows allelic and functional association with multiple sclerosis. Nat Genet 39(9), 1083–1091. Gudbjartsson, D.F., Arnar, D.O., Helgadottir, A., Gretarsdottir, S., Holm, H., Sigurdsson, A., Jonasdottir, A., Baker, A., Thorleifsson, G., Kristjansson, K. et al. (2007). Variants conferring risk of atrial fibrillation on chromosome 4q25. Nature 448(7151), 353–357. Gudmundsson, J., Sulem, P., Manolescu, A. et al. (2007a). Genome-wide association study identifies a second prostate cancer susceptibility variant at 8q24. Nat Genet 39, 631–637. Gudmundsson, J., Sulem, P., Steinthorsdottir, V., Bergthorsson, J.T., Thorleifsson, G., Manolescu, A., Rafnar, T., Gudbjartsson, D., Agnarsson, B.A., Baker, A. et al. (2007b). Two variants on chromosome 17 confer prostate cancer risk, and the one in TCF2 protects against type 2 diabetes. Nat Genet 39(8), 977–983. Gurwitz, D., Weizman, A. and Rehavi, M. (2003). Education: Teaching pharmacogenomics to prepare future physicians and researchers for personalized medicine. Trends Pharmacol Sci 24, 122–125. Hall, M., McEwen, J., Barton, J., Walker, A., Howe, E., Reis, J., Power, T., Ellis, S., Tucker, D., Harrison, B. et al. (2005). Concerns in a primary care population about genetic discrimination by insurers. Genet Med 7, 311–316. Haga, S. and Willard, H. (2006). Defining the spectrum of genome policy. Nature Rev Genetics 7, 966–972. Hanauer, D., Rhodes, D., Sinha-Kumar, C. and Chinnaiyan, A. (2007). Bioinformatics approaches in the study of cancer. Curr Mol Med 7(1), 133–141. Helgadotti, A., Thorleifsson, G., Manolescu, A., Gretarsdottir, S., Blondal, T., Jonasdottir, A., Sigurdsson, A., Baker, A., Palsson, A., Masson, G. et al. (2007). A common variant on chromosome 9p21 affects the risk of myocardial infarction. Science 316(5830), 1491–1493. Horig, H. and Pullman, W. (2004). From bench to clinic and back: Perspective on the 1st IQPC translational research conference. Journal of Translational Medicine 2, 44–51. Horig, H., Marincola, E. and Marincola, F. (2005). Obstacles and opportunities in translational research. Nat Med 11(7), 705–708. Humes, H. (2005). Translational medicine and the National Institutes of Health road map: Steep grades and tortuous curves. J Lab Clin Med 146, 51–54. Hunter, D., Khoury, M. and Drazen, J. (2008). Letting the genome out of the bottle – will we get our wish? N Engl J Med 358(2), 105–107. Hunter, D.J., Kraft, P., Jacobs, K.B., Cox, D.G.,Yeager, M., Hankinson, S.E., Wacholder, S., Wang, Z., Welch, R., Hutchinson, A. et al. (2007). A genome-wide association study identifies alleles in FGFR2 associated with risk of sporadic postmenopausal breast cancer. Nat Genet 39(7), 870–874.
■
273
IOM (2007). http://www.iom.edu/CMS/3740/44443.aspx Janssens, A., Gwinn, M., Bradley, L., Oostra, B., van Duijn, C. and Khoury, M. (2008). A critical appraisal of the scientific basis of commercial genomic profiles used to assess health risks and personalize health interventions. Am J Hum Genet 82(3), 593–599. Khoury, M., Gwinn, M.,Yoon, P., Dowling, N., Moore, C. and Bradley, L. (2007). The continuum of translation research in genomic medicine: How can we accelerate the appropriate integration of human genome discoveries into health care and disease prevention? Genet Med(10), 665–674. Klein, G. (2002). Standardization of health informatics – results and challenges. Meth Inform Med 41(4), 261–270. Klein, R., Zeiss, C., Chew, E.,Tsai, J., Sackler, R., Haynes, C., Henning,A., SanGiovanni, J., Mane, S., Mayne, S. et al. (2005). Complement factor H polymorphism in age-related macular degeneration. Science 308, 385–389. Lowe, C.E., Cooper, J.D., Brusko,T.,Walker, N.M., Smyth, D.J., Bailey, R., Bourget, K., Plagnol,V., Field, S., Atkinson, M. et al. (2007). Largescale genetic fine mapping and genotype-phenotype associations implicate polymorphism in the IL2RA region in type 1 diabetes. Nat Genet 39(9), 1074–1082. Lynch, T., Bell, D., Sordella, R., Gurubhagavatula, S., Okimoto, R., Brannigan, B., Harris, P., Haserlat, S., Supko, J., Haluska, F. et al. (2004). Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefitinib. N Engl J Med 350(21), 2129–2139. Ma, X., Zabriskie, N.A., Hau,V.S., Chen, H.,Tong, Z.,Yang, X., Gibbs, D., Katz, B.J., Luo, L., Pearson, E. et al. (2007). Genetic association of LOXL1 gene variants and exfoliation glaucoma in a Utah cohort. Cell Cycle 7(4). Marcom, P.K., Anders, C.K., Ginsburg, G.S., Nevins, J.R. and Potti, A. (2008). Utilization of genomic signatures for personalized treatment of breast cancer. In: “Pharmacogenetics of Breast Cancer”, B. Leyland-Jones, ed., Macmillan Publishing Solutions (in press). Marincola, F. (2003). Translational medicine: A two-way road. Journal of Translational Medicine 1, 1–2. McGuire, A., Cho, M., McGuire, S. and Caulfield, T. (2007). Free in PMC medicine. The future of personal genomics. Science 317(5845), 1687. McPherson, R., Pertsemlidis, A., Kavaslar, N., Stewart, A., Roberts, R., Cox, D.R., Hinds, D.A., Pennacchio, L.A., Tybjaerg-Hansen, A., Folsom, A.R. et al. (2007). A common allele on chromosome 9 associated with coronary heart disease. Science 316(5830), 1488–1491. Metcalfe, S., Hurworth, R., Newstead, J. and Robins, R. (2002). Needs assessment study of genetics education for general practitioners in Australia. Genet Med 4, 71–77. Moffatt, M.F., Kabesch, M., Liang, L., Dixon, A.L., Strachan, D., Heath, S., Depner, M., von Berg, A., Bufe, A., Rietschel, E. et al. (2007). Genetic variants regulating ORMDL3 expression contribute to the risk of childhood asthma. Nature 448(7152), 470–473. NIH (2007). http://www.ncrr.nih.gov/clinical_research_resources/ clinical_and_translational_science_awards/ Ordovas, J. and Corella, D. (2004). Nutritional genomics. Annu Rev Genom Hum Genet 5, 71–118. Parkes, M., Barrett, J.C., Prescott, N.J., Tremelling, M., Anderson, C.A., Fisher, S.A., Roberts, R.G., Nimmo, E.R., Cummings, F.R., Soars, D. et al. (2007). Sequence variants in the autophagy gene IRGM and multiple other replicating loci contribute to Crohn’s disease susceptibility. Nat Genet 39(7), 830–832.
274
CHAPTER 22
■
Translational Genomics: From Discovery to Clinical Practice
Personalized Medicine (2007). http://olpa.od.nih.gov/tracking/110/ senate_bills/session1/s-976.asp Plenge, R., Seielstad, M., Padyukov, L., Lee, A., Remmers, E., Ding, B., Liew, A., Khalili, H., Chandrasekaran, A., Davies, L. et al. (2007). TRAF1-C5 as a risk locus for rheumatoid arthritis – a genomewide study. N Engl J Med 357(12), 1199–1209. Potti, A., Mukherjee, S., Petersen, R., Dress, A.H., Bild, A., Koontz, J., Kratzke, R., Watson, M., Kelley, M., Ginsburg, G. et al. (2006). A genomic strategy to refine prognosis in early-stage non-small-cell lung cancer. N Engl J Med 355(6), 570–580. Remmers, E.F., Plenge, R.M., Lee, A.T., Graham, R.R., Hom, G., Behrens, T.W., de Bakker, P.I., Le, J.M., Lee, H.S., Batliwalla, F. et al. (2007). STAT4 and the risk of rheumatoid arthritis and systemic lupus erythematosus. N Engl J Med 357(10), 977–986. Rioux, J., Xavier, R., Taylor, K., Silverberg, M., Goyette, P., Huett, A., Green, T., Kubolla, P., Barmada, M., Datta, L. et al. (2007). Genome-wide association study identifies new susceptibility loci for Crohn disease and implicates autophagy in disease pathogenesis. Nat Genet 39, 596. Samani, N.J., Erdmann, J., Hall, A.S., Hengstenberg, C., Mangino, M., Mayer, B., Dixon, R.J., Meitinger, T., Braund, P., Wichmann, H.E. et al. (2007). Genomewide association analysis of coronary artery disease. N Engl J Med 357(5), 443–453. Sandhu, M.S.,Weedon, M.N., Fawcett, K.A.,Wasson, J., Debenham, S.L., Daly, A., Lango, H., Frayling, T.M., Neumann, R.J., Sherva, R. et al. (2007). Common variants in WFS1 confer risk of type 2 diabetes. Nat Genet 39(8), 951–953. Saxena, R., Voight, B., Lyssenko, V., Burtt, N., Bakker, N., Chen, H., Roix, J., Kathiresan, S., Hirschhorn, J., Daly, M. et al. (2007). Diabetes Genetics Initiative of Broad Institute of Harvard and MIT, Lund University and Novartis Institutes for BioMedical Research, Genome-wide association analysis identifies loci for type 2 diabetes and triglyceride levels. Science 316, 1331–1336. Scheuner, M., Sieverding, P. and Shekelle, P. (2008). Delivery of genomic medicine for common chronic adult diseases: A systematic review. J Am Med Assoc 299(11), 1320–1334. Scott, L.J., Mohlke, K.L., Bonnycastle, L.L. et al. (2007). A genomewide association study of type 2 diabetes in Finns detects multiple susceptibility variants. Science 316, 1341–1345. Secretary’s Advisory Committee on Genetics, Health, and Society (2007). Draft Report On US System of Oversight of Genetic Testing. Seo, D. and Ginsburg, G. (2005). Genomic medicine: Bringing biomarkers to clinical medicine. Curr Opin Chem Biol(4), 381–386. Sklar, P., Smoller, J.W., Fan, J., Ferreira, M.A., Perlis, R.H., Chambert, K., Nimgaonkar, V.L., McQueen, M.B., Faraone, S.V., Kirby, A. et al. (2008). Whole-genome association study of bipolar disorder. Mol Psychiatr 6, March 4 [Epub ahead of print] Sladek, R., Rocheleau, G., Rung, J., Dina, C., Shen, L., Serre, D., Boutin, P., Vincent, D., Belisle, A., Hadjadj, S. et al. (2007). A genome-wide association study identifies novel risk loci for type 2 diabetes. Nature 445, 881. Snyderman, R. and Williams, R.S. (2003). Prospective medicine: The next health care transformation. Acad Med 78(11), 1079–1084. Snyderman, R. and Yoediono, Z. (2006). Prospective care: A personalized, preventative approach to medicine. Pharmacogenomics 7(1), 5–9. Sonntag, K. (2005). Implementations of translational medicine. Journal of Translational Medicine 3, 33–35. Stallings, S.,Witt,W., Finkelstein, S., Hiller,A., Sinskey,A. and Ginsburg, G. (2006). A framework to evaluate the economic impact of pharmacogenomics. Pharmacogenomics 7, 853–862.
Staudt, L. (2003). Molecular diagnosis of the hematologic cancers. N Engl J Med 348(18), 1777–1785. Steinthorsdottir, V., Thorleifsson, G., Reynisdottir, I., Benediktsson, R., Jonsdottir, T., Walters, G., Styrkarsdottir, U., Gretarsdottir, S., Emilsson, V., Ghosh, S. et al. (2007). A variant in CDKAL1 influences insulin response and risk of type 2 diabetes. Nat Genet 39, 770–775. Stoughton, R.B. and Friend, S.H. (2005). How molecular profiling could revolutionize drug discovery. Nat Rev Drug Discov 4(4), 345–350. Sung, N., Crowley, W., Genel, M., Salber, P., Sandy, L., Sherwood, L., Johnson, S., Catanese, V., Tilson, H., Getz, K. et al. (2003). Central challenges facing the national clinical research enterprise. J Am Med Assoc 289(10), 1278–1287. Tenesa,A., Farrington, S.M., Prendergast, J.G., Porteous, M.E.,Walker, M., Haq, N., Barnetson, R.A., Theodoratou, E., Cetnarskyj, R., Cartwright, N. et al. (2008). Genome-wide association scan identifies a colorectal cancer susceptibility locus on 11q23 and replicates risk loci at 8q24 and 18q21. Nat Genet 40(5), 631–637. Topol, E., Murray, S. and Frazer, K. (2007). The genomics gold rush. J Am Med Assoc 298(2), 218–221. van Heel, D.A., Franke, L., Hunt, K.A., Gwilliam, R., Zhernakova, A., Inouye, M., Wapenaar, M.C., Barnardo, M.C., Bethel, G., Holmes, G.K. et al. (2007). A genome-wide association study for celiac disease identifies risk variants in the region harboring IL2 and IL21. Nat Genet 39(7), 827–829. West, M., Ginsburg, G., Huang, A. and Nevins, J. (2006). Embracing the complexity of genomic data for personalized medicine. Genome Res(5), 559–566. Westfall, J., Mold, J. and Fagnan, L. (2007). Practice-based research – ”Blue Highways” on the NIH roadmap. J Am Med Assoc 297(4), 403–406. Wilkins-Haug, L., Hill, L., Power, M., Holzman, G. and Schulkin, J. (2000). Gynecologists’ training, knowledge, and experiences in genetics: A survey. Obstet Gynecol 95, 421–424. Willard, H., Angrist, M. and Ginsburg, G. (2005). Genomic medicine: Genetic variation and its impact on the future of health care. Phil Trans Roy Soc Lond B Biol Sci 360(1460), 1543–1550. Wohlsen, M.(2008).http://news.yahoo.com/s/ap/20080331/ap_on_hi_ te/trolling_for_genes_startups Yates, J.R., Sepp,T., Matharu, B.K., Khan, J.C.,Thurlby, D.A., Shahid, H., Clayton, D.G., Hayward, C., Morgan, J., Wright, A.F. et al. (2007). Complement C3 variant and the risk of age-related macular degeneration. N Engl J Med 357(6), 553–561. Zeggini, E., Weedon, M., Lindgren, C., Frayling, T., Elliot, K., Lango, H., Timpson, N., Perry, J., Raynor, N., Freathy, R. et al. (2007). Replication of genome-wide association signals in UK. Samples reveals risk loci for type 2 diabetes. Science 316, 1336–1341. Zeggini, E., Scott, L., Saxena, R., Voight, B., Marchini, J., Hu, T., de Bakker, P.I., Abecasis, G., Almgren, P., Andersen, G. et al. (2008). Meta-analysis of genome-wide association data and large-scale replication identifies additional susceptibility loci for type 2 diabetes. Nat Genet 6. Zheng, S., Sun, J., Wiklund, F., Smith, S., Stattin, P., Li, G., Adami, H., Hsu, F., Zhu, Y., Bälter, K. et al. (2008). Cumulative association of five genetic variants with prostate cancer. New Engl J Med 358(9), 910–919.
CHAPTER
23 Principles of Study Design Peter Grass
INTRODUCTION The general procedure in scientific research is to formulate hypotheses and then to verify them directly or indirectly by their consequences. This verification necessitates the collection of observations, and the design of the experiment is essentially the pattern of the observations to be collected. A formal condition for a hypothesis is that it must be formulated in such a way that verification or lack of it may be achieved by direct observation with an experimental procedure, or that deductions made from the hypothesis lead to predictions that may be verified. The purpose of the theory of the design of experiments is to ensure that the experimenter obtains data relevant to his hypothesis in as efficient a way as possible; it strives for optimization in terms of time, resources, information content, and degree of relevance. There are more and less efficient ways of using resources. A measure of accuracy is obtainable, and it is necessary as in all phases of practical life to consider the costs and time consumption of obtaining a particular accuracy, whether it is worth this cost, and at which stage the costs of obtaining increased accuracy are too great (Kempthorne, 1952). In the case of clinical and animal studies, the principles of optimal study design should also consider ethical reasons. During the planning phase of an experimental study it is good practice to prepare a written draft of the proposal for the experiment that includes a statement of the objectives, the description of the experiment covering such matters as the experimental treatment, the sample size, and the experimental material and finally an outline of the method of analysis of the Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
data. Since experimental design and statistical analysis of the data are very much interrelated, it is recommended to involve the statistician or seek his/her advice at the time of planning of the study (Cochran and Cox, 1962). The statement of the objectives may have the form of a hypothesis to be tested or an effect to be estimated. Alternatively, the study may be of exploratory nature with the objectives of generating a hypothesis rather than testing a hypothesis. In any case, the formulation of the objectives should be clear, specific, and devoid of ambiguity. Objectives should not be too ambitious and if necessary, they should be classified into main and secondary objectives because not all objectives can be met with the same level of accuracy with a given study design. There should also be a statement about the area over which generalizations are to be made, or the populations about which inferences will be made. Genomic medicine encompasses a wide range of study designs that are known from preclinical and clinical research and other life sciences. First, there are studies with an exploratory intention in order to generate hypotheses about, for instance, the mode of action of a drug or the biological pathway being impaired by a specific disease. These exploratory studies are followed by confirmatory studies in order to provide evidence for or reject these hypotheses and to obtain some confidence. For instance, a genomic biomarker signature of patients responding or non-responding to an investigational compound could be derived from blood samples collected from patients participating in a classical clinical trial. Usually the sample size of such trials is limited and the patient population is restricted by stringent inclusion and Copyright © 2009, Elsevier Inc. All rights reserved. 275
276
CHAPTER 23
■
Principles of Study Design
exclusion criteria. The design of the study is sub-optimal for the identification of biomarkers because biomarker identification is not the main objective of the trial. The level of confidence in such a genomic signature would be low due to these limitations. Consequently, it is necessary to validate the biomarker signature in one or more subsequent studies and to obtain reliable information about its predictive performance in terms of selectivity, specificity, and generalizability. This approach is well established and principally not different from the development of other diagnostic tools, such as diagnostic imaging techniques or laboratory tests. There is, however, one fundamental difference between genomic studies and classical studies. Genomic, proteomic, and metabolomic studies provide usually multiple endpoints, which can be up to 50,000 in case of microarrays or more than 100 different multiplexed assays, in contrast to a single endpoint of a classical study. Multiple endpoints need to be taken into account when planning a study because the sample size of the study has to be adjusted accordingly. In the following, the fundamental design concepts will be introduced before measures to increase precision and accuracy of experimental studies will be presented including randomization, blocking, and replication with reference to sample size calculation. There are many further measures that may help to optimize the study design such as blinding, appropriate choice of control groups, and deviation from baseline measurements. Finally, possible biases that are typical for genomic and proteomic studies will be presented.
PRINCIPLES OF EXPERIMENTAL DESIGN Design Concepts Several statistical design concepts have to be explained before we can apply the principles of experimental design to genomic medicine studies. These design concepts are very generic and hold true for any experimental study, be it a typical clinical trial or a clinical genomic study. The first design concept refers to the observation of an effect that usually involves a measurement process. We all know that each measurement process is inherently imperfect, and that the measured value usually deviates from the true value by some amount called the experimental or measurement error. This type of error may be due to technical variations such as temperature fluctuations, small inadvertent impurities of the reagents needed for an assay, or small differences in the manual handling of the instrument. Small inaccuracies of the treatment conditions such as small variations in dosing may also contribute to the experimental error. Any optimization of the measurement process leads towards a minimization of the measurement error and ultimately to a higher precision of the measurement result. Obvious measures to improve the precision include repeated measurement of the same sample, standardization of the measurement process, and automation.
Second, there is the biological variation of the parameter of interest. We know clinical parameters may vary between populations, for instance, between Caucasians and Blacks, and between individuals within a given population; for instance, male versus female. Even within a subject a parameter may vary over time due to circadian rhythms, according to their health conditions, demographic particularity of subjects or other nutritional and environmental factors. Normal ranges of laboratory parameters have been defined within which the measurement values may vary under normal physiological conditions. Generally, biological variation cannot be minimized by a more precise measurement, but it can be reduced by specifying it for more homogeneous groups; for example, for males and females separately and within each gender for different age groups. Measurement error as well as biological variation are known to be random and both are very often normally distributed; that is, small positive and negative-random variations around an average value are more frequent than large variations which yields the well-known bell-shaped Gaussian distribution curve. In addition to random variation, there may be systematic deviation of the measurement value from the true value which is called bias. Such systematic deviations may be due to faults in the measurement process. For instance, incorrect calibration of an instrument would yield concentrations of a laboratory parameter being measured systematically either too high or too low, which is an example of measurement bias. A systematic error may also be introduced at the level of recruitment when patients or subjects are not representative of the entire population. For instance, only young healthy males are included in a study with the objective to assess the concentrations of a given parameter, but conclusions about the parameter distribution of the entire population will be made. This is a typical example of selection bias. Ultimately, the goal of any experimental study is to draw general conclusions about the behavior of a parameter of interest in an entire collective. Therefore, another important item refers to the design concepts of population and random sample. A population encompasses the entirety of subjects under a given experimental condition. An experimental study, however, will never be conducted in the whole population, but rather in a small sub-population, referred to as a sample, which would include a limited number of N subjects only. It is assumed that the sample is a valid representation of the entire population if the sample elements are drawn randomly from the entire population and the sample size N is sufficiently large. Within a population, the parameter of interest has a given distribution which may be characterized by its population mean () and standard deviation ( ) in the case of a normal distribution. Mean and standard deviation as calculated from the sample data is only an estimation of the true population parameters where the precision of the estimate is influenced by the sample size. The individual elements of a sample that undergo a predetermined treatment or belong to a certain condition are called experimental units (e.g., patients, subjects, animals) while repeated measurements of a parameter in the same experimental unit are called observational units; for instance, several readings of blood pressure in the same patient obtained at the same time under the same treatment condition.
Principles of Experimental Design
Further design concepts in the context of genomic studies include experimental conditions or treatments where conditions may refer to a disease status of a patient and treatment may refer to different medications. The investigation of the influence of the different conditions or treatments on a target parameter is usually the main objectives of a study. A main objective of experimental design is to determine which conditions or treatments and which levels should be chosen in a systematic way. Random allocation of the experimental units to the different conditions or treatments is mandatory. The choice of an appropriate control group is very important and affects the outcome of the study. Obviously, it makes a difference whether you take a placebo treatment or the reference treatment as a control condition. In contrast, blocking factors are conditions or treatments that are expected to influence the measurement systematically but are of no particular interest, so called nuisance variables. Typical examples of nuisance variables in genomic studies are ethnicity and gender of the patients, or the centers of a multicenter trial, but also operators of the laboratory where the mRNA extraction is performed. Confounding means the “mixing” of sources of variation in data so that their effects cannot be distinguished from each other. The sources of variation can be factors introduced by the experimenter or unknown variables introduced by the patient or the environment (Wooding, 1994). A general rule (Box et al., 2005) says “Block what you can, randomize what you cannot.” One of the most prominent objectives of experimental design is to identify possible nuisance variables and account for them by using them as blocking factors. Measures to Increase Precision and Accuracy Accuracy and precision are two characteristics that determine the quality of an experimental study. Precision describes how well the measurement can be reproduced and is usually reported as an average replication error. It can be determined by performing repeated measurements of the same experimental unit. Accuracy describes how close to a true population value a measurement lies. In many situations the true population value is unknown; this is in particular true for microarray experiments due to the absence of calibration samples. In experiments, however, where calibration samples of known concentrations are used, the accuracy can be estimated (Moreau et al., 2003). Bias effects accuracy and if the bias is large, a measurement may be of high precision but of low accuracy. Basically, there are three ways to increase the precision and accuracy of experimental studies: (a) refinement of experimental techniques; (b) careful selection of subjects with additional information about them that can account for a part of variation or by grouping them according to certain characteristics; and (c) enlargement of the size of experiment by employing more replicates. Fundamental techniques applied in statistical design (Kerr, 2003) include randomization, blocking, and replication. Randomization A random sample is one in which each individual of a population to be sampled has an equal chance of being selected. This
■
277
ensures that there is no bias; that is, on the average, the estimates of the population parameters will be accurate (Bolton and Bon, 2004). Furthermore, we assume that individuals being randomly sampled from a population are valid representatives of this population. Randomization encompasses both random selection from a population and random assignment to a treatment. Another reason for randomization is that statistical theory is based on the idea of random sampling. In a study with random allocation, the differences between treatment groups behave like the differences between random samples from a single population. We know how random samples are expected to behave and can compare the observations with what we would expect if the treatments were equally effective (Altman and Bland, 1999). However, randomization does not eliminate the possibility that treatment and control groups will differ according to some unknown factors that affect outcome (confounding variables). We can expect the random assignment of treatments to ensure some rough balancing of those factors. It is important to recognize that because randomization relies on the averaging of sampling variation, in studies with small numbers of subjects it may not effectively reduce bias (Sica, 2006). Blocking Blocking is another measure to increase precision and to reduce bias. In practice, subjects of similar characteristics are grouped together in blocks and randomly assigned to treatments. In a multi-center trial, for example, patients would be blocked by center, and within each block they would be randomly assigned to the different treatments. Other common blocking variables include gender, age groups, and ethnicity. In crossover trials, where each patient represents a block, each patient receives each of the two or more treatments in random order. Such a randomized block design is advantageous when the level of responses are very different between patients, but the within-patient variability is relatively small. If the subjects are properly randomized within the blocks, data analysis techniques, such as analysis of variance, are able to separate the variability due to the different blocking factors resulting in a decreased experimental error. Replication In analogy to biological and technical variation we can specify biological and technical replication. Biological replication is directly linked to the number of experimental units (e.g., subjects or patients) included in the study. Technical replication refers to repeated measurements of a particular parameter derived from one and the same experimental unit under the same experimental condition. Multiple observations made on material from the same experimental unit should not be confused with biological replications. Whatever the source of the experimental error, replication of the experiment by increasing the sample size steadily decreases the error associated with the difference between the average results for two treatments/conditions. Precautions have to be taken to ensure that one treatment/condition is no more
278
CHAPTER 23
■
Principles of Study Design
δ
Number of Subjects
1200
Populations μσ males 15.5 2.0 females 13.5 2.5
900
T A B L E 2 3 . 1 Results of a Monte Carlo Simulation: Mean, standard deviation and coefficient of variations (CV) of the differences between the 5000 sample means drawn from a male and female Sample Differences (n ⴝ 5000)
600
N 300
0 10
12
14
16
18
Hemoglobin (g/dL)
Figure 23.1 Frequency distribution of hemoglobin concentrations (g/dL) in healthy female and male subjects.
likely to be favored in any replicate than another, so that the errors affecting any treatment tend to cancel out as the number of replications is increased (Cochran and Cox, 1962). Monte Carlo Simulation The effect of replication on the precision of an experimental study can be demonstrated by a Monte Carlo simulation. Assuming that the true distribution of the hemoglobin concentrations in male and female healthy subjects is known by measuring hemoglobin in thousands of representatives in both populations, the results of this investigation are depicted in Figure 23.1. The hemoglobin concentrations are normally distributed within both populations with population means and standard deviations of 13.5 2.5 g/dL in females and 15.5 2.0 g/dL in males. Hence, the true difference ( ) between the two population means is exactly 2 g/dL, and the standard deviations of both populations are relatively large compared to the difference between the means as indicated by the ratio ⁄ 1.125. The objective of the Monte Carlo simulation is to provide an estimate of the difference between the two populations based on samples being drawn randomly from these populations and to demonstrate how the sample size influences the reliability or confidence of the estimated parameter. To do so, we draw small samples of N individuals from both populations and we assume that the samples are valid representatives of the entire populations. Based on the statistical characteristics of the samples we want to draw conclusions that hold true for the entire populations. Intuitively, we know that the individuals of each sample should be drawn randomly from the entire populations. Let us generate two samples by drawing randomly 3 individuals from a female population and 3 individuals from a male population. We calculate the mean x f , N 3 of the female sample and xm, N 3 of the male sample, and then the difference between the two sample means N 3 x f , N 3 xm, N 3 .
Mean
Stdev
CV (%)
3
2.05
1.86
91
6
2.01
1.29
64
12
2.03
0.91
45
24
1.99
0.65
33
48
2.00
0.46
23
96
2.00
0.33
16
192
2.00
0.23
12
The mean converges to the true difference ( 2.0) between the two populations and simultaneously, the CV decreases from 91% for N 3 down to 12% for N 192.
Let us repeat this procedure for n 5000 times and thereafter, calculate the mean and standard deviation of all n 5000 differences among the sample means. N 3
SN 3
n
1 n
∑ i , N 3
and
i1
∑ (i,N 3 N 3 )2 i
n 1
For N 3 per group the average difference and standard deviation is 2.05 1.86 g/dL. We can see that the average difference is close to the true difference, but the estimate is not very precise as indicated by the standard deviation; the coefficient of variation amounts to 91% (see Table 23.1). We continue the Monte Carlo simulation by increasing the sample size from N 3 to N 6, 12, 24, 48, 96, and 192 and perform the same procedure as described above. Figure 23.2 displays the results of the simulation study; the mean differences between the average sample hemoglobin concentrations in males and females as a function of sample size are quite accurately close to the true value of 2 g/dL. The standard deviation of the difference is indicated by the error bars, and the trumpetlike shape of the envelope clearly demonstrates how the estimation error decreases with increasing sample size. We realize that the estimate of the hemoglobin difference between male and female becomes more and more precise with increasing sample size. With a sample size of N 192 the coefficient of variation has decreased to a level of 12% (see Table 23.1). If the variability of the populations is increased and/or the true difference between the populations is decreased a higher sample size is needed in order to obtain the same level of precision. Figure 23.2 displays the result of the simulation with
Principles of Experimental Design
Hemoglobin Difference (g/dL)
4
Difference of Sample Means (g/dL) between Males and Females from 5000 Monte Carlo Simulation per Sample Size Means Stdev
3
2
1
0 0
48
96 144 192 Sample Size per Group
240
288
Figure 23.2 Mean differences (solid diamonds) between mean hemoglobin concentrations in males and females as a function of sample size which was successively increased (N 3, 6, 12, 24, 48, 96, and 192) with ⁄ 1.125. The standard deviation of each difference is indicated by the error bars. The trumpet-like shape of the envelope clearly demonstrates how the error bars decrease with increasing sample size. In consequence, the precision of the estimate of the hemoglobin difference between male and female increases. Dotted lines are the results of a simulation study with increased variability and/or decreased difference with
⁄ 1.625.
the ratio ⁄ 1.625; the estimation errors as indicated by the dashed lines are wider for a given sample size compared to the previous simulation with a ratio of ⁄ 1.125. In summary, the Monte Carlo simulation study showed us how the estimate of a population mean becomes more and more precise with increasing sample size N. In addition, we have noticed that – with a given sample size N – the difference between the two populations and the variability , in particular the ratio / play a crucial role in the estimation of the population parameters. Large ratios due to higher variability and/or smaller differences yield less reliable estimates and vice versa. Sample Size Calculation The reliability of the estimate of the difference between the two conditions can be related to the probability of committing an error. When comparing two samples obtained from two conditions, there are principally two different types of error that may occur and that we want to control. A type I error () is defined as the risk of concluding that conditions differ when, in fact, they are the same. The level is usually set at 5%. A type II error () is the risk of erroneously concluding that the conditions are not significantly different when, in fact, a difference of a given size or greater exists. Commonly chosen values of are between 5% and 20%. The type II error is directly linked to the power of test (1-) which implies the probability of being able
■
279
to demonstrate a statistically significant difference between the samples if a true difference of the estimated size actually exists in a larger population. In the planning phase of an experimental study we can use the Greek Quadruple ( , , , and ) in order to calculate the sample size N of the experiment needed to detect a difference of size with given a standard deviation . The difference should be of “practical significance” and should make sense in the context of the experimental setting. The standard deviation of the parameter of interest is usually taken from already existing data; that is, from a pilot study or from data published in literature or by an educated guess. It is the ratio / that impacts the sample size at a given - and -level. Intuitively, we know that a greater sample size is needed when we want to detect a smaller difference , or when the variability is greater. If we want to have more confidence in our results by minimizing type I and type II error we would increase the sample size too. There are formulas to calculate sample size that are specific for the different experimental designs. It is beyond the scope of this article to provide explicit formulas for sample size calculations under the different experimental designs and it is recommended to get advice from a statistician or use an appropriate computer program. In the context of genomic studies it is important to note that classical sample size calculations are only valid for studies with a single endpoint. Genetic, genomic, proteomic, and metabolomic studies, however, are characteristic for their multiple endpoints which need to be accounted for in sample size calculations, for instance by choosing false discovery rate (Benjamini and Hochberg, 1995) or adjusting the Type I error for multiple comparisons. In genomic studies, the estimation of the standard deviation is aggravated by the facts that the number of genes with a reliable signal and the associated standard deviation are different for each of the thousands of features on a chip, that they deviate from tissue to tissue, and that they differ between chip technologies (Lee et al., 2000). Further Measures to Optimize Study Design Accuracy as well as precision can strongly be influenced by how the different treatments or conditions are allocated to the experimental units. Well-known designs of clinical trials include parallel group design and cross-over design. Many other design concepts are employed for the purpose of bias elimination and reduction of the experimental error; for instance, blinding techniques, inclusion of positive control groups, repeated measurements, and others. The objective is always to reduce the heterogeneity of the experimental units by careful selection of patients by specifying inclusion and exclusion criteria and check of patient’s eligibility. It must be pointed out, however, that patients fulfilling the eligibility criteria for homogeneous groups are usually not representatives of all those with the disease or all those for whom the therapy is intended; that is, the target patient population. The art of good experimental design is to find the right balance between taking into consideration as many sources of variation as possible on one hand and the simplicity and practicability
280
CHAPTER 23
■
Principles of Study Design
of an experimental study on the other. Without doubt, more complex study designs have more restrictions and they lack flexibility; they are difficult to implement and they are bias-prone. For instance, the number of patients participating in cross-over studies is usually smaller in comparison to parallel studies, but the impact of patients dropping out from the study is much higher. Another important aspect is related to the symmetry of study designs. Although most data analysis methods are able to cope (to a limited extent) with unbalanced data and missing values, a certain symmetry should be strived for; for example, equal number of patients per treatment group, equal number of visits per patient, balanced order of administration, and equal number of replicates per patient. Symmetry of design increases the power of the analysis and – even more important – makes the conduct of the study less error prone and facilitates the quality control of the data. Optimal study design is also concerned with maximization of the information content of the study data. In a placebo controlled trial, the relative deviation of a clinical parameter under a given treatment condition from its baseline value is more informative than comparing the mean parameter of a placebo treated patient group with that of a different group of actively treated patients. Note that deviation from baseline in the absence of a placebo control group may be due to changes in environmental conditions during the intervening time period; therefore, the inclusion of a placebo group is strongly recommended. Classical pharmacokinetic/pharmacodynamic studies are very often designed as placebo-controlled cross-over studies. Their objective is to assess the time course of the drug concentration in blood and tissue, and simultaneously to measure the pharmacological effects that might be elicited by that compound or placebo in the same patient. Indisputably, such related concentration-effect time profiles provide more information than a single reading of the effect at a predetermined visit. However, in many situations the design is too complex for a clinical routine situation, and another caveat is a possible carry-over effect if the wash-out phase between treatments is not long enough. The majority of studies are designed as comparative trials, and the appropriate choice of the control treatment or condition is crucial. Several types of comparisons are possible: between a patient group being treated with an active compound and another patient group being treated with placebo; or between a patient group being treated with a developmental compound and another patient group being treated with the current clinical gold standard as a positive control. Even the simultaneous inclusion of placebo treatment, positive control and negative control may make sense in certain situations. Not only the choice of the control treatment is important but also the choice of the control population should not be neglected. It makes a difference whether the control group consists of age-matched healthy volunteers or age-matched co-morbid patients. Blinding, sometimes called masking, is another measure to help avoid bias. Human behavior is influenced by what we know and what we believe. In research there is a particular risk of expectation influencing findings, most obviously when there is some subjectivity in the assessment, leading to biased results.
Blinding is used to try to eliminate such bias. Blinding is particularly important when the response criteria are subjective; a typical example is the assessment of pain. Double-blind usually refers to keeping patients, those involved with their management such as nurses and study monitors, and those collecting and analyzing clinical data unaware of the assigned treatment, so that they should not be influenced by that knowledge (Day and Altman, 2000). It is important that the treatments look similar: same taste and smell, same color, same number of capsules to be taken, same route of administration which is called the double dummy method. Blinding in the assessment of the performance of a diagnostic test is crucial: those who do the assessment must be blinded; also when testing the reproducibility of a diagnostic test the observer must be unaware of the previous measurements. If not blinded, clinical trials show larger treatment effects and diagnostic test performance is overestimated. Possible Bias in Genomic Studies Two kinds of bias may be of particular importance for genomic studies. The first kind refers to the selection of the experimental units, that is which subjects or patients are to be recruited to participate in a study, and the second kind of bias may be introduced at the measurement level. Again, the process of patient selection should ensure that they are representative for the target population. Inclusion and exclusion criteria should not be too stringent because they may limit the generalization of the study findings. Selection bias in biomarker development are manifold and they usually lead to an overestimation of the sensitivity of the assay. Selection bias examples include the exclusion of patients with mild disease who are usually difficult to diagnose. The limitation of the patient population to only those who are volunteering to participate in the study might introduce the bias that health-conscious patients are over-represented. Patients dropping out from a study may be different from patients who remain in the study; consequently, the exclusion of those patients from the analysis will introduce a selection bias. The choice of a control group may also introduce bias. It makes a difference whether you choose young, disease-free subjects or non-diseased, but age-matched patients as a control group to be compared with a diseased group (Sica, 2006;Whiting et al., 2004). Observational biases may be caused by differences in methods in which information is collected and data are obtained or due to the fact that patients may under- or over-report their medical history or personal habits if they are aware of their diagnosis. Bias due to different methods being applied to different experimental groups occurs in clinical studies quite frequently. For instance, a reference test is only applied in case of a positive study test result. The consequence is that more sick subjects than healthy subjects undergo the reference test. Patients with negativetest results and those with benign-appearing lesions do not typically undergo an invasive reference test, for instance a kidney biopsy. On the other hand, bias may be introduced because patients with positive test results undergo more intensive follow-up investigations and therefore, missing data are present nonrandomly for disease-free subjects. Another issue is represented by
Design Issues in Genomic Medicine
the fact that the reference standard against which a new genomic biomarker is to be compared is not 100% accurate or a surrogate reference test is used instead of the clinical endpoint; for instance, coronary angiography for the diagnosis of coronary artery disease. Between-group differences in reporting may happen because patients with positive diagnostic results may better recall or even exaggerate their presentation symptoms. When an unblinded investigator is the interviewer and also involved in the interpretation of the results he/she may inquire details of medical history more rigorously in the case of positive test results. There are several sources of variation that are specific to genomic and proteomic experiments. The process of tissue sampling should be randomized across conditions in order to ensure uniform handling of the tissue specimen. In a toxicology study, for instance, it makes a difference when vehicle treated animals are all sacrificed first, followed by the low, medium and high dose group animals in a sequential way. Due to a considerable time differences between last dosing and tissue harvest the tissue exposure to the drug may be different, and in consequence, mRNA response may be changed due to this systematic error in sample collection. Another sample collection bias refers to the time of food intake in relation to specimen collection which should be blocked for, or if blocking is impossible, should be randomized. The sample collection scheme should also take circadian and seasonal changes into account (Rudic et al., 2005) and the stress exposure of the patients and animals should be balanced across treatments. Brain is a complex organ, with a heterogeneous distribution of distinct subpopulations of cells, intricate signaling and regulatory circuits, and exquisite lifelong sensitivity to environmental variation. These factors result in high levels of interindividual variability of gene expression (Karssen et al., 2006). Due to a difficult sampling situation in pathology possible tissue contaminations may occur; for instance, rat duodenum samples are often contaminated by pancreas tissue. The tissue sample volume is related to the amount of mRNA that can be extracted and should therefore be balanced across conditions. In clinical trials, co-medication can considerably influence the gene expression pattern of the patients. The way specimens are stored may also impact the outcome of the experiment; nuisance variables include storage duration and temperature, and whether tissue samples are conserved in paraffin blocks or as snap frozen samples. Another source of variation is how specimens are harvested from heterogeneous tissues such as brain or kidney. The mRNA extraction protocol should be standardized and the samples should be randomized across operators. Batch effects may be introduced by age and impurities of reagents. It is a well-known fact that microarray chips vary between production batches and it is strongly recommended to use one batch of chips for one study. If this is not possible the samples should be blocked by batches. It is good experimental practice to record temperature, experimenter, and date/time of hybridization (Li et al., 2002; Zakharkin et al., 2005). Systematic variation can be due to the use of different scanners or scanner settings, or different photo-multipliers and gains. Different spot-finding software and different grid alignments may also introduce bias (Coombes et al., 2002).
■
281
Needless to say, the entire technical process should be as standardized as possible and tissue sample processing should be as homogeneous as possible. If for any reason this is not possible, a proper blocking or randomization of the sample processing should be taken into consideration. Obviously, technical variation of the measurement process can be controlled to a certain extent (Tinker et al., 2003).
DESIGN ISSUES IN GENOMIC MEDICINE The basic concepts of experimental design are also applicable to genomic medicine studies; randomization, blocking, and replication are the most important measures to improve the accuracy and precision of the experimental outcome. Typical sources of random as well as systematic variation known from classical preclinical and clinical studies could possibly be identified also in genomic studies. There are, however, additional sources of bias that are directly related to the new technologies, and this requires the extension of the design concepts also to the tissue sampling and measurement level. From an experimental design point of view there are some fundamental issues that are not yet completely solved and need further investigation. Genomic and proteomic studies produce information of unparalleled wealth.Without doubt is the high sensitivity of these technologies in combination with the high number of features – genes or proteins – being interrogated simultaneously a great achievement for genome-wide screening purposes. The downside of this is that nuisance variables which are negligible in the clinical routine may gain importance on the molecular level; factors such as co-medication or co-diseases or personal habits like cigarette smoking may interfere with the molecular signature of a disease or a drug treatment under investigation. When designing a genomic study one should keep in mind that gene expression is a highly dynamic phenomenon that develops over time and may be subjected to feed-back and feedforward mechanisms (Yang and Speed, 2002). Therefore, the time point when the tissue specimen is collected in relation to a stimulus such as the administration of a drug is important. For instance, the experimental design of a toxicology study needs to take into consideration that immediate response genes are differentially expressed within minutes after injection of a toxic compound and may be back to normal levels after a few hours followed by mid term response genes that may react hours later. Late response genes mainly including residual persistent effects or compensatory repair mechanisms may react even days later. The gene signature is dependent on the time point of sampling. Another open issue represents the lack of microarray validation in a form as we know it from other bioanalytical assays, although promising attempts in this direction are made (Van Bakel and Holstege, 2004). Microarray technologies, whether single- or dual-channel microarray designs, always produce relative gene expression values. At the moment it is impossible to obtain an absolute measure of the concentration or the amount
282
CHAPTER 23
■
Principles of Study Design
of a specific mRNA segment. Validated assays would employ calibration samples for the calculation of a calibration curve and to determine upper and lower limit of quantification and detections, respectively. Quality control samples of known content could be randomly distributed among the unknown samples in order to provide in-process quality control. As long as we do not know the true expression values of the genes it is extremely difficult to identify measurement bias that is introduced at the mRNA extraction and hybridization level. Real-time PCR confirmation will always be limited to a subset of genes and cannot replace the validation of an entire microarray (Wang et al., 2006). Furthermore, it cannot provide information about the selectivity and precision of the microarray itself. From an experimental design point of view, dualchannel microarrays do not really provide advantages over single-channel microarrays; quite the reverse is true, due to possible dye effects special designs of the hybridization need to be used which include dye swapping and dye balancing in order to take these systematic errors into account (Churchill, 2002; Kerr and Churchill, 2001a, b;Wit and McClure, 2004). A related open issue is the poor reproducibility of genomic results across studies. Even the same tissue specimen being analyzed by different laboratories may produce different genomic signatures even when using the same microarray technology. The reason is only partly due to variation in mRNA extraction protocols, hybridization process and scanning. In particular in the field of biomarker research it had to be realized that many peerreviewed studies publishing genomic biomarker signatures based on statistically significant differences are not reproducible, and the conclusions drawn are overly optimistic (Halloran et al., 2006). Validation studies need to be conducted by several completely independent teams with appropriate sample size. Small sample sizes might actually hinder the identification of truly important genes (Ioannidis, 2005). In contrast to the current standard, thousands of samples are needed to generate a robust gene/ protein signature for predicting outcome of disease (Ein-Dor et al., 2006). If the identification and validation of gene signatures is the objective the approach of sample size calculation needs rigorous reconsideration. In the context of personalized medicine we can expect more and more clinical trials that will include genomic or genetic markers for population enrichment. Genomic biomarkers will be developed for the identification of patients likely to respond to treatment with a developmental compound. The availability of such response biomarkers will greatly influence the design of clinical phase III trials and will have a major impact on the sample size and statistical power. Furthermore, enrichment of clinical studies by selecting patients who are more likely to exhibit an effect fits with the growing appreciation that patients who seem similar may differ in their risk and likelihood of response to treatment (Temple, 2005). If responders represent only a small fraction of the patients of a particular disease, it may be difficult, if not impossible, to find a treatment effect in an unselected-non-enriched-population.
Let us assume a valid biomarker of response to a developmental compound exists before commencing a clinical phase III trial – which is, admittedly, rarely the case-, then patients with positive test result will be the only ones being included in the study. The response biomarker must be available and, in conjunction with the compound, has clinical utility. Under these circumstances, the approval will be limited to responders, only. However, if the test is commercially not available or is not validated safety assessments in both-test positive and test negativepatient groups is needed (White et al., 2006). An interesting adaptive clinical trial design for generating and prospectively testing a gene expression signature for sensitive patients was proposed by Freidlin and Simon (Freidlin and Simon, 2005; Simon, 2005). According to this proposal a phase III trial is designed as usual, that is randomization of patients to new drug and control treatment, but dividing the study into first part involving N1 patients and second part involving N2 patients. After completion of the first part of the study, an interim data analysis is conducted based on the available data. The aim of this analysis is to identify a genomic or genetic marker of response. Then, the study will be completed by entering all planned patients, responder as well as non-responder. The final analysis is conducted in a two step approach. First, the new drug is compared to the control treatment including all patients and ignoring the responder classification. If the treatment effect on the primary pre-specified endpoint is significant at the 0.04 level then effectiveness for the eligible population as a whole can be claimed. However, if the overall test is not significant at the 0.04 level, a single subset analysis will be performed by evaluating the new drug in the responder patients accrued in the second part of the study relative to control treatment. If the treatment effect is significant at the 0.01 level, effectiveness for responder population can be claimed. In this situation, the diagnostic test to identify responders to treatment needs to be validated and made available. A critical requirement of this design is the a priori planning of the data analysis. Obviously, this adaptive design does not reduce the sample size; it requires even slightly more patients for a given statistical power since the overall significance level is at 0.04 rather than 0.05. However, the design has some advantages that might compensate for the slightly higher sample size. Safety assessment, for instance, is performed in the whole study population as in any classical trial. In addition, the design provides an opportunity to demonstrate a significant treatment effect even in a small fraction of responders. Based on their simulation results (Freidlin and Simon, 2005) highest statistical power is obtained with a N1/N2 ratio slightly less than or equal to one provided that sensitivity and specificity of the response biomarker is in the range of 0.9–1.0. We are in the beginning of trying to incorporate pharmacogenomics and pharmacogenetics into clinical trials and there is very limited practical experience yet. Certainly, it will not always be possible to find genetic associations that would relate to drug response, and if an association is found, it may be difficult to replicate it.
Recommended Resources
■
283
REFERENCES Altman, D.G. and Bland, J.M. (1999). Treatment allocation in controlled trials: Why randomize. BMJ 318, 1209. Benjamini, Y. and Hochberg, Y. (1995). Controlling the false alarm discovery rate: A practical and powerful approach to multiple testing. J Roy Stat Soc B57, 298–300. Bolton, S. and Bon, C. (2004). Pharmaceutical Statistics. Practical and Clinical Applications, Fourth Edition. Marcel Dekker, New York. Box, G.E.P., Hunter, J.S. and Hunter, W.G. (2005). Statistics for Experimenters. Design, Innovation, and Discovery, 2nd Edition. John Wiley & Sons, Hoboken, NJ. Churchill, G.A. (2002). Fundamentals of experimental design for cDNA microarrays. Nat Genet 32, 490–495. Cochran, W.G. and Cox, G.M. (1962). Experimental Design. John Wiley & Sons, New York. Coombes, K.R., Highsmith, W.E., Krogmann, T.A., Baggerly, K.A., Stivers, D.N. and Abruzzo, L.V. (2002). Identifying and quantifying sources of variation in microarray data using high-density cDNA membrane arrays. J Comp Biol 9, 655–669. Day, S.J. and Altman, D.G. (2000). Blinding in clinical trials and other studies. BMJ 321, 504. Ein-Dor, L., Zuk, O. and Domany, E. (2006). Thousands of samples are needed to generate a robust gene list for predicting outcome in cancer. PNAS 103, 5923–5928. Freidlin, B. and Simon, R. (2005). Adaptive signature design: An adaptive clinical trial design for generating and prospectively testing a gene expression signature for sensitive patients. Clin Cancer Res 11, 7872–7878. Halloran, P.F., Reeve, J. and Kaplan, B. (2006). Lies, damn lies, and statistics: The perils of the p value. Am J Transplant 6, 10–11. Ioannidis, J.P.A. (2005). Microarray and molecular research: Noisy discovery. Lancet 365, 454–455. Karssen, A.M., Li, J.Z., Her, S., Patel, P.D., Meng, F., Evans, S.J., Vawter, M.P., Tomita, H., Choudary, P.V. and Bunney, W.E. (2006). Application of microarray technology in primate behavioral neuroscience research. Methods 38, 227–234. Kempthorne, O. (1952). The Design and Analysis of Experiments. Robert E. Krieger, Malabar. Kerr, M.K. (2003). Design considerations for efficient and effective microarray studies. Biometrics 59, 822–828. Kerr, M.K. and Churchill, G.A. (2001a). Experimental design for gene expression microarrays. Biostatistics 2, 183–201. Kerr, M.K. and Churchill, G.A. (2001b). Statistical design and the analysis of gene expression microarray data. Genet Res 77, 123–128. Lee, M.L.T., Kuo, F.C., Whitmore, G.A. and Sklar, J. (2000). Importance of replication in microarray gene expression studies: Statistical methods and evidence from repetitive cDNA hybridizations. Proc Natl Acad Sci 97, 9834–9839.
Li, X., Gu, W., M, S. and Balink, D. (2002). DNA Microarrays: Their use and misuse. Microcirculation 9, 13–22. Moreau, Y., Aerts, S., De Moor, B., De Strooper, B. and Dabrowski, M. (2003). Comparison and meta-analysis of microarray data: From the bench to the computer desk. TrendGenet 19, 570–577. Rudic, R.D., McNamara, P., Reilly, D., Grosser,T., Curtis,A.M., Price,T.S., Panda, S., Hogenesch, J.B. and FitzGerald, G.A. (2005). Bioinformatic analysis of circadian gene oscillation in mouse aorta. Circulation 112, 2716–2724. Sica, G.T. (2006). Bias in research studies. Radiology 238, 780–789. Simon, R. (2005). A roadmap for developing and validating genomic classifiers for treatment selection. www.fda.gov/cder/genomics/ presentations_20051006/051007_15_Simon.pdf Temple, R. (2005). Use og genomic biomarkers in a regulatory environment. DIA Genomics Biomarker Conference, October 7, 2005. www.fda.gov/cder/genomics/presentations_20051006/051007_ 10_Temple.pdf Tinker, N.A., Robert, L.S., Butler, G. and Harris, L.J. (2003). Data pre-processing issues in microarray analysis. In A Practical Approach to Microarray Data Analysis (D.P. Berrar, W. Dubitzky and M. Granzow, eds), Kluwer Academic Publishers, Norwell, pp. 47–64. Van Bakel, H. and Holstege, F.C.P. (2004). In control: Systematic assessment of microarray performance. EMBO Rep 5, 964–969. Wang,Y., Barbacioru, C., Hyland, F., Xiao,W., Hunkapiller, K.L., Blake, J., Chan, F., Gonzales, C., Zhang, L. and Samaha, R. (2006). Large scale real-time PCR validation on gene expression measurements from two commercial long-oligonucleotide microarrays. BMC Genomics 7, 59. White, T.J., Clark, A.G. and Broder, S. (2006). Genome-base biomarkers for adverse drug effects, patient enrichment and prediction of drug response, and their incorporation into clinical trial design. Person Med 3, 177–185. Whiting, P., Rutjes, A.W.S., Reitsma, J.B., Glas, A.S. and Bossuyt, P.M.M. (2004). Sources of variation and bias in studies of diagnostic accuracy. Ann Intern Med 140, 189–202. Wit, E. and McClure, J. (2004). Statistics for Microarrays. John Wiley & Sons, Chichester. Wooding,W.M. (1994). Planning Pharmaceutical ClinicalTrials.John Wiley & Sons, New York. Yang, Y.H. and Speed, T. (2002). Design issues for cDNA microarray experiments. Nat Rev Genet 3, 579−588. Zakharkin, S.O., Kim, K., Mehta, T., Chen, L., Barnes, S., Scheirer, K.E., Parrish, R.S., Allison, D.B. and Page, G.P. (2005). Sources of variation in Affymetrix microarray experiments. BMC Bioinformatics 6, 214.
RECOMMENDED RESOURCES Statistics for Experimenters. Design, Innovation, and Discovery by G.E.P. Box, J.S. Hunter, and W.G. Hunter (2005) is a new edition of their classical work on experimental design. It is an excellent presentation of the concepts of experimental design with many illustrative examples from different areas of experimentation. W.M. Wooding’s (1994) book is a valuable source of information for the planning of clinical trials.
G.T. Sica’s (2006) about bias in research studies is recommended because it is a comprehensive presentation of all possible biases that may occur in the context of genomic and biomarker studies. Kerr and Churchill (2001a, b) are recommendable for planning dualchannel experiments.
CHAPTER
24 Biobanking in the Post-Genome Era Theresa Puifun Chow, Chia Kee Seng, Per Hall and Edison T. Liu
INTRODUCTION The common definition of a Biobank is a collection of specimens of human bodily substances linked, or linkable, with personal data and information on their donors “that are intended for medical research, set up with a broad purpose, accessible to a variety of research groups, contain genetic data and are of substantial size” (Knoppers and Newton, 2001). In addition, a Biobank could also be described as the system for the efficient collection of information associated with the biological materials, such as lifestyle factors, therapy, etc. Previously, collections in tissue banks were mainly that of remnants from surgical and clinical processes to characterize physiological, morphological, and histological abnormalities. The collections were mostly ad hoc, clinic- or hospital-based. The processes and procedures vary widely, and many tissue banks are under-funded. Biobanks are perhaps best viewed as part of the continuum of tissue repositories. Starting with small investigator sponsored tissue banks, to institutional tissue repositories, to national repositories or biobanks, the difference is one of scale. Rapid advances in genomics-based technologies have radically changed the tissue banking practices in the last decade. Moreover, studies require large numbers to achieve statistical power and the management of thousands to tens of thousands of samples per study raises challenges based on the scale of the operations. Thus, larger biobanks with dedicated infrastructures
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 284
have superseded the investigator-operated freezer sample storage. Such large sample numbers requires automation for laboratory processes with informatics frameworks to manage sample processes, inventories, and associated data. Biobanks are fundamentally systems for information management. A prerequisite for a management system is standardized procedures and ontology (common language) for all levels of information. Quality control and information security systems with data encryption and access control are an essential part of this enterprise. Above this all, transparent governance in the management and distribution of these tissue resources are needed to protect donor privacy and confidentiality and to ensure good utilization. This also means that an open but controlled access to these tissue resources be made available to the larger scientific community. This data-intensive repository of individuals followed over time is also often thought of as a large-scale longitudinal cohort study. The distinction between such longitudinal studies and a biobank is becoming progressively blurred. However, a biobank is tissue centric whereas a standard longitudinal cohort study can be primarily based on epidemiological data. The greatest scientific power, of course, is derived when the two are combined. A review of all current biobanking practices is beyond the scope of this chapter. Instead, we wish to illustrate principles common to biobanking and discuss hands-on issues facing the contemporary biobank, specifically describing our experience in Singapore as an instructive model. The accelerated nature of development in Singapore provides a unique perspective on the Copyright © 2009, Elsevier Inc. All rights reserved.
The Evolving Face of Biobanking
interconnections between tissue repositories, the health and legal systems, and social culture.
THE BIOBANKING EVOLUTION Biospecimen collection has evolved from that of supporting basic discovery type of work where minimal clinical annotation is required, to the translational phase, where validation and clinical development is pursued. The challenge today is to create prospective, longitudinal studies with repeated measurements and blood samplings that allow studies of early markers of disease based on proteomics, analysis of treatment prediction, leading to individualized medicine, and most likely molecular phenotypes substituting the prevailing 20th century disease classification. Such a prospective, longitudinal cohort study allows comparison of large numbers of people who develop a disease with those who do not and a comparison with those that survive the disease to those who do not. Comparisons need to be based on knowledge of genetic alterations, high-quality information on nongenetic factors, and the interaction between the two of them. The advantage of a prospective longitudinal cohort study, sometimes (misleadingly) also called a Biobank, are numerous. An inherent strength of a prospective cohort study is the ability to efficiently investigate multiple endpoints. If cancer is used as an example, it is possible to study not only the risk factors for cancer but issues related to cancer survivorship such as local recurrences, distant metastases, second cancers, and quality of life issues. The prospective design adds to the flexibility of the study although a large cohort of individuals is needed to produce reliable results for a range of effects both overall and in small but important subgroups. Prospective cohort studies with longitudinal data collection also have the invaluable benefit of allowing studies of a host of exposures, information about which may be updated over time. Selection bias and population stratification are problems in association studies. Many of these problems could be reduced, but not avoided, by adopting a prospective, populationbased cohort design when setting up a biobank. The quality of a prospective cohort study builds on the prerequisite of rigorous follow-up mechanisms that could trace patients over long periods of time. The final translation into clinical practice lies not only in validation, but in actual proof of clinical utility based on long-term follow up of patient survival status found in public health registries like disease and death registries. Record linkage for research purposes is necessary to complete the full cycle of research data to convince policy makers and health care providers the value of new targeted therapies. However, as registries contain personal identifiers, special precautions are needed to be in place to ensure participant’s confidentiality and to encourage public participation. Since public health registries are government initiatives, mainly for epidemiological reasons, a biobank that has the capability to access public registry’s data will be a highly valuable resource for researchers’ effort in advancing genomics medicine. An advantage of a prospectively collected biobank is the possibility to generate results that are better generalized relative to
■
285
those derived from clinical trials and case-control studies because of the reduction in potential selection biases. These types of studies at most times address narrow ranges of patients regarding age, exposure, outcome, performance status, etc. By contrast, studies designed to test a particular hypothesis, such as case-control studies, have advantages but can only consider an a priori hypothesis and are thus somewhat inflexible. Some methodological concerns have to be taken into consideration when establishing a longitudinal cohort study.The responsibility, work, and expense of creating a cohort study may be best taken on by an organization set up for this specific purpose. Individual research groups will be given access data on scientific merits. The approach is in the interests of the public, the scientific community, and the funding agencies because a much wider range of research groups can be given access to information. This is one of the fundamental applications of a national biobank.
THE PAST IMPERFECT RESOURCES Two separate reports relating to a survey of tissue banking of 147 institutes in six European countries conducted from 1999 to 2001 (Hirtzlin et al., 2003) and another, a National Cancer Institute (NCI) USA commissioned report by the Rand Corporation of the existing tissue repositories in the United States of America (Eiseman et al., 2000), both highlighted the heterogeneity of tissue banks in size and purpose and the lack of standard practices across the tissue banking arena. Despite the NCI’s investments in biorepositories of $50 million per year involving 125 programs and projects that included biospecimen collection of 4 million human biospecimens in FY2003 alone, the major obstacle has been the lack of biorepository and data standards to enable the mounting of large scale studies from existing pools. The existing state of variation in consenting methods, accession policies, management principles, and operational processes affect sample quality and therefore interpretability, processes that do not provide useful biomolecules for analysis, the lack of common standard operating procedures (SOPs) or quality assurance/quality control (QA/QC) measures in sample handling and processing, the lack of a common database to facilitate common access to inventories and clinical data on available biospecimens present significant hurdles to mounting large scale or cross-comparison studies of statistical value.
THE EVOLVING FACE OF BIOBANKING Since 2003, a series of international biobank meetings and invitational workshops brought together repository managers, researchers, biotechnology executives, epidemiologists, information technology (IT) experts, funders, patient advocates, ethicists, regulatory bodies, and policy makers to actively engage in discussions toward understanding of the existing state of biobanks. Outputs of the meetings were presented in several reports between 2004 and 2005 on biobanking summits (Baig and
286
CHAPTER 24
■
Biobanking in the Post-Genome Era
Doyle, 2005; Louie and Reeve, 2005; Melnikova et al., 2005; Zimmerman et al., 2004) and the best practice guidelines were produced by the International Society of Biological and Environmental Repositories/ISBER (ISBER, 2005), Organization for Economic Co-operation and Development/ OECD (OECD, 2004) and the NCI, USA (NCI, 2006). Parallel to this development, large private or public biobanking initiatives have been established (Figure 24.1). These include collections focusing mainly on DNA samples by a single institutional entity as in deCode Genetics Inc. to broader collections that include serum/plasma for proteomics, and from single collection agencies to the establishment of large networked biobanks Position Papers & Best Practice Guidelines 1993–2002 2003 National Biospecimen Blueprint (NBN) USA 2004
2005 ISBER Best Practices 2006
2007
Biobanks Initiatives & International Harmonization 1993
NCTR Rand Report
IDC Report, Summit I,II OECD Guidance for Operation of Biological Resource IDC Report, Summit III,IV First Generation Guidelines for NCI supported Biorepositories (USA)
UK Biobank (1999) EGP (2001) KI Biobank 2003
EPIC (1993) HUNT-2 (1995) DeCode Genetics, NMCC (1999) ISBER (2000) Singapore Tissue Network (2002) Biobank Japan
2004 2005
2006 OnCore UK
2007
P3G Consortium CaBIGTM Prostate SPORE NBN Pilot SCCS Marble Arch Group UK Biobank Launch Taiwan Biobank CCB (UK) GAIN WAGHP Life Gene CARTaGENE
Figure 24.1 The biobanking evolution. Examples of key position papers on principles and best practice guidelines on biobanking and parallel development of world-wide biobank initiatives from the1990s. Biobank Japan; Cancer Bioinformatics Grid (caBIGTM); CARTaGENE; Confederation of Cancer Biobanks, UK (CCB); deCode Genetics, Inc.; Estonia Genome Project (EGP); European Prospective Investigation into Cancer Nutrition (EPIC); Genetic Association Information Network (GAIN); HUNT-2, Norway; IDC Reports; International Society for Biological Environmental Repositories (ISBER); Karolinska (KI) Biobank; Life Gene; Marble Arch Working Group (MAWG); National Biospecimen Network Blueprint (NBN); NCI First – Generation guidelines for NCI-supported repositories; National Cancer Tissue Resource (NCTR); Norwegian Mother and Child Cohort; oNCORE UK; Organization for Economic Co-operation and Development (OECD); Prostate Specialized Program of Research Excellence (SPORE) NBN Pilot; Public Population Project in Genomics (P3G); RAND Report on case studies of human tissue repositories: Best Practices; STN; Taiwan Biobank; Western Australian Genome Health Project (WAGH) (see References).
as proposed by the National Biospecimen Network Blueprint (USA), and piloted by the NCI’s Prostate Specialized Program of Research Excellence National Biospecimen Network Pilot (Friede et al., 2003; Prostate SPORE NBN Pilot) which is comprised of 11 centers in the United States linked through a federated informatics system, OnCore UK (oNCORE UK) for cancer clinical trials, Biobank Japan (The Biobank Japan project) for pharmacogenomics research (300,000 participants), and the UK Biobank (UK Biobank) for prospective cohort studies (500,000 targeted participants). Moreover, the range includes private biobank collections and large national biobanks with responsibility for national scale projects as in Singapore Tissue Network (STN) and Karolinska (KI) Biobank. The size of biobanks is increasing, and the diversity of biosamples collected is broadened for “omics” research. There is a trend toward collecting more extensive and higher quality clinical data and open source biobanking where IT tools are being developed for sharing encourages collaboration as seen in the US NCI Cancer Biomedical Informatics Grid (caBIGTM) and the p3G Consortium effort toward sharing and harmonization. Active alliances like the Confederation of Cancer Biobanks (CCB) in the United Kingdoms have been initiated. For example, the Public Population Project (p3G Consortium) has been established to facilitate international information exchange. Exemplifying this trend, megascale prospective cohort studies are being planned or have begun such as the Western Australian Genome Health Project (WAGHP), Life Gene (Sweden), Singapore Consortium of Cohort Studies (SCCS) and CARTaGENE (CARTaGene project). Other examples of large biobank initiatives include the European Prospective Investigation into Cancer Nutrition (EPIC), Estonia Genome Project (EGP), HUNT-2 (Norway) and the Norwegian Mother and Child Cohort (NMCC). The principles of clinical trial design with stringent requirements for clinical data is followed for the pre-clinical and/or laboratory research, but the challenges lie with the infrastructure requirements for establishing and maintaining biobanks of this size and scope, and in the distribution of large numbers of samples with consistent quality for various downstream analysis.
EXISTING MODELS: BIOBANKING IN EUROPE AND THE USA The following examples have been chosen for the level of organization and progress. The UK Biobank The UK Biobank is one of the number of prospective populationbased studies underway and aims at collection of baseline blood samples and data from 500,000 individuals aged 45–69 years with a follow up of at least 10 years. The project is supported by the Medical Research Council, Wellcome Trust, Department of Health, the Scottish Executives, and the Northwest Regional Development Agency with combined funding of £61 million for the 5 year collection phase. The objective is to create a research
Existing Models: Biobanking in Europe and the USA
resource for future investigations on the effects of genetic, environmental, and lifestyle factors on major diseases. A baseline questionnaire with a limited number of physical measurements is taken, and a 50 ml blood sample is taken and used for baseline clinical analysis and archiving. Subjects will be followed through general practitioners (GP) records and through linkage to public records for mortality, hospital admission, and cancer statistics. Issues such as consent, confidentiality, and security of the data are guided by an Ethics and Governance Framework overseen by an independent council in accordance with the UK Data Protection Act 1998. It is estimated that there would be 750 participants per day, generating more than 3750 tubes of blood and 750 tubes of urine per day. The large number of participants and samples requires specific IT systems to handle the recruitment, collection, and sample processing logistics. The UK Biobank has engaged Thermo Nautilus Laboratory Information System for sample hierarchy tracking through system of barcodes, and special automated setups with high throughput blood fractionator system (RTS Vision System – RTS Life Science UK). Sample storage is split into 80°C automated (and long-term archival in liquid nitrogen tanks). The 80°C storage is housed in a 20°C environment where samples are retrieved with robotic arms. Although hugely expensive, with the cost £4 million (for a capacity of 9 million cryovials), the automated retrieval of samples by automated robotic arms is the only solution for preventing repeated freeze thawings and is the state-of-the-art solution for preserving sample integrity for proteomics studies. Biobanks, like cohort studies, estimate the number of individuals needed to be recruited based on incidence rates for the diseases of interest. Therefore the more rare the disease, the larger the biobank collection that is required to make any observation statistically robust. Even with the large number of participants, it is estimated that the UK Biobank is just large enough to be of value as a stand alone research structure since for less common diseases like stomach cancer, it has been estimated that it would take 29 years to achieve 2500 cases. Thus, there is a call for cooperation amongst international biobanks to achieve 2–3 million recruits to have sufficient statistical power for research on rare disease within a reasonable time frame (Burton, 2006; Manolio et al., 2006). The proposed harmonization of the UK Biobank and other national cohort study into “The Last Cohort” (Potter, 2004) is one such call for coordination amongst national biobanks. The Karolinska Institute Biobank The Karolinska Institute Biobank (Sweden) is a national, noncommercial tissue resource that links biological samples and the large databases containing phenotype and genotype information of the individual donors. This information includes lifestyle, psychosocial, and clinical data in addition to molecular data, all of which require the development of appropriate database models and ontologies. An important task therefore has been to define, structure, and standardize collection of different kinds of information. Arising from these efforts is the Biobank Information Management System (BIMS), which is responsible for the linkage of the biological samples and databases containing information
■
287
Other Biobanks
DBs
Lab robot LIMS
Freezer
BIMS
Web interface
Figure 24.2 The Karolinska BIMS. The information management system designed to link biological samples information from the LIMS to databases (DBs) containing information on the individual sample donors.
on the individual sample donors following de-identification (Figure 24.2; Ölund et al., 2007). The KI Biobank also seeks to take advantage of communication tools such as the internet, and mobile phones for quick and cost-efficient assembly of patient data, and follow-up information. A key feature of the KI Biobank is in the ownership of the tissue and information. Due to the legal considerations, the owner of the data and at the same time the responsible authority is the KI Biobank, not the individual principal investigators. Nearly all Swedish patients use the tax-funded National Health Care system and therefore any material generated through this system shall not be restricted in use, but will be accessible solely based on scientific merits of a suggested project. The Prostate Specialized Program Of Research Excellence National Biospecimen Network (Prostate SPORE NBN) Pilot The Prostate SPORE NBN Pilot plans to establish a common biospecimen coordination system and informatics infrastructure/data-sharing for multiple Specialized Programs of Research Excellence funded by the US NCI SPOREs located at 11 different institutions. A key innovation is the National Biospecimen Network Blueprint (Friede et al., 2003) which spells out key concept that calls for a national “best practices” based tissue distribution of high-quality biospecimens and linked data to support and reduce variability through the use of standardized approaches for collecting, processing, storing, annotating, and distributing biospecimens. Initially, the program will support an Inter-SPORE Prostate Biomarker Study (IPBS) for biomarker validation, but biospecimens and the data collected will ultimately be available to qualified non-SPORE researchers. The caBIGTM is a voluntary network or grid connecting individuals and institutions to enable the sharing of data
288
CHAPTER 24
TABLE 24.1 ● ● ● ●
● ● ●
●
●
■
Biobanking in the Post-Genome Era
Key features of a centralized biobank
Mostly publicly funded initiatives of national and/or international scale. Complexity of multiple stakeholders requires implementation time. High set up cost. Complex technical requirements due to diversity of projects for example ability to process different sample types, molecular processes and clinical data. Central committees for policy, accession, and ethical oversight decisions. Quality management and operational efficiency. Standardization/harmonization for operating procedures (SOPs), collection workflows, clinical annotations, and IT frameworks. Robust information technology infrastructure for high volume sample tracking, processing, inventory, withdrawal, and automation management. Data privacy and confidentiality protection.
and tools. The Prostate SPORE NBN Pilot will interact with caBIGTM by using plug-in applications provided by caBIGTM and will share application development with caBIGTM. This pilot project adopts a federated model to address interoperability instead of requiring a single centralized database or a single standard system for all centers (Table 24.1). All personally identifiable information will be removed, and data will be submitted using common data elements. This is becoming the standard organizing principle for consortial databases. The resources and data moved to the interoperability layer will be organized and tracked in four modules: a searchable Web catalog; protocol tracking and biospecimen tracking modules, and a common results bank. Besides the issue of scalability, the information technology intensive approach is needed for multi-site collections separated by significant geographical distances.
SINGAPORE’S NATIONAL BIOBANK AND NATIONAL ASPIRATIONS IN BIOMEDICAL RESEARCH Singapore’s Biomedical Sciences Initiative In the year 2000, the government of Singapore announced the launch of a Biomedical Sciences Initiative to develop biomedical sciences as the country’s fourth pillar of economic growth, alongside the already successful electronics, engineering, and chemical manufacturing industries. An associated goal was to transition Singapore from a manufacturing-based nation to one focused on knowledge creation. The first phase of development (2000–2005) concentrated on establishing a firm foundation of basic biomedical research in Singapore and was supported by a S$4 billion fund. Five research institutes were established with the core public research capabilities in the areas of bioprocessing; chemical synthesis; genomics and proteomics; molecular
and cell biology; bioengineering and nanotechnology; and computational biology. The “extramural” research performed at the hospitals and the National University Medical School was bolstered by a significant increase in research funds as well. The initiative used a “systems approach” to create a total infrastructure to support biomedical industry. The components for success is based on high-upfront investment from the government to build infrastructure, fund research, develop human capital, attract foreign investments through tax incentives, foster local collaborations, and aggressive protection of intellectual property rights. The added advantage is Singapore’s flexibility as a small but well organized country willing to modify plans when needed. The vision was bold as it was risky. It was during this phase that Singapore’s national tissue repository, the STN was established. In 2006, with the success of Phase I of this initiative, the Government increased the commitments in Phase 2 (2006–2010) to S$12 billion over the next 5 years of which S$1.4 billion was earmarked to support translational and clinical research. Singapore’s Bioethics Governance National tissue repositories and biobanks function appropriately only in an environment with ethical rules and transparency of decision-making. In medical research, the absence of a moral framework in its governance places a nation on a dangerous slippery slope to unacceptable behavior. Governmental and academic leaders of this Biomedical Sciences Initiative recognized this danger very early in their planning. Biomedical research has social implications in its conduct and will generate outcomes that often challenge the ethical and legal framework of the country. So, also in 2000, the Singapore Cabinet appointed a high level Bioethics Advisory Committee (BAC) to openly deliberate and develop guidelines for the ethical, legal, and social issues for biomedical research in Singapore. The BAC guidelines are developed through a process of public consultation and its reports are then presented to the Steering Committee on Life Sciences, a panel established to oversee the “holistic” development of biomedical sciences in Singapore engaging the ministries of education, health, and industry and trade (Figure 24.3). The establishment of the BAC was perhaps one of the most important foundation pieces in the success of Singapore’s Tissue Network and national biobank. Without this national and governmental consensus, the STN’s work would inevitably diverge from national sensibilities. Among the reports already released, several are pertinent to the function of this national tissue repository, specifically surrounding the use of human tissues and human subjects research: Human Tissue Research (BAC, 2002), Research Involving Human Subjects: Guidelines for institutional ethics/review boards (IRBs) (BAC, 2004), Personal Information in Biomedical Research (BAC, 2007) and Genetic Testing and Genetic Research (BAC, 2005). Some of their deliberations already have had a significant impact in providing the necessary clarity for researchers to move forward. Of note, is the stand of the BAC that affirms the importance of biomedical research in addition to the primacy of patient rights. Specifically, it encouraged organizational flexibility and the use of information technology in resolving issues surrounding patient confidentiality.
Singapore’s National Biobank and National Aspirations in Biomedical Research
BAC’s recommendations process Consultations
Public
Religious & Community Groups
Deliberations
Recommendations
Examine current practices in major jurisdictions BAC
SCLS IAC
Consider responses from public and consultation parties
Professional Groups includes medical, scientific and legal
Public
International experts
Figure 24.3 The BAC of Singapore and its recommendation process. Appointed by the Cabinet and reporting to the Life Science Ministerial Committee (LSMC) with the expert advice of an international advisory committee (IAC), the BAC examines current ethical, legal and social issues arising from research on human biology and behavior , consults, deliberates and makes recommendations to the ministries of education, health, industry, and trade.
As an example, because Singapore is a small country that has a limited pool of clinical research expertise, IRBs within the hospitals had difficulty in continuously retaining high-quality reviewers. A solution devised by one of the hospital groups after the BAC recommendations was to establish four “Domain Specific” IRBs that covered all six hospitals within the health care group (Domain Specific IRB, 2004). Each application is triaged into one of the four broad but related disease subgroups thus maximizing the utilization of expertise within scientific domains for the review process. This innovation resulted in improved review times and review quality. Moreover, this arrangement has the potential of facilitating multi-institutional human subjects investigations pertinent to a national biobank. Singapore’s National Biobank: The STN The establishment of a publicly funded national tissue repository was considered a strategic step for the fledgling Singapore’s biomedical science initiatives. The STN was formed in the year 2002 as a joint initiative between the Ministry of Health (MOH), the Genome Institute of Singapore (GIS) and the Biomedical Research Council (BMRC) of the Agency of Science Technology and Research (A*STAR), then the key funding agency for public research in Singapore. The concept was that for Singapore to succeed in biomedicine, a robust and ethical tissue repository platform is needed to be available to laboratory and clinical investigators. Historically, like in many countries, tissue repositories were managed by small groups of academicians
■
289
supported by hospitals or universities commonly with meager funds. The planners and advisors of the Biomedical Sciences Initiative recognized that a larger, more professional unit was necessary to support the research demands and to enhance the quality of tissue repositories within the Republic. Moreover, as a small compact nation, Singapore can centralize many of the biobanking operations given the simplicity of sample logistics (e.g., accession and distribution) that would be challenging in a geographically larger country. Thus, both the need and the advantage of such a national tissue repository in Singapore were evident. In the initial planning for the STN, several key issues were identified: expertise/infrastructure, governance, flexibility, and patient confidentiality/data security. Expertise It was recognized that, in Singapore, there is little experience and expertise in starting and maintaining a high capacity national tissue repository at the beginning. Though several overseas companies offered to provide a complete service for the nation, the decision is to build the indigenous capabilities rather than outsource the operation. However, to accelerate the development, outside help was sought. A private genomics company specializing in tissue repositories, Genomics Collaborative Inc. (GCI) (Boston MA, now part of SeraLife Inc.) was contracted to train personnel and to provide SOPs for the start-up phase of the STN. As a commercial provider for tissues with annotated data, the GCI had developed one of biobank’s first Laboratory Information Management System (LIMS) to manage and track samples from “cradle-to-grave” and implemented stringent ethical policies and good QC processes for tissues and data. The GCI was ranked highly in a study on Human Tissue Repositories in the USA conducted by the Rand Corporation (Eiseman et al., 2003). Within 18 months, once the establishment of the STN was secure, the transition into independent operations went smoothly while additional designs to execute sample processing and improvement for more sophisticated tracking systems through LIMS continued. Though a major goal of the STN was to develop a national biobank for a large longitudinal cohort study, the reality was that smaller principal investigators (PIs) led projects that would constitute the majority of activity in Singapore for the foreseeable future. To this end, we built a technical infrastructure with capabilities to: (a) support a variety of projects (Figure 24.4), (b) expand capacities as needed, (c) provide quality sample management, and (d) foster collaborations, eliminating unnecessary duplications and wastage. The STN’s infrastructure is characteristic of a post-genome biobank and is distinguished from earlier tissue repositories by the engagement of robotics and information technology to manage large sample numbers, to link phenotype data with sample inventories, and to safeguard privacy and confidentiality. Tracking Samples Through a LIMS The core of any national biobank is its sample tracking and tissue archiving capabilities. All samples within the STN are tracked by a system of barcoding in the LIMS from kit building
290
CHAPTER 24
■
Biobanking in the Post-Genome Era
National Pool
Public
Diabetes cohort
Private Cancer specific groups (BCC, POG)
Multi-ethnic cohort S’pore Prospective Study S’pore Malay Eye Study
Twin
Tissues Blood DNA/RNA Buccal samples/saliva Immortalized cell lines Urine
Myocardial Infraction Study
Identifiers COLLECTION SITE
Tardive Dyskinesia
STN
Dengue Schizophrenia
Stroke Atrial fibrillation
Figure 24.4 STN collections. The collections of the STN constitute a national pool for researchers in the areas of cancer including unscheduled banking of a majority of cancer types to more specific cancer tissue like basal cell carcinoma (BCC) and leukemias from the pediatric oncology group (POG), cohort studies (diabetic and multi-ethnic cohorts), several disease specific projects and twin studies.
to distribution, eliminating transcript errors. Both the KI Biobank and the STN uses a system of coding and barcoding to track samples, and samples are identified with sample numbers and not personal identifiers. The LIMS connects workflows and processes, and captures all sample related data allowing instant querying of sample status from receiving through processing to archiving, aliquoting and distribution. LIMS, originally invented for QC purposes, has been widely adapted for the managing and manufacturing purposes and has recently been applied to biobanking as a solution to organized sample tracking. There are only a few commercial vendors offering LIMS software and many biobanks have resorted to customized solutions. However, the commercial solutions as offered by Thermo-Fisher, used both by the STN and the UK Biobank, and LabVantage, used by KI Biobank, represents easily configurable solutions to accommodate changes in workflows for a variety of sample types. Different from “home grown” customized solutions, they offer a multitude of benefits, including standardization on a single solution across multiple laboratories and adhering to more stringent regulatory requirements and validation (Perry and Reeve, 2004) (Figure 24.5). Work Flow Conceptually, the biospecimen collections encompass two major formats that must be handled differently: Unscheduled
Phenotype data
Clinical data
Specimens
Psoriasis
Hepatitis B • • • • • •
Pre-acquisition
Disease specific
Post-acquisition
Singapore Consortium of Cohort Studies (SCCS)
Cancer
DPF -Office
Phenotypic DB
Barcoded kits Sample log in Assign workflow
Work lists LIMS
Robots
Process Test & Storage Distribute QA Approval result QC entry 20°C, 80 °C, liquid nitrogen, security alarm, back-up site
Figure 24.5 Specimen and clinical data workflows at the STN. The STN collection work flow: Barcoded kits are sent to collection sites. Personal identifiers are sent with sample barcodes to the Data Protection Office that maintains link. Coded samples are processed with SOP with QC and QA processes. Workflows are assigned by the LIMS and sample updates are uploaded into the STN phenotypic database (DB) for display to researchers.
collections – defined as tissue collected in the course of standard medical care with no defined research hypotheses and collected outside a specific project to answer a scientific question (as in tumor collections present in the hospitals); and scheduled collections – defined as tissues collected as part of a prospective study or program (as in a cohort study or a clinical trial) in which a specific scientific question or goal is addressed. Motivations and incentives are also different. Unscheduled collections are fundamentally based on the altruism of clinicians who agree to participate in tissue collections that are above and beyond their standard delivery of care and their institutions who believe that clinical research is ultimately important to patient care. Scheduled collections already represent the aspirations of a funded cadre of investigators studying a specific disease or scientific question. Of the two, the unscheduled collections proved to be the most challenging to establish. Whereas unscheduled collections of tumor work adequately at the level of a single institution driven by an internal “champion”, the results have been equivocal on a national scale. Globally, with some exceptions, broad-based nationally mandated unscheduled collections of fresh human tumors for academic research have not worked well. In the United States, tissue services such as Cooperative Human Tissue Network or CHTN (Cooperative Human Tissue
Singapore’s National Biobank and National Aspirations in Biomedical Research
Network) that collect fresh tissues without an underlying scientific question function solely as a service entity and are limited to a network of dedicated hospitals. Moreover, the CHTN collects tissue prospectively when a project is negotiated by the administrators of the service with the requesting PI. Therefore, in effect, the CHTN uses a scheduled collection strategy. The difference in Singapore is that most of the cancer surgery in the public sector was performed in two large hospitals approximately 35 minutes apart by car: Singapore General Hospital and the National University Hospital. This compact configuration simplifies the organization of a national tissue repository. Through a negotiated agreement, the STN funded the base infrastructure (freezers and full time employees/FTEs for data-manager, phlebotomist, and consent nurse) in each hospital for a predetermined number of tumors. To encourage institutional buy-in, a 50/50 rule is applied, where, whenever feasible, half of each processed samples is returned to the contributing hospital’s tissue repository for distribution within the institution. The other half is banked as a national resource for future research use. The STN provides logistics and barcoded kits, and temperature controlled containers for transport. For the national resource, sample processing and storage, aliquoting and distribution is provided by the STN. Access to the banked tissue is determined by the STN Steering committee through a standardized process and review (see below) and requires approval by an ethics review or IRB and proof of funding support for the research. The scheduled collection has worked well in support of investigator driven clinical and epidemiologic studies. A principal investigator with a funded population study will approach the STN and request support for tissue collection, most commonly peripheral blood isolation and DNA preparation and storage. Upon receipt of IRB approval, the STN steering committee will review and approve the support of a study. The STN will bear the costs of accession, processing, sample storage, and distribution. Though originally, scheduled collections also followed the 50% rule (i.e., 50% of the samples “belong” to the principal investigator, and 50% belongs to the national repository), we are moving to one where only a predetermined access policy and process is required. Because of the success of this scheduled collection strategy, we are exploring the conversion of unscheduled tumor collections to a scheduled approach. This means that instead of random collections of tumor tissues, consortia of investigators and clinicians interested in a specific disease, like lung cancer, would propose to collect lung cancer tissues with clinical data and follow-up as part of a “registry” of lung cancers. The cancers will be stored at the STN. Fifty per cent of the tissues will be used by the consortium to perform molecular studies previously approved, the other 50% will be archived for potential future collaborations. In this manner, we believe that the impact of the tissues will be significantly augmented since they will be accompanied by both clinical information and molecular/genomic data (Table 24.2). The Governance The STN is run by an executive director and managing director and reports to a Steering Committee. Like many biobanks, the
TABLE 24.2
■
291
Scheduled versus unscheduled collections
Scheduled collections ● ● ●
●
Associated with specific research objectives for example disease specific collections in case-control studies. Study protocol defining collection period, target donor size, and demographics. Detailed clinical data and analysis with follow up on treatment outcomes according to study protocol and research objectives. Informed consent required.
Unscheduled collections ●
● ● ●
Regular collections without specific pre-designed research objectives for example broad-based tumor collection in hospital. Study protocols are submitted in future applications after biospecimens have been banked. Less specific clinical data collected as research objectives are not defined. Broad/general consent generally practiced with stipulations for further review and approval from Institutional Review Boards or appropriate Ethics Committees for future studies.
STN structures its central governance around participation from all respective stakeholders but is unique in that because the STN functions as the national tissue repository, we have diverse responsibilities and a range of stake holders. Thus, the oversight body or Steering Committee consists of policy and decision-makers representing executive members of the funding agency (BMRC), the MOH, research institutes, medical academic centers, universities, hospitals, IRB, the Economic Development Board (EDB), and the BAC. Because of the high level involvement, decisions are made at a level that can have far reaching implications. The Steering Committee votes on the approval of national scale projects based on internal and external reviews, subjected to approval from relevant ethics review boards, and the availability of peer-reviewed research funding. Emphasis is placed on ensuring focus and harmonization to maximize resources critical for a small country like Singapore (Figure 24.6). Confidentiality Protection: Data Privacy Issues in Biobanks A fully-functional national tissue repository/biobank collects and centralizes tissues and data from patients and normal research subjects. These are usually collected within the context of a doctor–patient or a researcher–study subject relationship. The collection of such data and tissue are subjected to approval and review by ethical review committees and or prevailing legislations/regulations of governmental authorities. Informed consent is the acceptable standard applied to such collections. To maximize the value of the collected data and tissue, some form of (electronic) record linkages is also included in the consent. Ultimately, the aim of a biobank is beyond being a data management and storage center for individual researchers.
292
CHAPTER 24
■
Biobanking in the Post-Genome Era
EDB
MOH
BAC Clinical experts
BMRC Steering Committee
Director
Collaborators
IRB Scientific advisory board
Research grant approval
IRB approval
Withdrawal
Withdrawal
Secretariat Management
Operations
QA
IT
Admin.
Facilities Safety
Phenotypic DB
LIMS
Lab QC
Figure 24.6 STN governance and infrastructure. Governance of STN is through a Steering Committee comprising members from the Ministry of Health (MOH), the Biomedical Research Council (BMRC), Collaborators (hospitals, collection sites), IEB Chairs, STN Director, clinical experts, BAC and executive members of the Economic Development Board (EDB). Key departments under the management include Laboratories (Lab), QA, QC, Informatics supporting STN’s phenotypic database (phenotypic DB) and LIMS and Administration (Admin).
To maximize its utility, the information collected should be a national resource for the research community for future research. To this end, the data and tissues need to be used for purposes and biomarker investigations beyond what was proscribed in the original informed consent. Technically, each patient/respondent would have to give reconsent for every new hypothesis to be tested on the cohort. Obviously, this would be impracticable for studies of the magnitude of a national cohort study. More recently attitudes toward reconsent requirements have changed both in Europe which favors broad consent and in the United States of America where the definition of non-identifiable samples have been expanded to include those that have been previously considered coded. Ethicists have argued in favor of using general consent after IRB approval together with the right to “opt out” as the preferred method (Elger and Caplan, 2006). But broad consents still have legal and ethical issues that are in debate (Hansson et al., 2006). For example, medical data collected within the context of a doctor–patient relationship will be made available to non-doctors. Furthermore, the initial consent between doctor and patient or researcher and study subject is legally seen as a contractual agreement. Further transfer of data/tissue to researchers downstream is then viewed as a new agreement. In this case,
disconcertingly, the later researchers would have no legal obligations to the original patient/respondent. Though the solutions to these issues will be dependent on the existing legal and ethical framework in the community contributing to the biobank, the technical solution involves some form of de-identification. The challenge is often not in the IT arena at the ground level but in the policy framework that needs to be drawn up at the macro-level. Unfortunately, such macrolevel policies are often hindered by a lack of understanding of technical issues and/or the imagined fears of privacy invasion. Given adequate technical safeguards and a strong data protection legal framework, the future benefit of science to humanity should outweigh the potential threat of personal privacy violation. In Singapore, the BAC has given national consensus statements that are now being crafted into legislative protections. Important in these regulations is the acknowledgement that scientific research and progress is important to our society, and that information technologies should be used help safeguard the privacy of this information. Conceptually, though we deemed that a national tissue repository should be centralized for efficiency, quality, and economy, the accompanying data repositories of clinical and epidemiologic information should be distanced from the tissue collection and should be constructed as a distributed but linked collection of databases. This is because of the recognition that information in these databases must be updated and managed by the local investigators. Moreover, distributed but federated databases pose less of a security risk than a single massive database of national proportions. The key, however, is that these databases should be linked. For security, a stringent de-identification process needs to be adopted. Unlike aggregate data, data from individuals with the potential for research use can be divided into two groups: ● ●
Personal identity data Research data
Personal identity data consist of data items that singly or in combination could potentially identify a specific individual. It is accepted that name and unique personal identification number (e.g., passport numbers, social security numbers, and other form of national registration numbers) are considered personal identity data. Some would extend the list to ethnicity, date and place of birth, and gender. In rare situations, simple combinations like date of birth (DOB) and diagnosis of an extremely rare condition may potentially reveal the identity of an individual. However, it is not necessary to invest vast resources to build a system that claims to be “fail-safe” for such cases. The performance of such “fail-safe” systems is cumbersome and tends to dissuade users. Prudent judgment is needed to balance efficiency and privacy. De-identification is a process where personal identity data are separated from the research data. There have been many different terminologies for the concept of de-identification: annonymized data, pseudo-annonymized data, partially de-identified
Singapore’s National Biobank and National Aspirations in Biomedical Research
data, etc. Conceptually, it will be easier to have three levels of de-identification: 1. Completely identifiable data: Personal identity data and research data are stored as single electronic or paper records. The data items may be coded or reversibly encrypted for security reasons. The personal identity and research data are physically linked after the items are decoded and de-encrypted. 2. Completely de-identified data: Personal identity and research data are de-linked in such a way that it is impossible to identify an individual from the research data. This is the same as anonymization. 3. Reversible de-identified data: Personal identity data are separated from the research data but linked via a newly created unique identifier. Each record in the research data is identified by this unique identifier which does not carry any personal identity information. For simplicity, this unique identifier could be termed a “private unique identification number” (PUIN). The personal identity data are linked to PUIN which serves as a bridge between the two datasets. To maximize the value of biobanks and cohort studies, the data should be managed as reversible de-identified data. The datasets and samples sent to researchers should be completely de-identified. However, the data managers of the biobank and cohorts should have some form of capability for reversing the de-identification as new data on the same individual may be obtained subsequent to the initial sample collection. A system that allows for active and passive follow-up of recruited subjects to a biobank and cohort study will need to meet three main criteria: 1. A trusted third party (TTP)/Data Protection Framework (DPF) Office that holds the link between PUIN and personal identity data. 2. A mechanism whereby the ground operation is partitioned such that no one is able to have all three sets of information: PUIN, personal identity data, and research data. 3. A mechanism of record linkage with external agencies such that they do not need to release completely identifiable data. Recruitment of subjects for a cohort study could be in the community or a clinical setting. Field workers will therefore have access to personal identity data and research data. Therefore, they should not have access to the PUIN. Furthermore, the personal identity data and the research data should be separated early, if possible, just after collection. The personal identity data should be sent to the DPF Office with the STN kit number where a PUIN is generated (if the subject has not been previously recruited) or the existing PUIN is retrieved. The “deidentified” phenotype and clinical data and tissue is sent to the biobank with the STN kit number. The PUIN is then sent by the DPF Office to the biobank with the STN kit number where the PUIN is now used to identify the subject. In this way, longitudinal data of the same subject could be constructed without revealing the actual identity of the subject.
■
293
The DPF Office can also serve as a portal for record linkages protecting external agencies from releasing completely identifiable data. The biobank will provide a listing of PUIN of cases that require external record linkage (e.g., follow-up information like death) to the DPF Office. The DPF Office creates a temporary listing consisting of the PUIN, personal identity data, and a DPF number. The DPF number is a unique number used only once (NONCE: number used only once) for the purpose of this linkage exercise. The DPF Office sends to the external agency a listing consisting of personal identity data linked to the DPF number. At the same time, the DPF Office sends back to the biobank a listing of PUIN linked to the DPF number. Once the external agency has retrieved the followup data, it sends back the follow-up data linked with the DPF number, without the personal identity data. The biobank could then attach the follow-up data with its original tissue and data using the PUIN–DPF number listing. Various other combinations and levels of sophistication could be added to this system to balance efficiency and confidentiality (Figure 24.7a, b). In Singapore, this structure is being implemented on a national scale to also assist with similar needs amongst government agencies. For example, the Ministry of Home Affairs holds the birth and death registries, but the MOH manages the disease registries. Though it would seem obvious that the union of the two datasets is an important linkage for research, the technical reality based on organizational history is that Ministerial databanks are rightfully highly protected both physically and by regulation. The structuring of an official Record Linkage Office (RLO) to administer a national TTP policy has been an important step in resolving this issue of citizen data security (Figure 24.8). Lessons Learned from the Singapore Experience In the 4 years since its inception, some lessons can be drawn from the STN experience. Once the decision to develop national tissue and data repositories was made, by far, the major challenges were the social and political issues. The greatest hurdle was in obtaining the buy-in of individual participants for contributing to a “national cause”. Fortunately, the problem was restricted only to the unscheduled collection of tumor tissues. Several solutions were devised: the “50/50” policy where the samples are divided between the STN and the submitting hospital, support from the STN to the local hospitals for personnel and reagents for the establishment of the satellite procurement stations, and the relatively low to no cost for processing and archiving samples. Patience, flexibility, and coordination were needed to overcome obstacles. When large scale projects engaging multiple institutions faced long institutional ethics review process, the establishment of Domain Specific IRBs helped to expedite the review process. When the issue of data-security and confidentiality became a concern, we embarked on a 2-year process of consensus building and problem solving with governmental and academic parties and emerged with the DPF and the establishment of an RLO.
294
CHAPTER 24
STN Kit No Tissue Clinical sample Data forms Form
■
Biobanking in the Post-Genome Era
Data Form Name, DOB, Ethic, etc
DPF Office NRIC-PUIN New PUIN for new subjects Old PUIN for existing subjects Personal data-Sample no. Name, NRIC, Gender, Ethnicity, DOB, address, date of collection, Sample no. (S/N)
Remote site keys in personal data form and scans all bar codes
STN scans in bar codes
Biomedical Record Linkage Office (RLO): National TTP
NUSDPF
NTUDPF
NUS
NTU
DPF
A*STAR
NHGDPF
SingHDPF
NHG
SingHealth
Returns PUIN
SCCS
DPF Office Personal data form PUIN Sample bar codes SAMPLE COLLECTION SITE Tissue sample form Clinical data form Personal data form NO PUIN
No Clinical data form No Tissue data form
STN Tissue data form Clinical data form PUIN NO personal data form
Figure 24.7 STN–DPF operations concept (a) The process of reversible de-identification where remote collection site keys in personal data like name, National Registry Identification Card (NRIC) number, gender, ethnicity, and date of birth (DOB) which are sent to the DPF Office. DPF Office returns a “PUIN” alongside the sample barcode number to STN. The PUIN is the substitute number for subject identity for future processes and linkage is kept at the DPF Office. (b) Segregation of information for increased security and privacy.
We also found that when the national funding agencies are aligned in strategic vision, incentives can be structured to ensure the success of programs. The STN have collaborated with a funding agency, the Singapore Cancer Syndicate to develop disease specific tissue and data repositories (Table 24.3).
Figure 24.8 Biomedical record linkage through TTP. Conceptual approach for record linkage through a TTP record linkage office (RLO). Individual institutes, hospitals, and tissue banks, for example, National University of Singapore (NUS), Nanyang Technology University (NTU), STN, Singapore Consortium of Cohort Studies (SCCS) and from the National Healthcare Group (NHG), SingHealth group each have their own DPF Office that communicates via a common RLO for external record linkage, for example follow-up information from hospital clusters (NHG, SingHealth) and from Ministry of Health disease registry data, and Ministry of Home Affairs’ birth and death data.
THE FUTURE OF NATIONAL BIOBANKS An emerging debate is on how large should be the scale of future biobank efforts. The dilemma in study design was recently highlighted by Willett and colleagues (Willett et al., 2007) who discussed the merits of merging existing cohorts versus a proposed single national cohort of several hundred thousands of Americans. They noted that since 1986, there are over 10 large scale cohorts that totaled 1.39 million donors with 845,000 stored biological samples, many with detailed annotation that could serve as the basis for a national biobank. Collins and Manolio (Collins and Manolio, 2007) responded that such a consortium of cohort studies is “necessary but not sufficient” due to the suboptimal aspects of the merging approach including issues of lack of standardization, and the failure to include new tools for genotypic and phenotypic measurements. In Singapore, we believe that both concepts have merit, but adhere to the Flaubert’s dictum that “perfection is the enemy of the good.” So, rather than to structure a national biobank scaled to several hundred thousand participants, we have begun our biobanking efforts by merging existing cohort studies funded
The Future of National Biobanks
T A B L E 2 4 . 3 Lessons learned from the Singapore experience for establishing a national biobank ● ● ●
● ● ●
●
●
Attention to social and political issues through process of consensus building Buy-ins for contributions to ‘national cause’ Clear incentives for driving participation, for example o equitable sharing between contributors and national bank, o manpower support o low to cost free sample processing, archiving and distribution o other non-research project needs Alignment with national funding agencies’ strategic vision to maximize coordination for large scale national projects Need for mandatory national data protection and legal framework Need for national guidelines for ethics governance for example Singapore’s Bioethics Advisory Committee to establish dialogue with public with recommendations for government policy-makers on specific issues for example consent requirements for research Need for pragmatic approaches for example establishment of Domain Specific Institutional Review Board to streamline ethics approval process Understanding concept of biobanking as long-term infrastructure building rather than project-based initiatives
by the government. This initiative is called the SCCS. The goal however, is to expand into a true national biobank/longitudinal cohort study when the requisite regulatory (such as the national TTP systems), social, and structural capabilities are mature. This transitional strategy is attractive because it is scalable and modular and can engage ongoing studies as well as new ones. For this strategy to work however, standardization and harmonization are keys. Luckily, global efforts are afoot in establishing best standard practices that includes not only sample quality and integrity (as practiced in Good Laboratory Practices), but in the more stringent Quality Management System (QMS) as seen in the adoption of internationally recognized quality standards like the ISO 9001 and the Current Good Manufacturing Practices (cGMP) (ISBER and NCI Best Practice Guidelines). This strong push toward standardization across institutional centers for critical data, but allowing individual variation within local centers, permits important harmonization and standardization as well as experimental flexibility. Progressively, there has been a focus on data access from cohort studies to make the genotype–phenotype data from large population studies available for the larger scientific community to analyze. The movement has been advanced in a number of association and linkage studies. The Genetics Association Information Network (Foundation for the National Institutes of Health, GAIN, 2007): a genotype–phenotype database for all of NIH supported genome-wide association studies, funded by Pfizer Inc. (New York City, USA), Affymetrix Inc (Santa Clara CA, USA), the Foundation of NIH (FNIH, USA), Perlegen
■
295
Sciences Inc. (Mountain View CA, USA), Abbott (Abbott Park IL, USA) and the Biomarkers Consortium, a public–private biomedical research partnership seeks to genotype approximately 18,000 cases in 6 diseases (psoriasis, schizophrenia, diabetic nephropathy, bipolar disease, depression, and attention deficit hyperactivity disorder. This genotype information will be coupled with extensive phenotype information which will be made available to the general research community through an access plan. Similarly, the Autism Genetic Resource Exchange (AGRE, 2007) is a DNA repository and family registry, housing a database of genotypic and phenotypic information on over 800 families that is available to the scientific community. Another similar project is the Wellcome Trust Case Control Consortium (WTCCC, 2005). Launched in September 2005, with the support of €9 million from Wellcome Trust, the WTCCC is a large collaborative network of geneticists, clinicians, statistical geneticists, and technologists across the United Kingdom. Phase I of the study involves the analysis of 675,000 SNP markers in each of eight common diseases (tuberculosis, coronary heart disease, type 1 diabetes, type 2 diabetes, rheumatoid arthritis, Crohn’s disease, bipolar disorder, and hypertension) in 1000 cases, followed by confirmation in Phase II where 5–10% of selected markers will be followed in 1000 new cases for each disease. Control samples will be from a set of 3000 common nationally-ascertained controls from the 1958 British Birth Cohort (~1500) and blood donors recruited by the three national UK Blood Services (~1500). A total of 17,000 samples will be typed with the 500 K Affymetrix chips. It is anticipated that the data from biobank initiatives will ultimately be released in a similar manner. If so, then these changes mirror those that have occurred in high-energy physics or cosmology where massive data collection is the responsibility of a national resource, and the data – which is accessed in an open fashion – is analyzed by the wider PI-based community. Unlike physics, however, the data have personalized information about people whose privacy might be compromised. For this reason, we believe the greatest challenges will be in the arena of information security. If solved, then biobanks not only have great scientific potential, but also represent a significant national resource. That such biobank initiatives can be used to help structure the health care of a nation is seen in the collaboration between the company deCode Genetics Inc. and the government of Iceland. The information from this national catchment provides a genetic understanding of disease origins within a homogeneous population and can be used for devising public health strategies. Though it is not likely that such an exclusive relationship can be or should be structured in larger countries, less commercial and exclusive arrangements will be as beneficial, but will require significant investment from government (e.g., UK Biobank initiative). In such large scale longitudinal studies, each subject volunteer functions as a health sentinel for the country. When genetics are added to this equation, then the possibility of defining genetic risk for specific diseases in a nation is real. For example, the prevalence of known at-risk alleles of diabetes, breast cancer, susceptibility to infectious diseases can be inferred from the frequency of the at-risk alleles in a population.
296
CHAPTER 24
■
Biobanking in the Post-Genome Era
The allele frequencies in specific populations of genes associated with adverse drug reactions are beginning to be used by pharmaceutical companies in the marketing plans for drugs with idiosyncratic reactions. Therefore, biobanks can potentially serve
as an important resource for national health planning. With both molecular and information technologies advancing at such remarkable speeds, it is likely that biobanks will play an even more significant role in the future.
REFERENCES Autism Genetic Resource Exchange (2007). AGRE program description. http://www.agre.org/program/descr.cfm?do=program Baig, S. and Doyle, A. (2005). From Biobanks to Biomarkers: Translating the potential of human population genetics research to improve the quality of health of the EU citizen. Proceedings of a conference held at the Wellcome Trust Conference Centre, Hinxton, Cambridge, 20–22 September. Bioethics Advisory Committee, Singapore (2002). Human tissue research http://www.bioethics-singapore.org/resources/reports2.html Bioethics Advisory Committee, Singapore (2004). Research involving human subjects: Guidelines for Institutional Review Boards (IRBs). http://www.bioethicssingapore.org/resources/reports3.html Bioethics Advisory Committee, Singapore (2005). Genetic testing and genetic research. http://www.bioethics-singapore.org/resources/ pdf/GT%20Report.pdf Bioethics Advisory Committee, Singapore (2007). Personal information in biomedical research. http://www.bioethics-singapore.org/ resources/reports5.html Burton, P. (2006). UK Biobank: how big is “big”? http://www.p3 gobservatory.org/download/ICHG.Aug.2006.ppt#1 Cancer Biomedical Informatics Grid (caBIGTM) http://cabig.cancer.gov CARTaGENE project. http://www.cartagene.qc.ca Collins, F. and Manolio, T. (2007). Merging and emerging cohorts necessary but not sufficient. Nature 445, 259. Confederation of Cancer Biobanks, UK. http://www.oncoreuk.org/ pages/about_confederation.html Cooperative Human Tissue Network (CHTN). http://www-chtn.ims. nci.nih.gov/ deCODE genetics Inc. http://www.decode.com Domain Specific IRB (2004). http://www.research.nhg.com.sg/ Eiseman, E., Bloom, G., Brower, J., Clancy, N. and Stuart, O. (2003). Case Studies of Existing Human Tissue Repositories Best Practices for a Biospecimen Resource for the Genomic and Proteomic Era. RAND Corporation, Santa Monica, California, USA. http://www. biospecimens.cancer.gov/nbn/rand.asp Elger, B. and Caplan, A. (2006). Consent and anonymization in research involving biobanks. EMBO Rep 7, 661–666. Estonia Genome Project. http://www.geenivaramu.ee/ European Prospective Investigation into Cancer Nutrition (EPIC). http://www.iarc.fr/epic Foundation for the National Institutes of Health (2007). Genetic Association Information Network. http://www.fnih.org/GAIN2/ home_new.shtml Friede, A., Grossman, R., Hunt, R., Li, R.M., Stern, S. and Andrew, S. (eds) (2003). National Biospecimen Network Blueprint. Constella Group Inc, Durham, NC.http://www.biospecimens.cancer.gov/ nbn/blueprint.asp Hansson, M., Dillner, J., Bartram, C., Carlson, J. and Helgesson, G. (2006). Should donors be allowed to give broad consent to future biobank research?. Lancet Oncol 7, 266–269. Hirtzlin, I., Dubreuil, C., Preaubert, N., Duchier, J., Jansen, B., Simon, J., Lobato, P., Perez-Lezaun, A., Visser, B., Williams, GD.,
Cambon-Thomsen, A. EURPOGENBANK Consortium et al. (2003). An empirical survey on biobanking of human genetic material and data in six EU countries. Eur J Hum Genet 11, 475–488. HUNT-2 Biobank. http://www.hunt.ntnu.no/index.php?side=english International Society for Biological Environmental Repositories (ISBER) (2005). Best practices for repositories I: Collection, storage, and retrieval of human biological materials for research. Cell Preserv Technol 3, 5–47. Karolinska Biobank. http://www.meb.ki.se Knoppers, B., and Newton, J. (2001). Creation of Population Biobanks: Design and Conduct. The Public Population Project in Genomics (P3G) publication. http://www.p3gconsortium.org Knox, K. and Ratcliff, C. (2002). A Strategic Framework for Establishing a National Cancer Tissue Resource for Cancer Biology and Treatment Development. National Translational Cancer Research Network Coordinating Centre, UK. Life Gene. http://www.lifegene.se/ Louie, A. and Reeve, B. (2005). Biobanks Strategies to Address NextGeneration Challenges for Biobanking. IDC, Framingham, Massachusetts. USA. Manolio, T., Bailey-Wilson, J. and Collins, F. (2006). Genes, environment and the value of prospective cohort studies. Nat Rev Genet 7, 812–820. Marble Arch Working Group. http://www.oncoreuk.org/pages/ MarbleArchWorkingGroup.html Melnikova, I., Reeve, B. and Swenson, M. (2005). Biobanks Collaborating for Cures. IDC, Framingham, Massachusetts. USA. http://www03.ibm.com/industries/healthcare/doc/content/bin/Biobanks.pdf National Cancer Institute, USA (2006). First-Generation Guidelines for NCI-Supported Biorepositories. http://www.biospecimens.cancer.gov/biorepositories/guidelines_full_formatted.asp Norwegian Mother and Child Cohort. http://www.fhi.no/eway/ default.aspx? Ölund, G., Lindqvist, P. and Litton, J-E. (2007). BIMS An information management system for biobanking in the 21st century. IBM Sys J 46, 171–182. oNCORE UK. http://www.oncoreuk.org/ Organization for Economic Co-operation and Development (2004). Guidance for the Operation of Biological Research Centres (BRCs): Certification and Quality Criteria for BRCs. Perry, R. and Reeve, B. (2004). Standardizing on LIMS: TCO and ROI for the Multilab Setting. IDC, Framingham, Massachusetts. http://www.thermo.com/www.thermo.com/com/cda/resources/ resources_detail/1,2166,111652,00.htm Potter, J. (2004). Toward the last cohort. Cancer Epidemiol Biomarkers Prev 6, 895–897. Public Population Project in Genomics. http://www.p3gconsortium.org/ Public Population Project in Genomics (2005). The Open Biobank Project. http://www.p3gobservatory.org/openBiobank.do Singapore Tissue Network. http://www.stn.org.sg Taiwan Biobank. http://www.ibms.sinica.edu.tw/biobank/biobank.htm The Biobank Japan project. http://www.biobankjp.org/
Recommended Resources
The Prostate Specialized Program of Research Excellence National Biospecimen Network (Prostate SPORE NBN) Pilot. http://www. prostatenbnpilot.nci.nih.gov/http://prostatenbnpilot.nci.nih.gov/ The UK Data Protection Act (1998). The Data Protection Act. http:// www.dh.gov.uk/PolicyAndGuidance/OrganisationPolicy/ RecordsManagement/DataProtectionAct1998Ar ticle/fs/ en?CONTENT_ID=4000489&chk=VrXoGe The Wellcome Trust Case Control Consortium. http://www.wtccc.org. uk/info/050928.shtml
■
297
UK Biobank. http://www.ukbiobank.ac.uk/ Western Australian Genome Health Project. http://www.genepi.org. au/waghp Willett, W., Blot, W., Colditz, G., Folsom, A., Henderson, B. and Stampfer, M. (2007). . Nature 445, 257–258. Zimmerman, Z., Swenson, M. and Brock, R. (2004). Biobanks Accelerating Molecular Medicine. Challenges Facing the Global Biobanking Community. IDC, Framingham, Massachusetts. USA.
RECOMMENDED RESOURCES Biobank Websites CARTaGENE project: http://www.cartagene.qc.ca.The CARTaGENE project will map genetic variation in a large reference population of Quebec that integrates and exploits the advantages of the existing Canadian health care and legal system, that is, universal health care, public health care, public health expertise, unique medical identifier numbers, comprehensive health and environmental record linkage with a genealogical/demographic database, extensive privacy and access legislation and a heterogeneous modern population. Cooperative Human Tissue Network (CHTN): http://www-chtn.ims. nci.nih.gov/. The CHTN is supported by the NCI to provide biomedical researchers with access to human tissues. Six member institutions coordinate the collection and distribution of tissues across the US and Canada in six regional divisions. The CHTN specializes in the prospective procurement, preservation and distribution of human tissues for research. Danubian Biobank Consortium: http://www.danubianbiobank.de. Danubian Biobank foundation for public utility in molecular medicine of aging disorders that networks and expands all biobanking activities and the core scientific competences of the Danubian universities, teaching hospitals and rehabilitation clinics between Ulm and Budapest and of associated collaborating regional centers in the field of non-cancer aging disorders. GenomeEUTwin: http://www.genomeutwin.org/. This project will apply and develop new molecular and statistical strategies to analyze unique European twin and other population cohorts to define and characterize the genetic, environmental and lifestyle components in the background of health problems like obesity, migraine, coronary heart disease, and stroke, representing major health care problems worldwide. Karolinska Biobank: http://www.nitordesign.se/biobank/about.php. A national biobank for collection, handling, and storage of human biological material has been established at Karolinska Institutet-KI Biobank. Singapore Tissue Network: http://www.stn.org.sg. The national tissue bank of Singapore. The STN is a non-profit tissue and DNA bank set up as a joint initiative between the Ministry of Health, the Genome Institute of Singapore and the Biomedical Research Council (BMRC) of the Agency of Science Technology and Research (A*STAR) supporting Singapore’s Biomedical Initiatives. UK Biobank: http://www.ukbiobank.ac.uk. UK Biobank is a longterm project aimed at building a comprehensive resource for medical researchers to gather information on the health and lifestyle of 500,000 volunteers aged between 40 and 69. Over the next 20–30 years UK Biobank will allow fully approved researchers to use these
resources to study the progression of illnesses such as cancer, heart disease, diabetes and Alzheimer’s disease. Western Australian Genome Health Project (WAGHP): http://www. genepi.org.au/waghp. The WAGHP aims to capture biospecimens on all consenting members of the living population of WA (approximately 2 million people) and leverage on a number of internationally unique, population-based datasets.
International Harmonization Websites Biobank Central: http://www.biobankcentral.org. A website sponsored by IBM, Affymetrix, Bioaccelerate, and Invitrogen that provides a working group venue, patient and public education programs, and a forum for international collaboration and harmonization of best practices. Confederation of Cancer Biobanks: http://www.info.cancerresearchuk. org/news/newsarchive/2006/october/17771016. Consortium of UK new tissue banks to standardize cancer samples established in October 2006. International Society for Environmental and Biological Repositories (ISBER): http://www.isber.org. The primary goal of the International Society for Biological and Environmental Repositories, ISBER, is to provide information and guidance on the safe and effective management of specimen collections. HealthNex: http://www.healthnex.typepad.com. The mission of this blog is to drive discussion and collaboration on some of the key elements necessary for a global, interconnected health care ecosystem: electronic health records, a new information infrastructure for improving the quality and efficiency of care, and a platform for transforming treatment by marrying personal genomics with care. HuGENet: http://www.hugenet.org.uk/. Human Genome Epidemiology Network, or HuGENet™ is a global collaboration of individuals and organizations committed to the assessment of the impact of human genome variation on population health and how genetic information can be used to improve health and prevent disease. The International Agency for Research on Cancer: http://www.euro. who.int/HEN/Resources/IARC/20030723_1. The IARC is part of the World Health Organization. IARC’s mission is to coordinate and conduct research on the causes of human cancer, the mechanisms of carcinogenesis, and to develop scientific strategies for cancer control. The Agency is involved in both epidemiological and laboratory research and disseminates scientific information through publications, meetings, courses, and fellowships. The International Society for Environmental and Biological Repositories (ISBER): http://www.isber.org. The primary goal of the ISBER is to provide information and guidance on the safe and effective management of specimen collections.
298
CHAPTER 24
■
Biobanking in the Post-Genome Era
Public Population Project in Genomics (p3G Consortium): http://p3gconsortium.org. The p3G Consortium is a not-for-profit international consortium to promote collaboration between researchers in the field of population genomics. It provides the research community with resources, tools and know-how to facilitate data management for improved methods of knowledge transfer and sharing. STROBE (STrengthening the Reporting of OBservational studies in Epidemiology): http://www.strobe-statement.org. International, collaborative initiative of epidemiologists, methodologists, statisticians, researchers and editors involved in the conduct and dissemination of observational studies, with the common aim of strengthening the reporting of observational studies in epidemiology. TuBaFrost: www.tubafrost.org. TuBaFrost is an infrastructure for a European Tumor Tissue Bank. It was developed in a 3-year Fifth Framework Program (FP5, 1998–2002) of the European Commission involving OECI members as participants as well as the EORTC and a Dutch law firm.
Australia, Canada, Council for International Organizations of Medical Sciences (CIOMS), France, Great Britain, Helsinki Declaration, Nuremberg Code, and the Belmont Report.
Legislation on Biobanks Estonia: Human Genes Research Act. Estonia/Government, Estonia, 2000, http://www.legaltext.ee/text/en/X50010.htm. Iceland: Ministry of Health and Social Security, Iceland, 2000, Government regulation on a Health Sector Database, http://www. bibliojuridica.org/libros/5/2292/54.pdf. Latvia: Human Genome Research Law, Latvia/Government, Latvia, 2002, http://bmc.biomed.lu.lv/gene/print/Human%20Genome% 20Research%20Law,%20Latvia.doc4. United Kingdom: Human Tissue Act 2004,United Kingdom/ Government, 2004, http://www.opsi.gov.uk/acts/acts2004/ 20040030.htm.
Legislation Relating to Biobanks Biobanking Tools Website caBIG TM (Cancer Biomedical Informatics Grid): http://www.cabig. cancer.gov/index.asp.caBIG™ is a voluntary network of infrastructure launched by the NCI in February 2004 to create a virtual community that shares resources with tools, and ideas that enables the collection, analysis, and sharing of data and knowledge along the entire research pathway from laboratory bench to patient bedside. Public Population Project in Genomics (p3G Consortium): http:// www.p3gconsortium.org. The p3G Consortium is a not-forprofit international consortium to promote collaboration between researchers in the field of population genomics. It provides the research community with resources, tools and know-how to facilitate data management for improved methods of knowledge transfer and sharing.
Policies Websites HumGen: http://www.humgen.umontreal.ca/int/. An international database on the legal, ethical and social aspects of human genetics, developed as a collaboration between academia, government and industry by the Centre de recherché en droit public at the University of Montreal. PHUGU Genetics Policy Database: http://:www.phgu.org.uk/policydb/index.html. A searchable web-based database of literature on policy development for genetics in health services and health care.
IRB Review Guidance http://www.hhs.gov/ohrp/humansubjects/guidance/reposit.htm, useful website with national and international guidelines including
Australia: Australian Law Reform Commission, Australia, 2003, Essentially Yours: The Protection of Human Genetic Information in Australia, http://www.austlii.edu.au/au/other/alrc/publications/ reports/96/. Canada: Network of Applied Genetic Medicine, Canada, 2003, Statement of Principles on the Ethical Conduct of Human Genetic Research Involving Populations, http://www.rmga.qc.ca/ en/index.htm. France: Opinion no 77 National Consultative Ethics Committee for Health and Life Sciences (CCNE), France, 2003, http://www. ccne-ethique.fr/english/start.htm. Germany: German National Ethics Council, Germany, 2004, Guiding principles of the legal position in Germany for biobanks used in medical research, http://www.ethikrat.org/_english/publications/ Opinion_Biobanks-for-research.pdf. Norway: Norway/Government, Norway, 2003, http://www.ec. europa.eu/research/biosociety/pdf/norwegian_act_biobanks.pdf. Scotland: Generation Scotland – Legal and Ethical Aspects, http:// www.129.215.140.49/gs/Documents/GSlawandethicsSep03.pdf. Singapore: Bioethics Advisory Committee, Singapore Bioethics Advisory Committee (BAC), 2002, Human Tissue Research, Research Involving Human Subjects: Guidelines for IRBs, 2004, http://www.bioethics-singapore.org8. Sweden: Sweden/Government, Sweden, 2003, http://www.sweden.gov. se/content/1/c6/02/31/26/f69e36fd.pdf. United Kingdom: UK Biobank Ethics and Governance Framework, 2006, http://www.ukbiobank.ac.uk/docs/EGF_Version2_July%2 006%20most%20uptodate.pdf.
CHAPTER
25 Application of Biomarkers in Human Population Studies Stefano Bonassi and Monica Neri
INTRODUCTION The introduction of biomarkers in human population studies is an effective strategy to gain knowledge about the occurrence of disease, to improve the understanding of etiology and pathogenesis, and to measure the effort to control the disease outcome. More specifically, under a gene–environment model of disease causation, biomarkers are used to measure exposures to causative agents that may induce or prevent the disease, to find out those genetic or acquired factors that may change the individual probability of disease, to predict the outcome of a disease through early damages, and to reduce the time interval between exposure to risk factors and identification of potentially harmful effects (Toniolo et al., 1997; IARC Workshop report, 1997; Bonassi and Au, 2002; Ugolini et al., 2007). The multidisciplinary nature of biomarkers is reflected in the difficulty to find a single definition that applies to the many fields in which biomarkers can be used. From a very general viewpoint, a biomarker is a measure of a chemical, cellular, molecular, immunologic, genetic, or physiologic signal, biologic event, or biologic state, measured in biologic material, although there are other definitions more oriented to disease diagnosis and treatment; that is, a biomarker is a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacologic response to a therapeutic intervention. In addition to assessing pharmacologic response, biomarkers can be applied as a
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
diagnostic tool, for staging disease, to predict and monitor clinical response to any intervention. The potential applications of biomarkers in biomedical research are numerous. They represent the most suitable alternative to traditional endpoints used in clinical research and in public health. This growing interest has been refueled by the increased accessibility to high-throughput techniques that can impact in many instances disease prevention – detecting diseases at early stages or identifying subgroups of susceptible individuals – and disease diagnosis and treatment-anticipating clinical diagnosis or surrogating hard endpoints commonly used in clinical trials. The use of biomarkers in human studies has been essentially directed in two directions, that is, to implement the traditional epidemiologic approach with new endpoints and to improve and accelerate clinical trials and drug development.
BIOMARKERS IN MEDICINE In a clinical and pharmacologic setting, biomarkers can provide a basis for selecting lead compounds for subsequent clinical evaluation. Furthermore, they can provide information to guide dosing and to minimize interindividual variation in response and in toxicity. Finally, some biomarkers can be employed as substitutes for clinical endpoints in clinical trials, under the assumption that they can predict clinical benefits (Downing, 2000).
Copyright © 2009, Elsevier Inc. All rights reserved. 299
300
CHAPTER 25
■
Application of Biomarkers in Human Population Studies
Before discussing some examples of the use of biomarkers in drug development and clinical trials, two major concepts must be properly described: surrogate endpoint and biomarker validation. A surrogate endpoint is a laboratory measurement or a physical sign used in a clinical trial as a substitute for an endpoint directly measuring how a patient feels, functions, or survives. For example, changes induced by a therapy on a surrogate endpoint are expected to reflect changes in a clinically meaningful endpoint (Piantadosi, 1997). The use of a “surrogate” variable in lieu of the true endpoint (usually survival or irreversible morbidity) allows one to evaluate the outcome more rapidly, less expensively, and/or less invasively. The use of a biomarker as surrogate endpoint requires validation, that is a demonstration of accuracy (correlation of the measure with the clinical endpoint) and precision (the reproducibility of the measure) (Downing, 2000). These properties are the basis for any biomarker validation; however, surrogate biomarker validation requires a more rigorous standard of evidence that links drug exposure to desired clinical drug effect (Prentice, 1989; Fleming and DeMets, 1996). The surrogate should be a correlate of the true clinical outcome and “fully capture the net effect of the treatment on the clinical outcome.” Validation of a surrogate endpoint requires not only that the surrogate correlates with the clinical endpoint, but also that changes in the clinical endpoint be reflected by changes in the surrogate (Racusen, 1999). Interestingly, there are a number of surrogate markers that are currently accepted by regulatory authorities such as blood pressure, serum cholesterol, and tumor response, that do not fully meet the standard for surrogate biomarkers, and that may be sensitive only to selected candidate drug mechanism of action in phase 3 evaluation. There are many potential valuable applications of biomarkers in medicine. These applications include: (1) diagnosis and differential diagnosis, (2) screening of potential new therapies either in vitro or in vivo, (3) measuring severity, progression of disease, and responses to therapy, (4) predicting prognosis, and (5) measuring toxicity. Some of these possible applications will be described with selected examples from nephrology, neurology, and cardiology. Renal biopsy is considered the gold standard for evaluation of renal allograph dysfunction in renal transplant recipients. Biopsies are examined for evidence of T cell and antibodymediated rejection according to standard criteria such as the Banff or the CCTT classification score (Racusen et al., 1999). These scores are in essence diagnostic markers, useful to discriminate immune-based rejection from other causes, to predict the probability of response to immuno-suppressive therapies, and to predict graft survival (Lachenbruch et al., 2004). Most markers used in renal transplantation can be considered predictive of renal allograph rejection, since they explain part of the variability of the endpoint. On the other hand, a surrogate biomarker is a substitute of the clinical endpoint, and changes in the outcome must be reflected in the surrogate. A few examples of candidate surrogates from different fields are reported in Table 25.1. A number of potential surrogates for renal allograph rejection have been proposed, especially for long-term graft survival,
TABLE 25.1
Examples of surrogate endpoints
Disease
Endpoint
Surrogate endpoint
Cancer
Mortality
Tumor size reduction CEA Level (Colon cancer) PSA Level (Prostate)
Disease progression Disease progression HIV infection
AIDS Disease progression
CD4count Plasma viremia
Cardiovascular disease
Hemorrhagic stroke
Blood pressure Myocardial infarction Cholesterol level
Glaucoma
Vision loss
Intraocular pressure
Adapted from Lachenbruch et al., 2004.
including interstitial fibrosis at 6–12 months, chronic allograft arteriopathy at 3–12 months, CADI at 12 months, etc. (Lachenbruch et al., 2004). However, none of them has been fully validated given the difficulty of planning and running longterm follow-up studies. Among these biomarker a great interest is focused on molecular markers, especially on the quantitative mRNA levels of immune effector molecules, like perforin, granzyme B, and cytokines (IL-10, IL-15, IL-7, IFN-). More recently microarray gene expression analysis has been proposed to identify those genes that most likely discriminate rejection sub-types. Among most exciting findings was the heterogeneous behavior of biopsies from acute rejection, with expression profiles resembling those of drug toxicity or chronic allograft nephropathy (Sarwal et al., 2003). Other expression arrays identified a set of genes predicting the development of chronic allograft pathology 6 months later (Scherer et al., 2003). Many examples of the use of biomarkers in clinical studies comes from neurology, especially in major diseases such as multiple sclerosis (MS), stroke, and neurodegenerative diseases including Alzheimer’s disease (AD), Parkinson’s disease (PD), and Huntington’s disease (HD). The use of these biomarkers relies on the understanding of the pathological basis for the disease under study at the molecular genetic, biochemical, or anatomical level. For example, the cytosine–adenosine–guanosine (CAG) repeat length in HD may contribute as a diagnostic test in symptomatic individuals (MacMillan et al., 1995), or as a prognostic test for presymptomatic individuals at risk for HD. Nevertheless, this marker can hardly be used as a surrogate clinical endpoint in a clinical trial, because it would be unlikely to change with a therapeutic intervention and can be abnormal in individuals
Biomarkers in Medicine
with no clinical signs of disease (presymptomatic HD) (Feigin, 2004). Biochemical markers used in clinical practice and clinical trials include measuring CSF oligoclonal bands to confirm the diagnosis of MS, and functional imaging of dopaminergic neurons in PD with dopamine transporter ligands, measures of dopamine metabolism (fluorodopa). Anatomical biomarkers include magnetic resonance imaging (MRI) of lesion burden or cerebral atrophy in MS, and confirmation of the clinical diagnosis of stroke with computed tomography (CT) or MRI. A major controversy in the field concerned the choice of best endpoints in clinical trials of neurodegenerative disorders aimed at slowing disease progression or delaying disease onset. The choice of clinical endpoints is quite controversial, and conditions like “no-change in clinical status,” or a rate of worsening slower than expected may be considered as a successful result. However, while clinical scales are commonly used, they may be really misleading in trials testing symptomatic therapies, as demonstrated by the misinterpretation of results from the DATATOP study (DATATOP, 1989; Feigin, 2004). The diagnosis and treatment of atherosclerosis is a classic example for the use of biomarkers, not only to identify at risk subjects, but also for the understanding of the molecular basis of disease pathogenesis. Although most clinical events associated with thrombosis are caused by plaque rupture or erosion, most plaques that rupture do not cause symptoms. Therefore, clinical trials measuring only clinical events do not detect the majority of asymptomatic plaque ruptures and have poor sensitivity to detect therapeutic effects on plaque stabilization and thrombosis. A major contribution to the understanding of etiology and pathogenesis of this disease comes from the Atherosclerosis Risk in Communities (ARIC) study, which included approximately 16,000 men and women between the ages of 45 and 64 years, and evaluated the incidence of coronary heart disease (CHD) events (fatal or nonfatal myocardial infarction and surgical revascularization procedures), and carotid artery atherosclerosis (assessed by B-mode ultrasonography). This study allowed the definition of a set of biomarkers that may impact clinical practice and pharmaceutical research. In particular the investigators found that a high level of intercellular adhesion molecule1 (ICAM-1) increased the risk for a coronary event 5.5-fold, after adjusting for an impressive list of potential and actual confounders. They also found that individuals with either elevated ICAM-1 or E-selectin had a higher risk for carotid artery atherosclerosis (Hwang et al., 1997). These results were confirmed in the Physicians’ Health Study, which suggested that a low-grade chronic inflammation may contribute to the development of atherothrombotic disease (Ridker et al., 1998). The progressive evolution of high-throughput techniques and their rapid application to clinical practice has changed the perspectives for the use of biomarkers in clinical trials as well as in the drug development process. Efforts have been dedicated to the identification of gene expression profiles, for the purpose of defining susceptible subjects, and allowing early detection of disease. These tools have allowed us to categorize otherwise
■
301
indistinguishable disease stages and histotypes, and the correlation of molecular features with the likelihood of treatment response or with prognosis. However, this technique has not been properly put to the test in prospective clinical trials. An example of these difficulties comes from oncology. The MINDACT (Microarray In Node-negative Disease may Avoid ChemoTherapy) trial was designed to provide definitive evidence on the clinical relevance of a 70-gene prognosis signature, comparing its performance with that of traditional prognostic indicators for assigning adjuvant chemotherapy to patients with node-negative breast cancer. The MINDACT investigators recently discussed the intrinsic difficulties in designing trials using gene expression profiling, reporting the major challenges inherent logistics, implementation, and interpretation of the results (Bogaerts et al., 2006). The effector role of proteins has been exploited by the recent availability of arrays for studying the proteome. The use of proteomics is in its infancy in clinical trials; however, the evolution of proteomic techniques may be of importance for the drug development process. While the complexity of structure and functions in the proteome represents a significant challenge, the pharmaceutical industry and its biotech and academic partners are expending a tremendous amount of resources to decipher and utilize proteomic data to make drug development more efficient and successful. In addition to using proteomic tools to identify an array of proteins that are modified in the disease state, a significant focus of proteomic activities in drug discovery has turned to the identification of biomarkers in easily accessible biological fluids. The development and validation of a biomarker aids not only in the understanding of the disease process and its progression, but also are critical as a monitoring tool in later stages of drug development. Good examples of such markers are the serum and urine biomarkers used to identify arthritis. Several biomarkers from synovial fluid, blood and urine have been used to identify and study the stages of osteoarthritis (OA) (e.g., urinary collagen fragments; Downs et al., 2001). A group of three markers best distinguished OA patients: one inflammatory marker and markers of both cartilage anabolism and catabolism. So, in this example, biomarkers that are directly indicative of the disease process (skeletal metabolism) are readily measured in body fluids and can be used throughout the drug development process (Walgren and Thompson, 2004). The authors emphasized that often more than one biomarker is necessary or preferable. A single marker may not readily track disease progression in each patient, especially if the marker is not directly linked to the molecular basis of the disease. The role of biomarkers in epidemiologic, non-clinical studies is similar to the examples discussed so far. Nevertheless, major differences exist, such as the reduced emphasis on the individual level and the focus on early stages of disease, which have a lower association with the outcome, but have a stronger potential for prevention. The effort to disentangle the complex interaction existing between the various steps of pathogenesis and the corresponding biomarker started during early 1980’s, when the National
302
CHAPTER 25
■
Application of Biomarkers in Human Population Studies
Research Council and other researchers proposed the classical picture of the black box and described the evolution of the detailed continuum of events eventually leading to a clinical diagnosis of disease (Perera and Weinstein, 1982; National Research Council, 1987). This figure has been represented in many versions and is now an icon of the molecular epidemiology approach. A comprehensive version of the figure describing the hypothesized continuum of events in the pathway leading to cancer is reported here (Figure 25.1). Exposure to environmental risk factors
Phase I, II reactions
Target dose
DNA damage promotion
DNA repair
Micronucleus
Normal cell
Metabolically activated cell
DNA adducts bearing cell
Markers evaluating biologically effective dose DNA adducts Protein adducts Tests evaluating internal dose Metabolites presence in body fluids or tissues Susceptibility markers Phase I, II activities
DNA damaged cell Early biological effect markers Cytogenetic markers UDS, DPXL, pADPRP Alkaline elution Comet assay
Mutated cell Early biological response markers Oncogenes TSGs Oncoproteins HPRT, HLA-A, Hb, GPA mutations
Figure 25.1 Intermediate endpoints used as biomarkers in molecular epidemiology as related to carcinogenesis process (Izzotti et al., 1997).
According to the classification proposed in those early papers and basically still adopted, biomarkers are divided depending upon their position in the disease natural history. The predominant model for the pathogenesis of complex diseases such as cancer is based on the interaction between environmental exposure to carcinogens and the individual genetic profile. This interaction, occurring at molecular level, determines early biological effects, which may progress to a clinical diagnosis of disease (Figure 25.2). Biomarkers are then classified as biomarkers of exposure, early effect (health risk), and susceptibility.
ENVIRONMENT
GENES
Biomarkers of exposure
Biomarkers of susceptibility
Biomarkers of early effect DISEASE
Figure 25.2 Human diseases are caused by some combination of environmental and genetic factors.
BIOMARKERS OF EXPOSURE An important piece of the information to gather for a molecular epidemiology study is the documentation of exposure to causal factors. Traditional epidemiology studies collect this information via questionnaire and historical information such as employment records and monitoring data. The use of biomarkers provides a new way to document the presence and the intensity of genotoxic exposures, no longer based on the amount of the agent in the environment, but on the dose reaching the target (DNA for genotoxic compounds): an individually tailored exposure measurement. Among the biomarkers currently used for this purpose, adducts are particularly useful for identifying chemicals that form covalent bonds with biological macromolecules (e.g., protein and DNA adducts). These biomarkers have become very popular because the presence of specific DNA adducts are associated with the development of disease (Vineis and Perera, 2000; Perera, 2000). Carcinogenic and mutagenic agents contained in most common carcinogens such as tobacco smoke and air pollution induce the formation of aromatic DNA adducts, which have been indicated to play an important role in the etiology of lung cancer (Peluso et al., 2005). However, tobacco smoke is also a major source of oxidative stress and may induce the production of endogenous DNA lesions, through oxidative DNA damage and lipid peroxidation of cell membrane (Panda et al., 2000). There are many features that should be considered when choosing a biomarker to measure exposure to carcinogenic agents in human populations. One of the most important is flexibility. The following example shows the possible application of the same biomarker in the same population to measure two major sources of DNA damage. A small population of subjects undergoing bronchoscopy for diagnostic purposes was included in a transitional study, aimed at evaluating the extent of DNA adduction due to exposure to tobacco smoke in target and surrogate tissues (Peluso et al., 2004). Samples of peripheral blood lymphocytes (PBL), nasal mucosa, and bronchial mucosa were collected during bronchoscopy in 55 subjects, both smokers and non-smokers, who gave their consent to participate to the study. The level of DNA adducts measured by 32P labeling assay in nasal mucosa and in PBL were correlated with those in bronchial mucosa (p 0.01 and p 0.05 respectively). DNA adducts in smokers were significantly increased in both nasal mucosa and PBL, with a significant dose-response linear trend (p 0.05). Many compounds among those present in tobacco smoke form DNA adducts, and their measurement is an excellent index of exposure to DNA-damaging carcinogens in humans. The formation of carcinogen-DNA adducts is an initial event in carcinogenesis (Peluso et al., 2005; Tang et al., 2001) and could be efficiently used to assess the oncogenic risk in exposed individuals. Besides the direct adduction of macromolecules to DNA, oxidative damage has been recognized to play a major role in early stages of carcinogenesis. Among oxidative pathways, most studies are those involving malondialdehyde (MDA). MDA is
Biomarkers of Early Disease Risk
(a)
(b)
U1
U2
U2 U1
U3 OR
U3 OR
Figure 25.3 Profiles of MDA–DNA adducts (U1–U3 adduct spots) in (a) bronchi and in (b) a standard MDA sample (from Peluso et al., 2006).
a natural product of membrane lipid peroxidation, formed also during prostaglandin biosynthesis via cyclooxygenase, capable to interact with DNA, forming exocyclic DNA adducts (Marnett, 2000). MDA-DNA adducts may be also induced by direct DNA oxidation, through the production of base propenal intermediates. MDA-DNA adducts have been shown to be premutagenic in mammalian cells and to induce frameshift and base pair substitution mutations (Benamira et al., 1995). Using the same material collected during the first study, the levels of MDA-DNA adducts were measured in the bronchial specimen of subjects undergoing a bronchoscopic examination for the clinical suspicion of lung cancer (Munnia et al., 2006). MDA-DNA adducts were higher in smokers than in never-smokers (Frequency Ratio (FR) 1.51, 95% Confidence Interval (CI) 1.01–2.26). MDA-DNA adducts were also increased in lung cancer cases with respect to matched controls, but only in the smoker subgroup (FR) 1.70; 95% CI (95% CI 1.16–2.51) (see Figure 25.3). In both studies the effect of exposure to tobacco smoke was evaluated in the context of the genetic profile of the study subjects, using a set of candidate genes directly involved in the metabolism of carcinogenic compounds or in the resistance to the oxidative damage. In the first study the frequency of bulky DNA adducts was not modified by the functionality of CYP1A1, GSTM1 and GSTT1 genes, while in the study on MDA a cyclin D1 (CCDN1) G870A polymorphism, previously associated with an impaired cell-cycle regulation and an accumulation of DNA damage, determined an increased frequency of MDA adducts in the heterozygotes FR 1.51 (95% CI 1.04–2.20) and in homozygotes (FR 1.45; 95% CI 1.02–2.07). Recent changes in the availability of high-throughput techniques have allowed the experimental use of expression arrays to highlight those genes that may constitute a fingerprint of a specific exposure. Workers exposed to benzene (Smith et al., 2005), and children exposed to air pollution (van Leeuwen et al., 2006) were among the first human populations to be screened with this assay, whose results are still to be validated for this purpose.
BIOMARKERS OF EARLY DISEASE RISK Biomarkers from this class represent a large number of biological events, including altered mRNA expression, chromosome damage,
■
303
cytotoxicity, somatic cell mutations, and immunologic alterations. These biomarkers, besides describing the early interaction of genotoxic agents with the human body, can be used as outcome measures in epidemiologic research, and can potentially allow risk factors for disease to be identified early in the exposuredisease process and allow for more timely interventions. This is particularly useful for diseases with long latency such as cancer. Biomarkers of early effect can be used at the population level, where research results may be incorporated into policy-making decisions and, potentially, at the individual level, where biomarkers may be employed for screening purposes and ultimately in clinical practice (Schulte and Rothman, 1998). Among this class, biomarkers measuring chromosome stability are among the most extensively used and best validated in population studies (Bonassi et al., 2005). Extensive literature exists on the use of chromosome aberrations (CA) to measure exposure to ionizing radiation and to genotoxic chemicals. More recently the causal role played by CA in the pathogenesis of cancer have been confirmed by epidemiological evidence, and this finding has opened a number of perspectives for the planning of cancer prevention policies (Bonassi et al. 2004; Hagmar et al., 2004). A drawback of traditional cytogenetic assays is the laborintensive laboratory procedure. Therefore, several molecular techniques have been developed to enhance the assay. The most significant improvement is the use of the fluorescencein-situ-hybridization procedure (FISH) to detect chromosomespecific aberrations. Chromosome-specific probes that are labeled with fluorescent dyes are used to hybridize to chromosomes for the determination of chromosome rearrangements and chromosome region-specific breaks. This approach, besides the significant reduction of the labor time, increased the specificity of the association exposure-damage, allowing the evaluation of specific chromosome damage and rearrangement (Swiger and Tucker, 1996). A classic example of this enhanced specificity are the studies on the effect of benzene exposure on selected chromosomes (Zhang et al, 2002) or studies of specific oncogene activation (Scaruffi et al., 2004). The most promising assay in the class of biomarkers of early biological effect seems to be the micronucleus (MN) assay. This assay has proved to be much simpler, quicker and cheaper than CA, and recently the capability of MN frequency in lymphocytes of healthy subjects to predict their risk of cancer has been published with parallel results as for CA (Bonassi et al., 2007). Moreover, its recently described multi-endpoint nature is extremely promising for a more extended use of the assay in exposed human populations. The cytokinesis-block micronucleus (CBMN) assay was originally developed as an ideal system for measuring MN; however, it can also be used to measure nucleoplasmic bridges (NPBs), nuclear buds (NBUDs), cell death (necrosis or apoptosis) and nuclear division rate. Current evidence suggests that (a) NPBs originate from dicentric chromosomes in which the centromeres have been pulled to the opposite poles of the cell at anaphase and are therefore indicative of DNA mis-repair, chromosome rearrangement
304
CHAPTER 25
■
Application of Biomarkers in Human Population Studies
CYTOTOXIC GENOTOXIC AGENT
G0
G1, S, G2
APOPTOSIS
NECROSIS CHROM. BREAKAGE DICENTRIC GENE and CHROM. AMPLIFICATION LOSS
Figure 25.4 The cytokinesis-block micronucleus (CBMN) assay has evolved into a “cytome” method for measuring comprehensively chromosomal instability phenotype and altered cellular viability (from Fenech et al., 2003).
or telomere end-fusions, (b) NPBs may break to form MN, (c) the nuclear budding process is the mechanism by which cells remove amplified and/or excess DNA and is therefore a marker of gene amplification and/or altered gene dosage, (d) cell-cycle checkpoint defects result in MN formation and (e) hypomethylation of DNA, nutritionally induced or by inhibition of DNA methyl transferase, can lead to MN formation either via chromosome loss or chromosome breakage. In this comprehensive mode, the CBMN assay measures all cells including necrotic and apoptotic cells as well as number of nuclei per cell to provide a measure of cytotoxicity and mitotic activity. The CBMN assay has in fact evolved into a “cytome” method for measuring comprehensively chromosomal instability phenotype and altered cellular viability caused by genetic defects and/or nutritional deficiencies and/or exogenous genotoxins (Figure 25.4).This great availability of endpoints, that allows measuring a number of various damages in the same (simple) assay, opens up an exciting future for the use of this methodology in emerging fields such as nutrigenomics, toxicogenomics, and their combinations (Fenech, 2006)
BIOMARKERS OF GENETIC SUSCEPTIBILITY TO DISEASE The role of genes in the etiology of cancer have been studied since long ago by family-based studies, and this approach led to the mapping of several cancer-related genes. However, these genes, which are characterized by high penetrance, high absolute, and relative risk, have a low impact on the total burden of cancer (Toniolo et al. 1997; IARC, 1997). Molecular epidemiology
in genetic research has greatly impacted the development of etiologic models based on the gene–environment interaction and their application to population studies. In the field of cancer research the most studied polymorphisms are those involving genes active in various steps of chemical metabolism of xenobiotics and in DNA repair. Biomarkers of susceptibility were first studied at phenotypic level, by quantifying enzyme proteins and functions or by administering test substances and looking for their metabolites in urine. These kinds of methods are still important in pharmacokinetic and pharmacodynamic studies; however, the development of molecular biology allowed the study of polymorphisms at genetic level by PCR, PCR/RFLP or PCR/SSCP in a molecular epidemiologic framework. The availability of these techniques lead to the identification of disease risks associated to genetic variation as small as single nucleotide polymorphisms (SNPs). In some cases the SNPs may have a direct causal significance, while sometimes they simply tag a region of the genome that is associated with the risk of disease. Despite the definition of polymorphisms as common genetic variation, from the epidemiological viewpoint these events are rather rare, and as a consequence most epidemiologic studies involving these biological markers are generally inconclusive. This common limitation has called attention to the need to summarize the results of individual studies, while waiting for the completion of larger studies, designed to answer questions that have been raised by preliminary studies. Recent papers have examined methodological issues related to meta-analyses and pooled analysis of biomarker studies (Taioli and Bonassi, 2002). Pooled analysis seems to provide a relevant improvement over meta-analysis in molecular epidemiology studies, although more research on methodology is needed. The need of larger databases of data on biomarkers has contributed to the start of various collaborative studies, aimed at providing results not available from the analysis of single studies. Among these large projects, GSEC and HuGE have contributed significant results to the literature on genetic susceptibility (Hung et al., 2005; Raimondi et al., 2005). One of the major advantages of these collaborative studies is the potential to investigate small population subgroups for which single studies could not reach enough statistical power. For example a recent paper published by the GSEC consortium investigated the association between metabolic gene polymorphisms and lung cancer risk in non-smokers (Raimondi et al., 2005). Since genetic factors may play an important role in lung cancer development at low dose of carcinogen exposure, non-smokers are a good model to study genetic susceptibility and its interaction with environmental factors. The authors evaluated the role of the metabolic gene polymorphisms CYP1A1MspI, CYP1A1lle462Val, GSTM1, and GSTT1 in non-smoker lung cancer patients. Non-smokers (defined as subjects who never smoked on a regular basis) were selected from the raw data of 21 case-control studies for a total of 668 cases and 2479 controls. A significant association between lung cancer and CYP1A1lle462Val polymorphism was observed in Caucasians (adjusted OR 2.04, 95% CI 1.17–3.54). GSTT1 deletion variants appear to be a risk
Biomarkers of Genetic Susceptibility to Disease
factor for lung cancer in Caucasian non-smokers only when the analysis was restricted to studies including healthy controls (adjusted OR 1.66, 95% CI 1.12–2.46). A protective effect on lung cancer was observed with the combination of CYP1A1 wild type, GSTM1 null, and GSTT1 non-null genotypes. Interestingly, pooled analyses of data allows one to evaluate the impact of genetic polymorphisms not only in major pathways, such as metabolism or DNA repair, but also in pathways that are much less common. A good example is a recent paper from the HuGE consortium that evaluated the role of genetic polymorphism in a DNA repair pathway in modulating cancer risk (Hung et al., 2005). In this case a meta-analysis of associations was conducted between genes in the base excision repair pathway and cancer risk, focusing on three key genes: 8-oxoguanine DNA glycosylase (OGG1), apurinic/apyrimidinic endonuclease (APE1/APEX1), and X-ray repair cross-complementing group 1 (XRCC1). Using a large number of cases and controls, they found increased lung cancer risk among subjects carrying the OGG1 Cys/Cys genotype (OR 1.24, 95% CI: 1.01, 1.53), and a protective effect of the XRCC1 194Trp allele for tobacco-related cancers (OR 0.86, 95% CI: 0.77, 0.95). The XRCC1 399Gln/399Gln genotype was associated with increased risk of tobacco-related cancers among light smokers (OR 1.38, 95% CI: 0.99, 1.94), but decreased risk among heavy smokers (OR 0.71, 95% CI: 0.51, 0.99), suggesting effect modification by tobacco smoking. Even with large datasets gathered by international pooling efforts, many association studies yields-negative results. Possible explanations for the lack of association are both methodological, that is, low-study power to demonstrate small effects and too few cases to investigate disease heterogeneity (e.g., by tumor histology), and substantive; that is the existence of multiple pathways that can compensate for each other. Moreover, the possibility of gene–gene or SNP–SNP interactions, or the possibility of linkage disequilibrium between polymorphisms are rarely considered. Further investigations of the haplotype effect of a gene and the study of multiple polymorphisms in different genes within the same pathway and different pathways are needed (Manuguerra et al., 2007). Recommendations for future studies include pooling of individual data to facilitate evaluation of multigenic effects and detailed analysis of effect modification by environmental exposure. One of the major problems in studying the influence of genetic polymorphisms is difficulty collecting large enough study populations. To make association studies more suitable, the use of cancer risk biomarkers (cytogenetic, DNA adducts, and others) instead of cancer incidence has been used (reviewed in Au et al., 2004). Recently a pooled re-analysis of data from seven laboratories that performed biomonitoring studies using the in vivo cytokinesis-block MN assay was performed to evaluate the influence of genetic polymorphisms in GSTM1 and GSTT1 genes on MN frequencies in human PBL (Kirsch-Volders et al., 2006). A total of 301 unexposed individuals and 343 workers exposed to known or suspected genotoxic substances were analysed by Poisson regression. The results of the pooled analysis
■
305
indicate that the GSTM1 and GSTT1 null genotypes are associated with reduced MN frequency in the studied populations. In both exposed and total populations, the GSTT1 null genotype seems to have a protective effect (p .016). The function of many genotypes is modified in some subgroups; for instance, the protective effect of the GSTT1 null genotype is reversed with increased age (above 40 years). This study confirmed the efficiency of pooled studies, which by increasing the statistical power, can assess the involvement of genetic variants on genome stability and resolve discrepancies among individual studies. Recently new techniques such as DNA microarray analysis allowed the detection of multiple SNPs in thousands of genes at once, testing many subjects at reasonable cost. The vast amount of data that is made available by these assays has required the development of new techniques for statistical analysis. The overwhelming amount of information and the cost of microchips has directed researchers involved in population studies to develop alternative array-based techniques, more specific and more adequate for large population surveys. A suitable technique for these purpose is the arrayed-primer extension (APEX), which allows the simultaneous analysis of a large number of specific cancer-related SNPs potentially relevant to the tumor etiology and pathogenesis. An example of this approach is a case-control study on 90 cases of malignant pleural mesothelioma (MPM) and 349 controls, which were extensively genotyped to evaluate the extent of the gene–environment interaction between these genes and asbestos exposure (Landi et al., 2007). APEX consists of a sequencing reaction primed by an oligonucleotide anchored with its 5 end to a glass slide and terminating just one nucleotide before the polymorphic site. A DNA polymerase extends the oligonucleotide by adding one fluorescently labeled dideoxy-nucleotide (ddNTP) complementary to the variant base. Reading the incorporated fluorescence identifies the base in the target sequence. Since both sense and antisense strands are sequenced, two probes were designed for each polymorphism. Five-prime (C-12) aminolinker oligonucleotides are synthesized and spotted onto silanized slides. Genomic DNA is amplified to enrich the fragments carrying the SNPs by using specific primer pairs. Then, PCR products are pooled, purified, concentrated, and fragmented. See (Landi et al., 2003; Pastinen et al., 2000) for details. The microarrays used in this case-control study were designed to genotype 300 SNPs occurring on selected highly polymorphic genes. To allow a proper evaluation of the overall dataset the statistical analysis was conducted in by addressing separate pathways. The first was the oxidative pathway, that has been extensively reported as to play a major role in the asbestos-mediated carcinogenesis of MPM. An increased risk of MPM was found in subjects bearing a GSTM1 null allele (OR 1.69, 95% CI 1.04–2.74; p 0.034), and in those with the Ala/Val and Ala/Ala genotypes at codon 16 within MnSOD (OR 3.07 in the recessive model, 95% CI 1.55 – 6.05; p 0.001). A stronger effect of MnSOD was observed among patients without a clear exposure to asbestos fibers. No effect was found for other GSTs genes. MPM cases with a MnSOD Val/Val genotype
306
CHAPTER 25
■
Application of Biomarkers in Human Population Studies
showed a significantly better survival (Hazard Ratio 0.47; 95% CI 0.25–0.89). These findings contribute substantial evidence to the hypothesis that oxidative stress and cellular anti-reactive oxygen species systems are involved in the pathogenesis and in the natural history of MPM (Landi et al., 2007).
CONCLUSIONS The introduction of biomarkers in the design of clinical and epidemiological studies is a recent achievement, but in the past few decades a number of discoveries have been made using these new tools (Perera, 2000). Among the challenges to be addressed in the future, two seem more urgent: the validation of biomarkers of disease risk, and the introduction of validated
biomarkers into disease prevention strategies, and eventually in clinical practice. The best example for this latter application is the use of total cholesterol to evaluate the risk of CHD. This model was validated many years ago in one of the first example of population cohort studies, and is still the most popular risk biomarker for the prevention of the CHD (Durrington et al., 1999). The second challenge is the adaptation and the validation of high-throughput techniques for the use in population studies. This effort, although close to the success, suffers from a number of limitations in the reliability of the technique and in the analysis of data that require a bigger final effort of validation. In summary, addressing these priorities will generate a group of biomarkers with a great potential to change our approach to the determination of the disease risk, which will offer a new perspective for strategies of disease prevention and treatment.
REFERENCES Au, W.W., Navasumrit, P. and Ruchirawat, M. (2004). Use of biomarkers to characterize functions of polymorphic DNA repair genotypes. Int J Hyg Environ Health 207, 301–313. Benamira, M., Johnson, K., Chaudhary, A., Bruner, K., Tibbetts, C. and Marnett, L.J. (1995). Induction of mutations by replication of malondialdehyde-modified M13 DNA in Escherichia coli: Determination of the extent of DNA modification, genetic requirements for mutagenesis, and types of mutations induced. Carcinogenesis 16, 93–99. Bogaerts, J., Cardoso, F., Buyse, M., Braga, S., Loi, S., Harrison, J.A., Bines, J., Mook, S., Decker, N., Ravdin, P. et al. (2006). Gene signature evaluation as a prognostic tool: Challenges in the design of the MINDACT trial. Nat Clin Pract Oncol 3, 540–551. Bonassi, S. and Au, W.W. (2002). Biomarkers in molecular epidemiology studies for health risk prediction. Mutat Res 511, 73–86. Bonassi, S., Znaor, A., Norppa, H. and Hagmar, L. (2004). Chromosomal aberrations and risk of cancer in humans: An epidemiological perspective. Cytogenet Genome Res 104, 376–382. Bonassi, S., Ugolini, D., Kirsch-Volders, M., Strömberg, U.,Vermeulen, R. and Tucker, J.D. (2005). Human population studies with cytogenetic biomarkers: Review of the literature and future prospectives. Environ Mol Mutagen 54, 258–270. Bonassi, S., Znaor, A., Ceppi, M., Lando, L., Chang, W.P., Holland, N., Kirsch-Volders, M., Zeiger, E., Ban, S. et al. (2007). An increased micronucleus frequency in peripheral blood lymphocytes predicts the risk of cancer in humans. Carcinogenesis 28, 625–631. DATATOP (1989). DATATOP: a multicenter controlled clinical trial in early Parkinson’s disease. Parkinson Study Group. Arch Neurol 46, 1052–1060. Downing, G.J. (2000). Biomarkers and surrogate endpoints: Clinical research and application. Elsevier Science BV. Amsterdam,The Netherlands. Downs, J.T., Lane, C.L., Nestor, N.B., McLellan, T.J., Kelly, M.A., Karam, G.A., Mezes, P.S., Pelletier, J.P. and Otterness, I.G. (2001). Analysis of collagenase-cleavage of type II collagen using a neoepitope ELISA. J Immun Meth 247, 25–34. Durrington, P.N., Prais, H., Bhatnagar, D., France, M., Crowley,V., Khan, J. and Morgan, J. (1999). Indications for cholesterol-lowering
medication: Comparison of risk-assessment methods. Lancet 353, 278–281. Feigin, A. (2004). Evidence from Biomarkers and Surrogate Endpoints. NeuroRx® 1, 323–330. Fenech, M., Chang, W.P., Kirsch-Volders, M., Holland, N., Bonassi, S. and Zeiger, E. (2003). HUMN project: Detailed description of the scoring criteria for the cytokinesis-block micronucleus assay using isolated human lymphocyte cultures. Mutat Res 534, 65–75. Fenech, M. (2006). Cytokinesis-block micronucleus assay evolves into a “cytome” assay of chromosomal instability, mitotic dysfunction and cell death. Mutat Res 600, 58–66. Fleming, T.R. and DeMets, D.L. (1996). Surrogate end points in clinical trials: Are we being misled. Ann Int Med 125, 605–613. Hagmar, L., Strömberg, U., Bonassi, S., Hansteen, I.L., Knudsen, L.E., Lindholm, C. and Norppa, H. (2004). Impact of types of lymphocyte chromosomal aberrations on human cancer risk: Results from Nordic and Italian cohorts. Cancer Res 64, 2258–2263. Hung, R.J., Hall, J., Brennan, P. and Boffetta, P. (2005). Genetic polymorphisms in the base excision repair pathway and cancer risk: A HuGE review. Am J Epidemiol 162, 925–942. Hwang, S.J., Ballantyne, C.M., Sharrett, A.R. et al. (1997). Circulating adhesion molecules VCAM-1, ICAM-1, and E-selectin in carotid atherosclerosis and incident coronary heart disease cases:The atherosclerosis risk in communities (ARIC) study. Circulation 96, 4219–4225. Izzotti, A., Neri, M.,Vecchio, D. and Puntoni, R. (1997). Molecular epidemiology in cancer research (Review). Int J Oncol 11, 1053–1069. Kirsch-Volders, M., Mateuca, R.A., Roelants, M., Tremp, A., Zeiger, E., Bonassi, S., Holland, N., Chang, W.P., Aka, P.V., Deboeck, M. et al. (2006). The effects of GSTM1 and GSTT1 polymorphisms on micronucleus frequencies in human lymphocytes in vivo. Cancer Epidemiol Biomarkers Prev 15, 1038–1042. Lachenbruch, P.A., Rosemberg, A.S., Bonvini, E., Cavaille’-Coll, M.W. and Colvin, R.B. (2004). Biomarkers and surrogate endpoints in renal ransplantation: Present status and considerations for clinical trial design. Am J Transplant 4, 451–457. Landi, S., Gemignani, F., Gioia-Patricola, L., Chabrier, A. and Canzian, F. (2003). Evaluation of a microarray for genotyping polymorphisms
References
related to xenobiotic metabolism and DNA repair. Biotechniques 35, 816–827. Landi, S., Gemignani, F., Neri, M., Barale, R., Bonassi, B., Bottari, F., Vanessa, P.A., Canzian, F., Ceppi, M., Filiberti, R. et al. (2007). Polymorphisms of glutathione-S-transferase M1 and manganese superoxide dismutase affect the risk of malignant pleural mesothelioma. Int J Cancer 120, 2739–2743. MacMillan, J.C., Davies, P. and Harper, P.S. (1995). Molecular diagnostic analysis for Huntington’s disease: A prospective evaluation. J Neurol Neurosurg Psychiatry 58, 496–498. Manuguerra, M., Matullo, G., Veglia, F., Autrup, H., Dunning, A.M., Garte, S., Gormally, E., Malaveille, C., Guarrera, S., Polidoro, S. et al. (2007). Multi-factor dimensionality reduction applied to a large prospective investigation on gene-gene and gene-environment interactions. Carcinogenesis 28, 414–422. Marnett, L.J. (2000). Oxyradicals and DNA damage. Carcinogenesis 21, 361–370. Munnia, A., Bonassi, S., Verna, A., Quaglia, R., Pelucco, D., Ceppi, M., Neri, M., Buratti, M., Taioli, E., Garte, S. et al. (2006). Bronchial malondialdehyde DNA adducts, smoking and cyclin D1 G870A polymorphism. Free Rad Med Biol 41, 1499–1505. National Research Council (1987). Biological markers in environmental health research. Environ Health Perspect 74, 1–191. Panda, K., Chattopadhyay, R., Chattopadhyay, D.J. and Chatterjee, I.B. (2000). Vitamin C prevents cigarette smoke-induced oxidative damage in vivo. Free Radic Biol Med 29, 115–124. Pastinen, T., Raitio, M., Lindroos, K., Tainola, P., Peltonen, L. and Syvanen, A.C. (2000). A system for specific, high-throughput genotyping by allele-specific primer extension on microarrays. Genome Res 10, 1031–1042. Peluso, M., Neri, M., Margarino, G., Mereu, C., Munnia, A., Ceppi, M., Buratti, M., Felletti, R., Stea, F., Quaglia, R. et al. (2004). Comparison of DNA adduct levels in nasal mucosa, lymphocytes and bronchial mucosa of cigarette smokers and interaction with metabolic gene polymorphisms. Carcinogenesis 25, 2459–2465. Peluso, M., Munnia, A., Hoek, G., Krzyzanowski, M.,Veglia, F., Airoldi, L., Autrup, H., Dunning, A., Garte, S., Hainaut, P. et al. (2005). DNA adducts and lung cancer risk: A prospective study. Cancer Res 65, 1–7. Perera, F.P. (2000). Molecular epidemiology: On the path to prevention. J Natl Cancer Inst 92, 602–612. Perera, F.P. and Weinstein, I.B. (1982). Molecular epidemiology and carcinogen-DNA adduct detection: New approaches to studies of human cancer causation. J Chron Dis 35, 581–600. Piantadosi, S. (1997). Clinical Trials: A Methodologic Perspective. John Wiley and Sons, New York. Prentice, R.L. (1989). Surrogate endpoints in clinical trials: Definition and operational criteria. Stat Med 8, 431–440. Racusen, L.C., Solez, K., Colvin, R.B. et al. (1999). The Banff 97 working classification of renal allograft pathology. Kidney Int 55, 713. Raimondi, S., Boffetta, P., Anttila, S., Brockmoller, J., Butkiewicz, D., Cascorbi, I., Clapper, M.L., Dragani, T.A., Garte, S., Gsur, A. et al. (2005). Metabolic gene polymorphisms and lung cancer risk
■
307
in non-smokers. An update of the GSEC study Mutat Res 592, 45–57. Ridker, P.M., Hennekens, C.H., Roitman-Johnson, B., Stampfer, M.J. and Allen, J. (1998). Plasma concentration of soluble intercellular adhesion molecule 1 and risks of myocardial infarction in apparently healthy men. Lancet 351, 88–92. Sarwal, M., Chua, M.S., Kambhan, N. et al. (2003). Molecular heterogeneity in acute renal allograft rejection identified by DNA microarray profiling. New Engl J Med 349, 125. Scaruffi, P., Parodi, S., Mazzocco, K., Defferrari, R., Fontana,V., Bonassi, S. and Tonini, G.P. (2004). Detection of MYCN amplification and chromosome 1p36 loss in neuroblastoma by cDNA microarray comparative genomic hybridization. Mol Diagn 8, 93–100. Scherer, A., Krause, A., Walker, J.R., Korn, A., Niese, D. and Raulf, F. (2003). Early prognosis of the development of renal chronic allograft rejection by gene expression profiling of human protocol biopsies. Transplantation 75, 1323. Schulte, P.A. and Rothman, N. (1998). Epidemiological validation of biomarkers of early biological effect and susceptibility. Biomarkers: Medical and Workplace Applications. Joseph Henry Press, pp. 23–32. Smith, M.T., Vermeulen, R., Li, G., Zhang, L., Lan, Q., Hubbard, A.E., Forrest, M.S., McHale, C., Zhao, X., Gunn, L. et al. (2005). Use of “Omic” technologies to study humans exposed to benzene. Chem Biol Interact 153-154, 123–127. Swiger, R.R. and Tucker, J.D. (1996). Fluorescence in situ hybridization: A brief review. Environ Mol. Mutagen 27, 245–254. Tang, D., Phillips, D.H., Stampfer, M., Mooney, L.A., Hsu,Y., Cho, S.,Tsai, W.Y., Ma, J., Cole, K.J. and She, M.N. (2001). Association between carcinogen-DNA adducts in white blood cells and lung cancer risk in the physicians health study. Cancer Res 61, 6708–6712. Taioli, E. and Bonassi, S. (2002). Methodological issues in pooled analysis of biomarker studies. Mutat Res Rev 512, 85–92. Toniolo, P., Boffetta P., Shuker, D.E.G., Rothman, N., Hulka, B. and Pearce, N. (1997) Consensus Report. In “Application of biomarkers in cancer epidemiology” (IARC Scientific Publications, Ed) No. 142. IARC, Lyon. Ugolini, D., Puntoni, R., Perera, F.P., Schulte, P. and Bonassi, S. (2007). A bibliometric analysis of scientific production in cancer molecular epidemiology. Carcinogenesis 28, 1774–1779. van Leeuwen, D.M., van Herwijnen, M.H., Pedersen, M., Knudsen, L. E., Kirsch-Volders, M., Sram, R.J., Staal, Y.C., Bajak, E., van Delft, J.H. and Kleinjans, J.C. (2006). Genome-wide differential gene expression in children exposed to air pollution in the Czech Republic. Mutat Res 600, 12–22. Vineis, P. and F. Perera, F. (2000). DNA Adducts as markers of exposure to carcinogens and risk of cancer. Int J Cancer 88, 325–328. Walgren, J.L. and Thompson, D.C. (2004). Application of proteomic technologies in the drug development process. Toxicol Let 149, 377–385. Zhang, L., Eastmond, D.A. and Smith, M.T. (2002). The nature of chromosomal aberrations detected in humans exposed to benzene. Crit Rev Toxicol 32, 1–4.
CHAPTER
26 Validation of Candidate Protein Biomarkers Ingibjörg Hilmarsdóttir and Nader Rifai
INTRODUCTION The completion of the Human Genome Project is one of the most remarkable achievements in biology and medicine. It offers the promise of insights into the molecular basis of pathologies and possible means for diagnosis, treatment, and prevention of disease. However, information about the genotype is often insufficient for the understanding of a pathological process. After the transcription of genetic materials to mRNA and the subsequent translation to proteins, the latter undergo various structural modifications; the human genome consists of only 30,000 genes while over a million proteins are estimated to exist. Proteins therefore represent the phenotype and are functionally closer to the disease process than genes. Proteomics technology has evolved over the past decade to identify protein biomarkers that are useful in diagnosing, staging, and screening disease, and monitoring treatment. The process of proteomic biomarker discovery has, however, proven to be uncertain and more difficult than initially anticipated. Of more than 100 protein markers used clinically today, only 2 have been approved by the FDA since 1998: B-type natriuretic peptide (BNP) for the diagnosis of congestive heart failure and cancer antigen (CA) 19-9 for monitoring disease activity in pancreatic cancer. Mass spectrometry has been the preferred approach for candidate protein discovery, and plasma, considered to be the most comprehensive human proteome, is the preferred medium.
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 308
Human plasma candidates are present in a vast range of concentrations, from 1 pg/ml for interleukin-6 (IL-6) to 10 109 pg/ml for albumin. Although only low abundance proteins are targeted in candidate protein discovery, several hundred candidate proteins may be identified at the early stage, with the false-positive rate expected to be quite high. Qualification and verification studies in well-characterized populations and larger numbers of patients than initially used in the discovery phase will considerably reduce the number of proteins to 5–20. With current methods, candidate proteins will most likely be evaluated individually through the laborious, complex, and costly processes of research assay optimization, analytical and clinical validation, before the commercialization of the clinical assay. This chapter addresses these processes with the hope that familiarity with their strengths and limitations will aid the proteomics researchers in their quest for introducing novel clinical biomarkers. This chapter is in part based on an article by Rifai, N., Gillette, M.A. and Carr, S.A. (2006).
OPTIMIZATION OF THE CANDIDATE PROTEIN RESEARCH ASSAY After a candidate protein is identified, suitable antibodies must be developed to enable its quantification. Because the candidate
Copyright © 2009, Elsevier Inc. All rights reserved.
Analytical Evaluation
protein is assumed to have a plasma or serum concentration of only picograms to nanograms per milliliter, sensitive immunotechniques such as Radio Immuno Assay (RIA) or Enzyme Linked Immuno Sorbant Assay (ELISA), rather than turbidimetry or nephelometry, should be employed. These techniques are more sensitive than the more sophisticated non-immuno-based technologies such as liquid chromatography/tandem mass spectrometry and are readily available in clinical laboratory research and development settings (Vitzthum et al., 2005). Because the use of isotopes is discouraged in the clinical laboratory, ELISA offers a desirable alternative to RIA. As indicated earlier, IL-6 is one of the least abundant proteins in plasma that can be reliably measured by ELISA at concentrations as low as 0.15 pg/ml (coefficient of variation [CV] of 5%, internal data), with alkaline phosphatase or peroxidase as a label. Two antibodies, one for capture and another for detection, that recognize distinct epitopes of the protein are needed to form the ELISA sandwich reaction. Either monoclonal or polyclonal antibodies can be used and can be generated with a synthesized, recombinant, or isolated and purified protein. Before antibodies are used in an immunoassay, their specificity must be established by Western blot and immunostaining or other suitable techniques. Development and optimization of an ELISA must be undertaken with great care because ELISA performance is affected by a wide variety of variables, including the avidity and concentration of the capture and detection antibodies, incubation time and temperature, sample volume and dilution, pH and composition of diluent, enzyme and substrate type, and detector quality (Wild, 2005). Although general guidelines are recommended for choosing the appropriate antibody concentrations, sample dilutions, and incubation temperature and time, the optimal conditions can be determined only empirically. If the desired sensitivity cannot be achieved with an enzymatic colorimetric reaction, other alternatives such as fluorescent (fluorescein, europium chelate) or chemiluminescent (acridinium ester, isoluminol) tags may be used (Price and Newman, 1996). If the candidate protein is not present in healthy subjects, the standard curve can be constructed by adding different concentrations of the purified protein to a pool of normal plasma. Otherwise, other matrices such as bovine serum albumin (5 g/dl) can be used. A parallel between matrix-based and non–matrix-based calibrators must be established before the latter is deemed acceptable (Blirup-Jensen et al., 2001; Dati and Brand, 2000; Johnson et al., 2003). Finally, the appropriate curve-fitting model (linear, semi-log, log/log, or 4 parameter logistic) for calculation should be determined (Price and Newman, 1996). The analytical performance of an assay can be assessed only after the assay is completely optimized. Multiplexing technology, which is designed to simultaneously evaluate several novel proteins, may become a viable option in the future. Although this technique is currently available (e.g., Luminex), the simultaneous optimization of several protein assays is seldom achieved; for example, an ideal buffer for one protein, rather than being ideal when used for another protein, is likely to result in an analytically suboptimal assay that could jeopardize the clinical findings (Liu et al., 2005).
■
309
ANALYTICAL EVALUATION Before a newly developed assay is evaluated for clinical utility, its analytical performance must be carefully examined. For established measurands, the performance characteristic limits are specified according to outcome studies, clinical requirements, published professional recommendations (e.g., National Cholesterol Education Program performance recommendations for lipid measurements), proficiency testing findings (external quality assessment), or goals set by regulatory agencies (e.g., Clinical Laboratory Improvement Amendments [CLIA]) (Fraser and Petersen, 1999;Vitzthum et al., 2005). For novel measurands, performance specifications may not be available, but an analytical performance comparable to that of an established similar assay system should be expected. In addition, a common practice for establishment of analytical performance goals takes into consideration the biological variation both within (CVI) and between subjects (CVG) and stipulates that imprecision should be 0.5 CVI and bias 0.25 (CVI2CVG2)1/2. The main aspects of the analytical evaluation are presented below (Fraser and Petersen, 1999). Indicators of Accuracy Trueness and accuracy are similar but not identical concepts for assessing assay performance. Trueness is the closeness of agreement between the average measurand value of different samples and the true concentration value, whereas accuracy is the closeness of the agreement between the value of a measurement and the true concentration of the measurand in that sample (Dybkaer, 1997; ISO, 1994). Trueness reflects bias, a measure of systematic error, and accuracy reflects uncertainty, which comprises both random and systematic errors. Trueness can be assessed by the use of unbiased regression analysis (Deming) (Linnet, 1993) and residuals plots (Bland–Altman) (Bland and Altman, 1986) to compare the values obtained by the newly developed method with those determined by a reference method (as illustrated in Figure 26.1), or by measuring reference materials, with values assigned by a reference method or proficiency testing materials. For novel analytes for which reference methods and materials do not exist, recovery can be used as an alternative. In this procedure, various concentrations of the purified protein are added to a set of plasma or serum samples, and the recovery is determined by comparing the measured with the added amounts. Ideally, a recovery of 100% should be seen. Because antibody specificity is initially established by Western blotting and immunostaining, no significant cross-reactivity with other proteins is expected. It is imperative, however, to demonstrate that the newly developed assay is not affected by endogenous or exogenous interferences (NCCLS, 2002). Because in ELISA the wells are washed prior to the addition of the detection reagents, this technique is not affected by endogenous interferents such as hemoglobin, bilirubin, or triglycerides, a problem that often occurs with turbidimetric or nephelometric immunoassays. The presence of heterophilic, or anti-animal, antibodies in the patient sample, however, may lead to values that are falsely increased or
310
CHAPTER 26
Validation of Candidate Protein Biomarkers
■
12
3 Tina-quant CRP-N High Sensitivity CRP (mg/l)
Tina-quant CRP method (mg/l)
(a)
9
6
3
2 1 0 1 2 3
0 0
3 6 9 N High Sensitivity CRP method (mg/l)
12
(b)
0
2
4 6 8 Mean concentration (mg/l)
10
Figure 26.1 Comparison of results from the N hsCRP and Tina-quant methods. (a) Deming regression analysis. The dashed line represents the line of identity and the solid line represents the Deming regression line. Slope 0.958 (95% confidence interval, 0.954–0.962); intercept 0.280 (95% confidence interval, 0.268–0.292); Sy/x 0.264, r 0.992. (b) Bland–Altman plot. The solid line is the zero line. The thick dashed line indicates the mean difference (0.19 mg/l) and the thin dashed lines indicate the limits of agreement (0.36 to 0.74 mg/l) (Reprinted with permission from Lolekha et al., 2005).
decreased, ultimately resulting in disease misdiagnosis or patient mismanagement (Boscato and Stuart, 1988). For example, when human–mouse antibodies (HAMA) are present in the human sample being tested with an human choriogonadotropin (hCG) assay that employs two mouse antibodies, a false-positive result will occur if the HAMA recognize both the capture and the detection antibodies and form complexes indistinguishable from the hCG–antibody complex. A false-negative result will occur if the HAMA recognize only one antibody and make it inaccessible to the other antibodies, thus preventing complex formation. This effect can be minimized by the addition of either non-immune serum or IgG from the species used to raise the antibodies to the immunoassay reagents (Kricka, 1999). Indicators of Precision Repeatability and reproducibility are indicators of precision used to quantitatively express the closeness of agreement among results of measurements performed under stipulated conditions (Dybkaer, 1997; ISO, 1993). Repeatability refers to measurements performed under the same conditions and reproducibility to measurements performed under different conditions. Imprecision, which reflects random error and is expressed as a standard of deviation (SD) or a CV%, is assessed by the use of pools with different measurand concentrations (low, normal, high) over a defined period of time. Ideally the matrix of the pools should be identical to, or an appropriate simulation of, the matrix of the patient samples (Miller and Kaufman, 1993). The range of measurand values used in the series of pools to be tested should reflect, most importantly, the clinically relevant range, and also the span of the standard curve. For example, when Creactive protein (CRP) concentrations are high (100 mg/l), traditional CRP assays are suitable for detecting and monitoring
active infection and inflammation but inadequate for assessing risk of cardiovascular disease. The latter task requires high reproducibility at the lower concentration range (3 mg/l). Although some of these assays may be able to measure CRP at these concentrations, the CV% is often unacceptable (30–50%). High sensitivity CRP (hsCRP) methods can achieve such determinations with reproducibility of CV 5% and therefore are recommended for cardiovascular disease risk prediction. One recommended method for assessing precision calls for two replicates per specimen per run and two runs per day for 20 days (NCCLS, 1999). This method permits the estimation of within-run, within-day, between-run, between-day, and total precision. In general, assays are less imprecise at higher concentrations of the measurand than at lower concentrations, because assay designs generally result in direct proportionality between quantity of measurand and signal generated. Indicators of Analytical Measurement Range Linearity and limits of detection and quantitation must be assessed to determine the range of values of the measurand over which measurements can be performed with acceptable precision and accuracy (Currie, 1995; ISO, 2000; NCCLS, 2003). Linearity refers to the range of measurand values over which there exists a constant relationship between observed and expected values. Linearity is usually assessed by making serial dilutions of a sample containing a large quantity of measurand, using a diluent with a matrix similar to that of the undiluted specimen. Using a visual (graphic) and/or statistical assessment, this procedure will reveal whether the measured concentrations decline as expected, in proportion to the amount of dilution (Linnet, 2005). Furthermore, linearity assessment will determine the highest measurable value within the specified conditions.
Pre-Analytical Variation
The limit of detection is defined as the lowest value that significantly exceeds the measured value obtained with a blank sample (in this case a plasma or serum sample that does not contain the protein of interest). The limit of detection is calculated as the mean value of repeated blank sample measurements plus 2 or 3 SD. Although this approach is practical and widely used in clinical laboratories, it has recently been challenged by the introduction of a more complicated scheme by International Standard Organization (ISO) (ISO, 2000). The limit of quantitation is the lowest reported quantitative patient value within the established acceptable total error of the assay (Currie, 1995). For example, if the acceptable total error (bias 2 CV) for an ELISA method was set a priori at 30% (10% bias and 10% imprecision), then the limit of quantitation is the lowest value measured that conforms to these specifications. Ideally, these performance criteria should be based on the clinical needs of the test. Often however, at least initially, they are established arbitrarily.
■
311
Approaches used to develop reference intervals for biochemical markers vary according to the biomarker in question. Some markers, such as myoglobin, are usually present in apparently healthy individuals and increase in concentration in the disease state. In contrast, troponin is not normally present in circulation, but only values above the 99th percentile of the reference population are associated with cardiomyocyte injury (Apple et al., 2002). For certain analytes such as cholesterol, the developed cutoff values for risk assessment are not based on a reference population but on clinical evidence associating these values with increased risk of cardiovascular disease ( JAMA, 2001).
PRE-ANALYTICAL VARIATION Characterization and control of pre-analytical variation is essential for correct analyte measurement. Pre-analytical variation is influenced by both physiologic and non-physiologic variables (Figure 26.2) and can be minimized by standardizing sample collection procedures and making adjustments based on the limitations of the assay in various physiologic and pathologic conditions.
REFERENCE INTERVALS The frequency distribution of the candidate protein in the target population must be examined in apparently healthy individuals to establish reference intervals to which patient results can be compared (Solberg, 1987; Solberg and PetitClerc, 1988). The examined population must be similar to that in which the test would be used, and at least 120 subjects should be included in the sample group (Harris and Boyd, 1995; Solberg and PetitClerc, 1988). In the event of statistically significant differences in the range of values related to differences in age, sex, race, or physiological states (e.g., puberty, menopause), the reference intervals must be stratified into appropriate subgroups and a larger sample size must be used (Harris and Boyd, 1990). For example, alkaline phosphatase activity changes differently with age among males and females in the first two decades of life, thus necessitating the use of sex- and age-specific cutoff values for clinical interpretation. In contrast, studies have demonstrated that age, sex, and ethnicity do not significantly affect the population distribution of hsCRP, indicating the appropriateness of using a single set of cutoff values for cardiovascular disease risk prediction in the entire population. If the distribution of the protein of interest is gaussian, then parametric analysis is used and the reference intervals are presented as mean 2SD (Harris and Boyd, 1995; Solberg, 1983). However, if the distribution is skewed, then non-parametric analysis is needed and the reference intervals may be presented as the lower 2.5th to the upper 97.5th percentiles. An alternative approach is to log transform the data to approximate the gaussian distribution, identify the 2.5th and the 97.5th percentiles, and then transform back the estimates of the upper and lower limits of the range to the original measurement scale for use as the reference intervals (Harris and Boyd, 1995; Solberg, 1983). Other statistical methods for estimating reference intervals, such as bootstrap methods, have also been used (Harris and Boyd, 1995).
Patient Status and Specimen Collection and Handling An important determinant for specimen collection is whether the sample needed for the measurement must be collected while the person undergoing the procedure is in a fasting state, because fasting versus non-fasting status can affect measurements both physiologically and analytically. An example is interference by postprandial hypertriglyceridemia, which is not a significant issue for ELISA, since unlike other techniques such as nephelometry and turbidimetry is unaffected by the presence of chylomicrons. Another consideration is time of collection, because many analytes, such as IL-6, exhibit significant circadian rhythm fluctuations and therefore, samples for their measurement must be collected at a set time (e.g., in the morning) (Gudewill et al., 1992). The diurnal variation of the candidate protein can be determined by measurement in samples collected from individuals over a 24-hour period (Meier-Ewert et al., 2001). During the sample collection process, it is important to note that factors such as the length of time of tourniquet application and the posture of the patient (e.g., supine or seated position) may affect the analyte concentration by up to 15% (Tan et al., 1973). Serum and plasma (anticoagulated with heparin, citrate, or EDTA) samples are widely used in the clinical laboratory. Although the choice of the appropriate type of specimen for the candidate protein should be empirically determined, one should be aware of certain caveats. EDTA and citrate cause osmotic shifting of fluid from cells to plasma, leading to value reductions of up to 10% compared to serum (Cloey et al., 1990). If tubes are not completely filled with blood during collection, larger effects can be expected, especially with citrate. EDTA-containing plasma is not suitable when the detection enzyme used requires a cation (Ca, Mg) to be active or when the protein itself contains a cation. Pregnancy-associated plasma protein
312
CHAPTER 26
■
Validation of Candidate Protein Biomarkers
Variables known to affect test results
Pre-analytic variables
Analytic and post-analytic variables
Non-physiologic
Physiologic
Patient preparation
Race
Time of collection
Age
Method of collection
Sex
Sample type and quality
Season
Transportation Sample treatment and stability
Laboratory methodology Sensitivity Precision Antigen excess Matrix effects
Lifestyle (exercise, smoking, obesity, alcohol, antiinflammatory drugs, hormone therapy, etc.)
Calibration and curve fitting Method correlation Interpretive report
Other (pregnancy, altitude)
Figure 26.2
Variables known to affect test results.
A (PAPP-A), for example, contains 16 zinc atoms (Apple et al., 2005), and chelation of the zinc by EDTA changes the configuration of the protein and reduces its immunoreactivity, leading to falsely reduced results. Heparin may mask the antigenic epitope and prevent appropriate binding with the antibody, as also occurs with PAPP-A (Apple et al., 2005). If the protein of interest is partially or mainly derived from platelets (e.g., plasminogen activator inhibitor-1, monocyte chemoattractant protein-1, or CD40L), platelet-poor plasma is required to prevent the results from being falsely increased. Finally, length of sample storage time, short- or long-term, and storage conditions are of paramount importance because the clinical utility of the candidate protein will likely be assessed using stored samples from an appropriate clinical trial. For example, CRP is a very stable protein that can be stored up to 20 years at 20°C, whereas tumor necrosis factor (TNF) is a labile protein that requires sample collection on ice and storage at 70°C or in liquid nitrogen (Ledue and Rifai, 2003). Because of the instability of TNF, researchers have used TNF receptors I and II, which are stable, as surrogates for TNF. In some cases, such as for atheranol, ascorbic acid and butylated hydroxytoluene are added immediately after plasma separation to stabilize the analyte. Such steps, however, may hinder the utility of the marker. Physiologic Considerations Within- and between-subject variation must be considered when characterizing the clinical utility of a new analyte. Biological variability differs greatly among analytes (42% for triglyceride and 6.5% for cholesterol) and is usually higher than the analytical variability (imprecision of 1–2% for triglyceride and cholesterol) (Cooper et al., 1995).Total error, which encompasses
both biological and analytical variability, is used to determine the number of measurements needed to make a diagnosis. For example, for both cholesterol and hsCRP, the mean of two independent measurements made several weeks apart is used to classify individuals into categories of risk for cardiovascular disease (Ledue and Rifai, 2003). Data from reference interval studies enable determination of whether age, sex, ethnicity, or race significantly affect protein concentrations. As indicated earlier, this information is important in developing interpretive criteria and determining whether single or multiple cutoff values specific for the various subgroups need to be established (Rifai and Ridker, 2003). Furthermore, data from the clinical evaluation of the protein assay can be used to determine whether lifestyle factors (smoking, alcohol intake, diet, exercise, and obesity), ancillary pathological conditions, or certain drugs influence protein concentrations. Studies specifically designed to address these issues may have to be conducted in the future.
CLINICAL EVALUATION As indicated earlier, diagnostic markers are used for the diagnosis, staging, screening, and prediction of a disease, for the prediction and monitoring of treatment response and measure of treatment compliance (Vitzthum et al., 2005). The clinical evaluation of the candidate protein’s potential for one or more of these indications starts when the analytical evaluation is complete. The clinical evaluation is done through a series of carefully planned diagnostic studies involving patients with and without the target condition. This section will focus on the indicators of
Indicators of Diagnostic Accuracy and Predictability
■
313
T A B L E 2 6 . 1 Diagnostic performance of BNP testing for the diagnosis of congestive heart failure in dyspneic patients in the emergency room (adapted from Maisel et al. [2002]) Diagnostic test results (BNP 100 pg/ml)
Disease status (CHF)
Positive
Negative
Total
Positive
673 TP
71 FN
744 TP FN
Negative
FP 198
TN 644
FP TN 842
TP FP 871
FN TN 715
TP FP FN TN 1586
Total
Sensitivity TP/TP FN 673/744 90% Specificity TN/TN FP 644/842 76% LRpos sensitivity/(1 specificity) 0.9/(1 0.76) 3.75 LRneg (1 sensitivity)/specificity (1 0.9)/0.76 0.13 Predictive value of a positive test TP/TP FP 673/871 77% Predictive value of a negative test TN/TN FN 644/715 90% Pre-test probability of having CHF (prevalence) TP FN/TP FN TN FP 673 71/673 71 644 198 47% Pre-test odds of having CHF prevalence/(1 prevalence) 0.47/(10.47) 0.89 Post-test odds of having CHF pre-test odds LR pos 0.89 3.75 3.34 Probability odds ratio/(odds ratio 1) Post-test probability of having CHF post-test odds/(post-test odds 1) 3.34/(3.34 1) 77% Abbreviations: BNP, B-type natriuretic peptide; TP, true-positive; FN, false-negative; FP, false-positive; TN, true-negative; LRpos, likelihood ratio for a positive test result; LRneg, likelihood ratio for a negative test result; CHF, congestive heart failure.
diagnostic accuracy and predictability, diagnostic research studies, importance of design quality, and issues regarding the transferability of diagnostic test performance. For simplification, the term “disease” will be used for the target condition under study.
INDICATORS OF DIAGNOSTIC ACCURACY AND PREDICTABILITY The diagnostic performance of a test involves its diagnostic accuracy and predictability. These are commonly measured with a handful of basic methods; some of which have been elaborated further for complex situations and are then best performed by the use of statistical computer programs. Table 26.1 shows the calculation of diagnostic performance of BNP testing for the diagnosis of congestive heart failure. Diagnostic accuracy estimates how accurately the test discriminates between individuals who have the disease and those who don’t. Diagnostic accuracy is measured by calculating the test’s sensitivity and specificity, which can be further examined through the likelihood ratio and receiver operating characteristic (ROC) curve. Sensitivity is the proportion of the diseased that are correctly identified by the test while specificity is the proportion of the nondiseased that are correctly identified. When test results in the two groups do not overlap, the test achieves perfect discrimination. In most instances this is not the case though; therefore, a trade-off between sensitivity and specificity is inevitable (Figure 26.3).
Sensitivity and specificity characteristics of a test are not absolute; they depend on several phenomena. Firstly, sensitivity and specificity are affected by the distribution of results in the diseased (which usually correlates with the spectrum of disease) and the non-diseased groups, respectively, and the distribution varies between populations. Therefore, the diagnostic accuracy characteristics observed for a given population are not necessarily applicable to other populations (Moons et al., 1997). The example in Table 26.2 shows how the spectrum of left ventricular systolic dysfunction affects the sensitivity of BNP testing for the diagnosis of congestive heart failure. Likewise, the specificity is affected by the spectrum of values in the non-diseased group. For example, BNP values are higher in individuals with impaired renal function (McCullough et al., 2003), but free of left ventricular systolic dysfunction, than in healthy ones. If the non-diseased group (i.e., free of left ventricular systolic dysfunction) includes such patients, the curve for the distribution of BNP values will be shifted to the right on the x-axis (see Figure 26.3). A larger proportion of the group would thus fall into the false-positive category, and the specificity of the test would be reduced compared to that seen in a healthy group. Secondly, the reference standard (“gold standard”) against which the test is measured should provide a clear and binary – present or absent – definition of the disease of interest. It can be simple or composite, consisting of laboratory tests and/or clinical data. In rare occasions, the true disease state is represented by a continuous scale of values, which requires the choice of cutoff values
314
CHAPTER 26
■
Validation of Candidate Protein Biomarkers
in order to obtain a binary standard (Obuchowski et al., 2004).The reference standard for a particular disease may vary across studies; such variation could lead to different diagnostic accuracy. Ideally, the reference standard should identify correctly the true disease condition of all study subjects; however, this is not always the case. Thirdly, the study test itself requires a definition of positive and negative test results. In the case of continuous scale results, this definition is often unknown in the early phases of the clinical evaluation and is determined after evaluation of different cutoff values against the reference standard. An example of how diagnostic performance varies with the cutoff value is shown in Table 26.3. More complex situations in which the disease and/ or the test results are non-binary can be analyzed with the use of multiple and logistic-regression models (Altman, 1991).
Frequency in population
Non-diseased Threshold Diseased TN
TP
FN
Negative
FP
Test value
Positive
Figure 26.3 Hypothetical distribution of test results in a population of diseased and non-diseased. The location of the threshold for a positive test (cutoff value), as well as the location and shape of the curves affect the diagnostic performance of the test.
Sensitivity and specificity are calculated from two different groups, that is, the diseased and the non-diseased, respectively, and are therefore not dependent on prevalence; the same is true for their derivatives, the likelihood ratio and the ROC curve. Likelihood ratio for a positive test result (LRpos) is the ratio of the proportion of the diseased that are correctly identified by the test (true-positives) to the proportion of the non-diseased that are wrongly identified (false-positives). Inversely, the likelihood ratio for a negative test result (LRneg) is the proportion of the diseased that are missed by the test (false-negatives) to the proportion of the non-diseased that are correctly identified (true-negatives). As likelihood ratios combine information about a test’s sensitivity and specificity, they can be used to calculate post-test probability of a disease from its pre-test probability. This is done by the use of Bayes’ theorem according to which the post-test odds pre-test odds LR (Altman, 1991). The pre-test probability represents the prevalence of a disease in the population that is subjected to the test. The post-test probability represents the chances of having a disease after test results have been taken into account. The LRpos and LRneg provide an estimate of how much a positive or negative test result will increase or decrease, respectively, the pre-test odds of having a disease. While high LRs increase the certainty about a positive test being truly positive, test results for each subject still need to be interpreted in light of the pre-test odds of having the disease. A positive test result in a low prevalence population is more likely to be falsely positive than the same result in a high prevalence population (Altman, 1991). The example in Table 26.1 shows that a cutoff value of 100 pg/ml for BNP for the diagnosis of congestive heart failure had an LRpos of 3.75, which means that this result is almost four times as likely to occur in patients with the disease as in
TABLE 26.2 A study that examined the usefulness of BNP for screening for LVSD reveals how the spectrum of the target condition affects sensitivity of the test % (95% confidence interval) Severity of LVSD
Sensitivity
Specificity
PPV
NPV
LRpos
LRneg
LVEF 50%
14 (4–25)
95 (94–96)
7 (2–12)
98 (97–98)
2.94
0.9
LVEF 40%
40 (10–70)
95 (94–96)
5 (0.2–9)
96 (99–100)
8.08
0.63
The table shows the diagnostic performance for women (n 42) at a cutoff value of 50 pg/ml ( Vasan et al., 2002). The “LVEF 50%” and “LVEF 40%” groups differ with regard to spectrum of disease; the former includes mild, moderate, and severe cases, whereas the latter includes only moderate and severe ones. The curve for the distribution of BNP values for the group including mild cases is shifted to the left on the x-axis (see Figure 26.1), compared to the other group, and therefore the chosen cutoff value increases the fraction of false-negatives and detects fewer cases of impaired LVEF (sensitivity of 14% versus 40%). Abbreviations: BNP, B-type natriuretic peptide; LVSD, left ventricular systolic dysfunction; LVEF, left ventricular ejection fraction.
T A B L E 2 6 . 3 A study of BNP testing for the diagnosis of CHF in dyspneic patients in the emergency room (n 1586) reveals how diagnostic performance characteristics vary with different cutoff values for the test (Maisel et al., 2002) % (95% confidence interval) BNP pg/ml
Sensitivity
Specificity
PPV
NPV
LRpos
LRneg
50
97 (96–98)
62 (59–66)
71 (68–74)
96 (94–97)
2.55
0.05
150
85 (82–88)
83 (80–85)
83 (80–85)
85 (83–88)
5.00
0.18
Indicators of Diagnostic Accuracy and Predictability
persons without it. In this case a positive test result increases the probability of having congestive heart failure from 47% to 77%. In contrast, an LRneg of 0.13 decreases the probability of having congestive heart failure from 47% to 10% in those patients having a negative test. This example involved a likelihood ratio for only one test. When two or more independent tests are used, their likelihood ratios can be multiplied to assess the combined value of the tests in increasing the diagnostic certainty. Although the likelihood ratio incorporates information about sensitivity and specificity, it does not show the actual percentages and should therefore not be used alone to assess or compare test performance (Zweig and Campbell, 1993). This was demonstrated in the study cited in Figure 26.4 in which the LRpos for a clinical diagnosis of congestive heart failure was 11.5, whereas for BNP it was 3.4. The sensitivity and specificity were 49% and 96% for the former, respectively, and 90% and 73% for the latter (McCullough et al., 2002). An additional interesting feature of the likelihood ratio, besides estimating post-test odds, is that unlike sensitivity and
1.0 Combined BNP
0.9 0.8
E.D. Probability
Sensitivity
0.7 0.6
Area Under ROC Curve 0.86 (0.84–0.88) E.D. Probability 0.90 (0.88–0.91) BNP 0.93 (0.92–0.94) Combined
0.5 0.4 0.3 0.2 0.1 0.0 0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
1-Specificity
Figure 26.4 A study of BNP testing in dyspneic patients in the emergency room (n 1586) examined whether BNP added to clinical judgement in the diagnosis of congestive heart failure. The graph showing the ROC curves for clinical diagnosis (E.D. Probability), BNP testing, and the combination of both demonstrates how the area under the curve can be used to compare diagnostic accuracy for two or more tests and test combinations. When the criterion of BNP values at 100 pg/ml was added to a high clinical suspicion of congestive heart failure (80 100% probability, as estimated by the emergency department physicians), the area under the curve increased from 0.86 to 0.93 and the proportion of correctly diagnosed cases increased to 81.5% from 74% (TP TN/TP FN FP TN) (McCullough et al., 2002).
■
315
specificity calculations, which require binary test results, the LR can be obtained for multicategory test results (Deeks and Altman, 2004). The ROC curve provides a graphic index of the accuracy of quantitative tests. The majority of tests in the clinical laboratory yield a numerical value on a continuous scale, and, in order to differentiate between the diseased and non-diseased in the population, an appropriate cutoff value needs to be chosen. A convenient way to examine the potential cutoff values is to plot their performance on a graph to obtain the ROC curve. The curve shows the fraction of true-positive results (sensitivity) over the fraction of false false-positive results (1–specificity) for the entire range of values and thus depicts the overlap between the distribution of test results in the diseased and non-diseased groups. When there is no overlap, and thus perfect discrimination between the two groups, the curve passes through the upper left corner, and the area under the ROC curve (AUROC) is 1. In case of a complete overlap, that is, the true-positive fraction equals the false-positive fraction, the plot generates a 45° diagonal line from the lower left corner to the upper right corner, and the AUROC is 0.5. Most often the curve falls between these two extremes. The larger the AUROC the better, although it should always be appreciated along with the curve, since the area by itself does not convey information about the curve’s shape. The main advantage of the ROC curve is to enable: (i) visualization of the performance of potential cutoff values in the same graph, although the actual values are not directly visible; (ii) comparison of the curves generated from two or more tests, as demonstrated in Figure 26.4; (iii) measure of accuracy by calculating the AUROC, and (iv) discriminant analysis and logistic regression that allow the incorporation of other tests or covariates to obtain the optimal cutoff value and improve the diagnostic approach (Obuchowski et al., 2004; Zweig and Campbell, 1993). Predictive values show the ability of the test to predict the presence or absence of disease for a given test result. The positive predictive value is the proportion of patients with positive test result who have the disease, and the negative predictive value is the proportion of patients with negative test who do not have the disease. The predictive values of a test vary with the prevalence of the disease in the population examined. As prevalence increases the positive predictive value increases and the negative predictive value decreases, and vice versa. To demonstrate how predictive values change with prevalence of the condition under study, let’s imagine that the number of patients with congestive heart failure in Table 26.1 were 1000 instead of 744, and the prevalence therefore 63% instead of 47%. Given that sensitivity and specificity remain unchanged, the new positive predictive value would increase to 86% from 77% and the negative predictive value would decrease to 82% from 90%. As mentioned before, the Bayes’ theorem allows one to calculate the post-test probability (i.e., positive predictive values) for any prevalence of disease, by using the prevalence and the likelihood ratio derived from previous studies. A nomogram that enables an easy appreciation of post-test probability was created from Bayes’ theorem (Deeks and Altman, 2004; Fagan, 1975).
316
CHAPTER 26
■
Validation of Candidate Protein Biomarkers
The use of Bayes’ theorem is limited, however, by the influence of test result distribution in the diseased and non-diseased groups on diagnostic accuracy and therefore likelihood ratio, and this should be borne in mind when post-test probabilities are calculated using likelihood ratios from studies on other populations.
following both patient groups with regard to post-test diagnostic and therapeutic interventions and the subsequent effect on health outcome (Sackett and Haynes, 2002).
DESIGN OF DIAGNOSTIC STUDIES DIAGNOSTIC RESEARCH STUDIES In order to establish its clinical utility, the candidate protein needs to be evaluated in sequential phases that emphazise different performance characteristics of the test and therefore require different study populations (Sackett and Haynes, 2002). The exploratory phase (Phase I) usually involves patients with confirmed disease and people known not to have the disease. The question here is: are test results different for these two groups? The ROC curve is often used to examine the discriminative potential of the test. If the AUROC is equal or less than 0.5 the test is not useful as it does not distinguish the diseased from the non-diseased. Further studies are not indicated in these cases. In contrast, satisfactory discrimination is suggestive of a useful test, and Phase II studies will subsequently determine its diagnostic accuracy (Obuchowski et al., 2004; Sackett and Haynes, 2002). The challenge phase (Phase II) often involves similar patient groups as Phase I, but here the question is: do test results predict disease or absence of disease? This may be addressed by assessing different cutoff values for the calculation of sensitivity and specificity. Hence, the diagnostic accuracy of the test is determined for this well-defined study population. At this stage, it is still not known how the test will perform in the population that will ultimately benefit from it, that is, those suspected of having the disease. This is addressed in Phase III studies (Obuchowski et al., 2004; Sackett and Haynes, 2002). The choice of cutoff values is based on the most appropriate sensitivity/specificity combination with regard to the purpose of the test, the prevalence in the target population, cost and the potential harm and benefit associated with incorrect and correct disease classification, respectively (Zweig and Campbell, 1993). The advanced phase (Phase III) involves patients who are suspected of having the disease. The question here is: what are the diagnostic accuracy and predictive values in the target population for whom the test is intended in routine clinical practice? Whereas earlier research phases are conducted on well-defined and controlled study groups, Phase III involves patients that may vary greatly across different health care and geographical settings. It follows that test performance may vary accordingly and performance characteristics may not be transferable between settings. Diagnostic tests should be validated in independent studies before they are adopted in clinical practice (Obuchowski et al., 2004; Sackett and Haynes, 2002) Outcome phase (Phase IV) involves patients who undergo the study test and a comparable group of patients who do not. The question here is: does the test influence positively the ultimate health outcome of tested patients? The answer is obtained by
It is the responsibility of those who plan diagnostic studies to strive toward the highest quality in order to maximize the health benefit for the individual patient and the population, and minimize costs related to inefficient diagnostic testing and misuse of resources in poor research. A well-designed study can be analyzed in a number of ways, but no conclusion can be derived from a poor one. The major components of a study design include the patient group, reference standard, interpretation of test and standard with regard to blinding, data collection, and analysis. Any of these can lead to biased study results, usually toward an overly optimistic performance of the test. Meta-analyses of the effect of potential bias factors on the estimates of diagnostic accuracy showed that a study population consisting of severe cases and healthy control (as opposed to a cohort population), non-consecutive inclusion of subjects, retrospective data collection, different reference standards for different study test results, non-blinding of test interpretation, incorporation of test results into the reference standard, and post hoc definition of cutoff value for positivity tend to overestimate the diagnostic odds ratio (Lijmer et al., 1999; Rutjes et al., 2006). Rutjes et al. concluded that “studies of the same test can produce different estimates of diagnostic accuracy depending on choices in design” (Rutjes et al., 2006). Furthermore, it is increasingly recognized that advanced phase studies need to reflect a realistic clinical setting. The diagnostic process is a multifactorial one that includes patient history, physical examination, and diagnostic tests that vary in complexity, invasiveness, performance, and cost. The design of diagnostic studies should be adapted to real life in the clinical setting, where the choice of a particular test is usually guided by a suspicion of a disease, based on previous examination results. Diagnostic tests may have excellent diagnostic accuracy without being clinically helpful. More convenient or cheaper diagnostic processes may perform equally well or the nature of the disease may not justify the inconvenience and cost of a particular test. The real contribution of a test can be assessed by the use of multivariable analyses. Discerning mutual dependencies between different examination results permit to establish whether the test is redundant or has an independent predictive value for the diagnosis of the disease (Moons et al., 2004). This was done in a study of BNP testing for the diagnosis of congestive heart disease in dyspneic patients in the emergency room (n 1586; see Table 26.3). A multiple logistic-regression analyses including clinical, radiological, and laboratory data revealed that the test was not only the strongest independent indicator of congestive heart disease, but also added significant additional information when it was entered after other clinical indicators (Maisel et al., 2002).
Assay Transfer to a Diagnostic Company
TRANSFERABILITY OF TEST PERFORMANCE When a test shows a promising diagnostic performance in a known population, it is tempting to adopt the test for use in other populations, on the sole basis of its performance characteristics. This simple approach has pitfalls though, because performance characteristics are not “fixed.” The diagnostic accuracy, measured by sensitivity, specificity, LR and the ROC curve, and predictive values of a test may vary across patient populations (Moons et al., 2004; Sackett and Haynes, 2002). The transferability of a test depends on several criteria that affect diagnostic accuracy and predictive values (Irwig et al., 2002). In order to obtain equivalent diagnostic performance in a new clinical setting, all of the following characteristics need to be comparable to the study setting: test characteristics, including calibration (see “Indicators of Accuracy”) and manufacturer, disease definition, and the population to be tested. Populations may differ with regard to distribution of test results in the diseased and non-diseased groups (Irwig et al., 2002), thus giving rise to different diagnostic accuracy and predictive values. Table 26.4 demonstrates how diagnostic performance can vary between two populations. Whenever test characteristics and disease definition in the new setting differ from those used in reported studies it is wise to validate the test with regard to diagnostic accuracy before its implementation. The question to be asked is whether this is the appropriate test for the setting. Furthermore, as predictive values are dependent on prevalence, they need to be recalculated in the new population. As explained above, the predictive values can be estimated in a new setting by the use of the Bayes’ theorem.
ASSAY TRANSFER TO A DIAGNOSTIC COMPANY Once evidence for the clinical usefulness of a novel marker has been obtained, firms producing in vitro diagnostics (IVD) will determine whether the marker is worth pursuing
■
317
commercially. Many technical, medical, financial, and legal factors are considered in making the decision (Vitzthum et al., 2005). Intellectual property issues relating to the protein, the antibodies used, and the intended clinical use may limit the availability of the marker and hinder commercial pursuit. For example, Roche Diagnostics has exclusivity for troponin T, which is measured in clinical laboratories only on the Elecsys platform. As a result, fewer than 300 labs in the United States are using this assay compared to more than 2500 labs using troponin I, measured on platforms from eight different manufacturers (CAP Survey, 2005). Dade Behring has the exclusive license for marketing hsCRP for the prediction of cardiovascular disease. Other manufacturers of hsCRP reagents (30 worldwide) must obtain a sublicense from Dade Behring if their product is intended to be used in risk assessment of heart disease. These legal matters tend to complicate the process and increase the cost of introducing new tests to the market, potentially limiting wide availability. The developed assay must be suitable for modern clinical laboratories and meet basic requirements (e.g., robustness, reagent stability, ease of use, acceptable turnaround time, adaptability to automated immunoassay systems, and low cost). Most major IVD manufacturers have immunoassay instrument platforms with non-isotopic detection systems that enable the transfer of a research ELISA to an automated clinical laboratory test format. The manufacturer then must evaluate the optimized assay in the context of the new platform and protocol, analytically and clinically, to satisfy regulatory agencies. It is important to note that often multiple generations of a particular assay must be developed as the perceived clinical utility of the marker evolves and/or the technology advances. For example, a third-generation troponin T assay is currently in clinical use, with a fourth generation under evaluation. The sensitivity and the performance of the assay improved over the years, thus enabling the detection of minor cardiomyocyte injury and permitting the clinician to use the assay as a diagnostic and a prognostic tool in patients with acute coronary syndromes. Although improvements in assays are useful to clinicians and beneficial to patients, they are quite costly to IVD
TABLE 26.4 Two studies evaluated the use of BNP for the diagnosis of CHF in dyspneic patients in the emergency room, using the same commercialized test and cutoff value % (95% confidence interval) Studies
Sensitivity
Specificity
PPV
NPV
LRpos
LRneg
Area under ROC curve
Study 2001 (Dao et al., 2001)
94 (89–97)
94 (89–97)
92 (85–96)
96 (91–98)
15.7
0.06
0.98
Study 2002 (Maisel et al., 2002)
90 (88–92)
76 (73–79)
79 (76–81)
89 (87–91)
3.75
0.13
0.91
The studies differed with regard to sex distribution of the patients (95% versus 56% male) and perhaps race (unknown versus 45% African Americans), and number of study sites and teams evaluating true disease status (1 versus 7). Although the test was shown to be useful in both studies, the impressive diagnostic performance reported in the earlier study was not repeated in the second one.
318
CHAPTER 26
■
Validation of Candidate Protein Biomarkers
manufacturers, requiring development, evaluation, and validation of the assay, and often do not lead to a significant increase in the marketability of the test. These issues can only make IVD manufacturers more cautious in pursuing novel markers.
REGULATORY REQUIREMENTS In many countries, for an IVD device to enter the market, it must conform to a set of rules and regulations such as those of the FDA in the United States (FDA, 2005, 2006), the Pharmaceutical and Medical Safety Bureau in Japan (www. mhlw.go.pj/english/org/policy/p13-4.html), and the European Directive in the European Union (Dati, 2003). Regulations are somewhat similar among these agencies, and here we discuss the FDA process as an example. According to the Medical Device Amendments of 1976 (Greenberg, 1976), an IVD device to be commercialized in the US market must undergo one of two primary regulatory processes stipulated by the FDA; 510(k) clearance or Pre-Market Approval (PMA). The 510(k) process is used when the new test measures an existing FDA-classified analyte (Class I or II) for which there exists a commercially available predicate test method that has been cleared by the FDA or was in commercial distribution prior to May 28, 1976 (pre-amendment devices [2005]). The submitted information required for the new test includes its intended use and classification, data comparing its performance to the predicate device, and characterization of its analytical capability (e.g., precision, linearity, specificity, and accuracy compared to the predicate device, determined via patient correlation studies). This process requires 100 days for FDA review time and User Fees of $3800 (www.fda.gov/oc/
mdufma/coversheet.html). The PMA process is used when the test is classified as Class III (the riskiest devices [e.g., cancer diagnostics]), or the clinical utility of the marker or the technology of the measurement are novel and no predicate device can be identified (2006). For these tests the required information includes the same data required for 510(k) as well as clinical outcomes data that present independent clinical criteria establishing that the concentration of the marker is related to disease status. This process requires 180–360 days to review, and often requires an FDA panel meeting and User Fees of ~$240,000 (www.fda. gov/oc/mdufma/coversheet.html). The FDA recognizes that some newly discovered analytes that have no obvious predicate devices do not have safety concerns that automatically trigger a Class III designation. To address these situations, in 1997 the agency created a new hybrid classification termed “de novo” or “risk-based” (FDA, 1998). This process allows a new analyte to be regulated as in a 510(k) but requires the demonstration of clinical effectiveness. Although the de novo process provides a viable third option for novel marker registration, the regulatory route is not always obvious and is determined ultimately by the FDA. BNP is an example of a novel marker that was cleared by the FDA using the de novo route. This process requires 150–180 days for FDA’s review and clearance, and User Fees of $3800. Therefore, in the absence of a predicate device and depending on the intended use and clinical utility, either a PMA or a de novo process will be required before a new protein of interest becomes commercially available. Because of the nature of protein measurement, new proteins are classified in the United States according to CLIA in the “high complexity testing” category and subjected to all analytical performance requirements for that class (www.phppo.cdc.gov/clia/regs/).
REFERENCES Altman, D.G. (1991). Practical statistics for medical research. Chapman & Hall. Apple, F.S., Wu, A.H. and Jaffe, A.S. (2002). European Society of Cardiology and American College of Cardiology Guidelines for redefinition of myocardial infarction: How to use existing assays clinically and for clinical trials. Am Heart J 144, 981–986. Apple, F.S., Wu, A.H., Mair, J., Ravkilde, J., Panteghini, M., Tate, J., Pagani, F., Christenson, R.H., Mockel, M., Danne, O. et al. (2005). Future biomarkers for detection of ischemia and risk stratification in acute coronary syndrome. Clin Chem 51, 810–824. Bland, J.M. and Altman, D.G. (1986). Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1, 307–310. Blirup-Jensen, S., Johnson, A.M. and Larsen, M. (2001). Protein standardization IV: Value transfer procedure for the assignment of serum protein values from a reference preparation to a target material. Clin Chem Lab Med 39, 1110–1122. Boscato, L.M. and Stuart, M.C. (1988). Heterophilic antibodies: A problem for all immunoassays. Clin Chem 34, 27–33.
Cloey,T., Bachorik, P.S., Becker, D., Finney, C., Lowry, D. and Sigmund,W. (1990). Revaluation of serum-plasma differences in total cholesterol concentration. JAMA 263, 2788–2789. Cooper, G.R., Smith, S.J., Myers, G.L., Sampson, E.J. and Magid, E. (1995). Biological variability in the concentration of serum lipids: sources, meta-analysis, estimation, and minimization by relative range measurements. J Int Fed Clin Chem 7, 23–28. Currie, L. (1995). Nomenclature in evaluation of analytical methods including detection and quantification capabilities (IUPAC Recommendations 1995). Pure Appl Chem 67, 1699–1723. Dao, Q., Krishnaswamy, P., Kazanegra, R., Harrison, A., Amirnovin, R., Lenert, L., Clopton, P., Alberto, J., Hlavin, P. and Maisel, A.S. (2001). Utility of B-type natriuretic peptide in the diagnosis of congestive heart failure in an urgent-care setting. J Am Coll Cardiol 37, 379–385. Dati, F. (2003). The new European directive on in vitro diagnostics. Clin Chem Lab Med 41, 1289–1298. Dati, F. and Brand, B. (2000). Standardization activities for harmonization of test results. Clin Chim Acta 297, 239–249. Deeks, J.J. and Altman, D.G. (2004). Diagnostic tests 4: Likelihood ratios. BMJ 329, 168–169.
References
Dybkaer, R. (1997). Vocabulary for use in measurement procedures and description of reference materials in laboratory medicine. Eur J Clin Chem Clin Biochem 35, 141–173. Executive Summary of The Third Report of The National Cholesterol Education Program (NCEP) (2001). Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol In Adults (Adult Treatment Panel III). JAMA 285, 2486–2497. Fagan, T.J. (1975). Letter: Nomogram for Bayes theorem. N Engl J Med 293, 257. FDA (1998). Classification of devices intended for human use. In Code Federal Register, Vol. 21 USC. FDA (2005). Establishment registration and device listing for manufacturers and initial importers of devices. In Code Fedral Regulations, Vol. 21 CFR. FDA (2006). Premarket approval of medical devices. In Code Federal Regulations, Vol. 21 CFR 814.1. Fraser, C.G. and Petersen, P.H. (1999). Analytical performance characteristics should be judged against objective quality specifications. Clin Chem 45, 321–323. Greenberg, R. (1976). Medical device amendments of 1976. Am J Hosp Pharm 33, 1308–1311. Gudewill, S., Pollmacher, T., Vedder, H., Schreiber, W., Fassbender, K. and Holsboer, F. (1992). Nocturnal plasma levels of cytokines in healthy men. Eur Arch Psychiatry Clin Neurosci 242, 53–56. Harris, E.K. and Boyd, J.C. (1990). On dividing reference data into subgroups to produce separate reference ranges. Clin Chem 36, 265–270. Harris, E. and Boyd, J.C. (1995). Statistical Bases of Reference Values in Laboratory Medicine. Marcel Dekker, New York. Irwig, L., Bossuyt, P., Glasziou, P., Gatsonis, C. and Lijmer, J. (2002). Designing studies to ensure that estimates of test accuracy are transferable. BMJ 324, 669–671. ISO (1993). Statistics – Vocabulary and symbols– Part 1: Probability and General Statistical Terms (3534-1). ISO, Geneva. ISO (1994). Accuracy (Trueness and Precision) of Measurement Methods and Results (ISO 5725) – Part 1: General Princiles and Definitions. ISO, Geneva. ISO (2000). Capability of Detection – Part 2: Methodology in the Linear Calibration case (11843-2). ISO, Geneva. Johnson, A.M., Ledue, T.B. and Collins, M.F. (2003). Commutability of the CRM 470 C-reactive protein value in the Dade Behring N High Sensitivity CRP assay. Clin Chem Lab Med 41, 177–182. Kricka, L.J. (1999). Human anti-animal antibody interferences in immunological assays. Clin Chem 45, 942–956. Ledue, T.B. and Rifai, N. (2003). Preanalytic and analytic sources of variations in C-reactive protein measurement: Implications for cardiovascular disease risk assessment. Clin Chem 49, 1258–1271. Lijmer, J.G., Mol, B.W., Heisterkamp, S., Bonsel, G.J., Prins, M.H., van der Meulen, J.H. and Bossuyt, P.M. (1999). Empirical evidence of design-related bias in studies of diagnostic tests. JAMA 282, 1061–1066. Linnet, K. (1993). Evaluation of regression procedures for methods comparison studies. Clin Chem 39, 424–432. Linnet, K.B.J. (2005). Selection and analytical evaluation of methods – with statistical techniques. In Tietz Textbook of Clinical Chemistry and Molecular Diagnostics (C.A. Burtis and E.R. Ashwood, eds), Elsevier Saunders, Philadelphia, pp. 352–407. Liu, M.Y., Xydakis, A.M., Hoogeveen, R.C., Jones, P.H., Smith, E.O., Nelson, K.W. and Ballantyne, C.M. (2005). Multiplexed analysis of biomarkers related to obesity and the metabolic syndrome
■
319
in human plasma, using the Luminex-100 system. Clin Chem 51, 1102–1109. Lolekha, P.H., Chittamma, A., Roberts, W.L., Sritara, P., Cheepudomwit, S. and Suriyawongpaisal, P. (2005). Comparative study of two automated high-sensitivity C-reactive protein methods in a large population. Clin Biochem 38, 31–35. Maisel, A.S., Krishnaswamy, P., Nowak, R.M., McCord, J., Hollander, J.E., Duc, P., Omland, T., Storrow, A.B., Abraham, W.T., Wu, A.H. et al. (2002). Rapid measurement of B-type natriuretic peptide in the emergency diagnosis of heart failure. N Engl J Med 347, 161–167. McCullough, P.A., Nowak, R.M., McCord, J., Hollander, J.E., Herrmann, H.C., Steg, P.G., Duc, P., Westheim, A., Omland, T., Knudsen, C.W. et al. (2002). B-type natriuretic peptide and clinical judgment in emergency diagnosis of heart failure: analysis from Breathing Not Properly (BNP) Multinational Study. Circulation 106, 416–422. McCullough, P.A., Duc, P., Omland, T., McCord, J., Nowak, R.M., Hollander, J.E., Herrmann, H.C., Steg, P.G., Westheim, A., Knudsen, C.W. et al. (2003). B-type natriuretic peptide and renal function in the diagnosis of heart failure: An analysis from the Breathing Not Properly Multinational Study. Am J Kidney Dis 41, 571–579. Meier-Ewert, H.K., Ridker, P.M., Rifai, N., Price, N., Dinges, D.F. and Mullington, J.M. (2001). Absence of diurnal variation of C-reactive protein concentrations in healthy human subjects. Clin Chem 47, 426–430. Miller, W.G. and Kaufman, H.W. (1993). College of American Pathologists Conference XXIII on matrix effects and accuracy assessment in clinical chemistry: Introduction. Arch Pathol Lab Med 117, 343–344. Moons, K.G., van Es, G.A., Deckers, J.W., Habbema, J.D. and Grobbee, D.E. (1997). Limitations of sensitivity, specificity, likelihood ratio, and Bayes’ theorem in assessing diagnostic probabilities: A clinical example. Epidemiology 8, 12–17. Moons, K.G.M., Biesheuvel, C.J. and Grobbee, D.E. (2004). Test research versus diagnostic research. Clin Chem 50, 473–475. NCCLS (1999). Evaluation of Precision Performance of Clinical Chemistry Devices: Approved Guideline. NCCLS,Wayne, PA. pp. 1–42 NCCLS (2002). Interference Testing in Clinical Chemistry: Approved Guideline. NCCLS,Wayne, PA. NCCLS (2003). Evaluation of the Linearity of Quantitative Measurement Procedures: A Statistical Approach; Approved Guideline. NCCLS, Wayne, PA. Obuchowski, N.A., Lieber, M.L. and Wians, F.H., Jr. (2004). ROC curves in clinical chemistry: Uses, misuses, and possible solutions. Clin Chem 50, 1118–1125. Price, C. and Newman, D.J. (1996). Principles and Practice of Immunoassays. Stockton Press, New York. Rifai, N. and Ridker, P.M. (2003). Population distributions of C-reactive protein in apparently healthy men and women in the United States: Implication for clinical interpretation. Clin Chem 49, 666–669. Rifai, N., Gillette, M.A. and Carr, S.A. (2006). Protein biomarker discovery and validation: The long and uncertain path of clinical utility. Nature Biotech 24, 971–983. Rutjes, A.W., Reitsma, J.B., Di Nisio, M., Smidt, N., van Rijn, J.C. and Bossuyt, P.M. (2006). Evidence of bias and variation in diagnostic accuracy studies. CMAJ 174, 469–476. Sackett, D.L. and Haynes, R.B. (2002). The architecture of diagnostic research. BMJ 324, 539–541.
320
CHAPTER 26
■
Validation of Candidate Protein Biomarkers
Solberg, H.E. (1983). The theory of reference values Part 5. Statistical treatment of collected reference values. Determination of reference limits. J Clin Chem Clin Biochem 21, 749–760. Solberg, H.E. (1987). International Federation of Clinical Chemistry (IFCC), Scientific Committee, Clinical Section, Expert Panel on Theory of Reference Values, and International Committee for Standardization in Haematology (ICSH), Standing Committee on Reference Values. Approved Recommendation (1986) on the theory of reference values. Part 1. The concept of reference values. J Clin Chem Clin Biochem 25, 337–342. Solberg, H.E. and PetitClerc, C. (1988). International Federation of Clinical Chemistry (IFCC), Scientific Committee, Clinical Section, Expert Panel on Theory of Reference Values. Approved Recommendation (1988) on the theory of reference values. Part 3. Preparation of individuals and collection of specimens for the production of reference values. J Clin Chem Clin Biochem 26, 593–598.
RECOMMENDED RESOURCES http://www.fda.gov/oc/mdufma/coversheet.html http://www.mhlw.go.pj/english/org/policy/p13-4.html http://www.phppo.cdc.gov/clia/regs
Tan, M.H., Wilmshurst, E.G., Gleason, R.E. and Soeldner, J.S. (1973). Effect of posture on serum lipids. N Engl J Med 289, 416–419. Vasan, R.S., Benjamin, E.J., Larson, M.G., Leip, E.P., Wang, T.J., Wilson, P.W. and Levy, D. (2002). Plasma natriuretic peptides for community screening for left ventricular hypertrophy and systolic dysfunction: The Framingham heart study. JAMA 288, 1252–1259. Vitzthum, F., Behrens, F., Anderson, N.L. and Shaw, J.H. (2005). Proteomics: From basic research to diagnostic application. A review of requirements and needs. J Proteome Res 4, 1086–1097. Wild, D. (2005). The Immunoassay Handbook. Elsevier,Amsterdam. Zweig, M.H. and Campbell, G. (1993). Receiver-operating characteristic (ROC) plots: A fundamental evaluation tool in clinical medicine. Clin Chem 39, 561–577.
CHAPTER
27 Pharmacogenetics and Pharmacogenomics Iris Grossman and David B. Goldstein
INTRODUCTION Western medicine has been extremely successful in producing mass pharmaceutical solutions by following the “one-drug-fitsall” paradigm. Indeed, the majority of individuals plotted on a Gaussian distribution show reasonable efficacy and safety profiles, justifying the overall validity of this approach. However, alarming reports are amassing on the frequency and gravity of adverse drug responses, as well as inadequate efficacy, clearly signaling the need to shift from “blockbuster” to more individualized treatment models. The objective of pharmacogenetic research is to identify a genetic marker, or a set of genetic markers, that serve as prognostics for the fashion in which a given person is likely to respond to a given medicine. The appropriate formulation, dose, and regimen are presumed to be predicted, at least partially, by genetic determinants, thus serving as diagnostics that can be tested prior to initiation of drug treatment. In this fashion, treatment would be allocated only, or mainly, to individuals who are expected to benefit from it, thus reducing dramatically the suffering and costs associated with adverse drug reactions (ADRs) and inadequate efficacy. As summarized by Wilkinson, “most major drugs are effective in only 25–60% of patients, and more than two million cases of ADRs occur annually in the US, including 100,000 deaths” (Wilkinson, 2005). It should be noted that two terms exist in the scientific literature, pharmacogenetics and pharmacogenomics, each with its own
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
connotation and a range of alternative and poorly defined distinctions. Here, we use the term pharmacogenetics (PGx), which has historical priority, according to its broadest meaning – relating heritable variation to inter-individual variation in drug response (Goldstein et al., 2003). It should be mentioned that the term pharmacogenomics is often used to describe the study of differential gene-expression profiles in tissues of interest and their relation to drug-response phenotypes. While such expression signatures may be affected by underlying genetic variation, non-inherited environmental determinants play an important role as well. We therefore focus here solely on inherited sequence variation (i.e., single nucleotide polymorphisms, SNPs), insertions–deletions, copy number variations, microsatellites, Alu repeats, and so on) and its association with treatment response end-points. A comprehensive review of geneexpression signatures, their association with response to chemotherapy, and the search for underlying genetic determinants is provided in Nevins and Potti (2007). The complexity of genetic relationships and its interplay with drug-response phenotypes has been recently acknowledged by the US Food and Drug Administration (FDA) in a rather crude and controversial decision to stratify patients’ populations by race in the case of severe heart failure adjunct treatment with BiDil (Meadows, 2005). Although little beneficial effect has been registered in the general population, BiDil treatment showed a dramatic reduction in death and hospitalization in African
Copyright © 2009, Elsevier Inc. All rights reserved. 321
322
CHAPTER 27
■
Pharmacogenetics and Pharmacogenomics
Americans. Indisputably, this stratification criterion is simplistic, biased, and ethically provocative, but it illustrates the concept of inherent differences between individuals in the general population and serves as a bridging step from indiscriminant allocation of therapeutics to genetic-based stratification into risk/benefit subgroups. This example also illustrates the fact that the popular description of PGx as aspiring to “individually tailor” therapy is an exaggeration. The field is not centered around each individual specifically, but rather seeks to increase homogeneity among patients to an acceptable efficacy and safety level. In other words, PGx aspires to generate an individualized guide to proper therapeutic option among predictable segments of the patient population (Roses, 2004). The discipline of PGx is closely tied to human genetics and genomics as a whole, but focuses on specific phenotypes of interest. While disease predisposition research in the realm of non-Mendelian traits is faced with extremely complex etiology, dictated by interplay between various pathways in the human body, it is assumed that pharmacogenetic phenotypes are largely governed by a limited number of molecules (an oligogenic structure), most of which interact with the drug as pharmacokinetic or pharmacodynamic entities. Although genetic factors associated with underlying disease etiology may impact treatment response features, these effects are hypothesized to have a more modest effects in comparison to genetic variations in molecules that come in physical contact with the drug upon its administration. Moreover, the phenotypic changes that are induced by drugs are clearly and reproducibly caused by treatment exposure, which renders them less difficult to relate to genotype in comparison to studies of other complex traits. Thus, PGx research is equipped with the enormous advantage of being more TABLE 27.1
readily and fruitfully realized in clinical practice. In addition, the (potential) direct application of pharmacogenetic findings in the daily management of disease burden prompts this field to be attractive, not only to academic circles but also to the pharmaceutical industry, patients, health care providers, regulators, and health systems as a whole. Any treatment for which inter-individual variability is significantly larger than intra-individual diversity indicates an underlying genetic effect associated with response variables. In order to maximize the potential benefit from PGx research, it is helpful to focus and prioritize the analyzed therapies as follows (Table 27.1): ●
●
Chronic disease states: in these cases disease burden on both the patients and health systems translates into compromised patients’ quality of life, continuous suffering, low medication adherence rates, polypharmacy, and ultimately considerable care management costs. Therefore, management of chronic diseases comprises an attractive aim for optimization of available treatments and development of new therapeutic solutions (e.g., hypertension, diabetes, asthma, and so on). Narrow therapeutic index drugs (i.e., those with little difference between toxic and therapeutic doses) require close and constant dosage adjustment procedures to achieve therapeutic levels and minimize toxicity (usually by repeated measurements of therapeutic drug monitoring (TDM) protocols). Utilization of predictive genetic tests that may identify individuals at risk – either prior to treatment initiation or during the initial dose adjustment period – will save patients’ lives, while diminishing substantially the costs of care management (e.g., warfarin, 6-mercaptopurine, and so on).
Circumstances defining high priority targets for pharmacogenetic investigation
Circumstances The underlying condition being treated is chronic, incurable, and/or recurrent Current clinical practice of the underlying condition follows a “trial and error” paradigm
Rationale
Epilepsy ( Tate et al., 2005) Multiple sclerosis (Grossman et al., 2007a) Arrest the progression of the underlying disease, reduce patients’ suffering, and decrease disease management burden on the health system
Asthma ( Weiss et al., 2006) HIV ( Telenti and Goldstein, 2006) Hypertension (Arnett et al., 2006) Warfarin (Aquilante et al., 2006) TPMP (van den Akker-van Marle et al., 2006)
The studied drug has a narrow therapeutic index The studied drug is associated with severe, lifethreatening side-effects
Examples
Save patients’ lives and decrease hospitalization costs associated with ADRs
Herceptin (Krejsa et al., 2006)
The predictive value and clinical utility of the studied drug are high
Develop accurate and useful clinical diagnostic tools
Abacavir ( Warren et al., 2006) TPMT (van den Akker-van Marle et al., 2006)
Rescue drugs post Phase II development
Market drugs to patient subpopulations that are predicted to benefit from the treatment, protecting potential non- or adverse responders
GW320659 (Spraggs et al., 2005)
Pharmacogenetic Studies: From Concept to Practice
●
●
Disease states for which current practice follows a “trialand-error” paradigm due to absence of predictive markers of efficacious and safe treatment regimens often require long adjustment periods, thus resulting in months or even years of incrementing doses and drug switching until a successful course of treatment is identified. Predictive genetic tests may assist in reducing this elongated adjustment period, resulting in increased compliance rates, and reduction of side effects, disease management burden and costs. As an integral part of the drug research and development process, rescuing formulations efficacious in subgroups of the patient populations, but failing to exhibit overall benefit in the all-inclusive cohort.
Noteworthy is the fact that most of the current examples in the PGx literature, including those discussed here, investigate marketed drugs retrospectively, when the drug and its clinical responses are available to be studied by academic investigators. A comprehensive review of PGx research along the drug development process with examples from discovery through phase III trials is presented elsewhere (see Chapter 29 and Roses et al., 2006).
The Ultimate Goal: Definition of the Key Research Question A PGx study aims at identifying genetic determinants associated with drug–response features. The latter focus mainly on treatment efficacy, dosing, and safety. The definition of each of these elements is not trivial and is tightly dependent upon the specific parameters of a given dataset: the underlying disease and relevant clinical follow-up measurements dictate the available tests and clinical information that is not only relevant, but also obtainable, especially with respect to retrospective data acquisition. For example, in an investigation of the response to multiple sclerosis, treatment efficacy may be determined by clinical observations evaluated by the Extended Disability Status Scale (Kurtzke, 1983) (EDSS), reduction in relapse rate, severity of relapses, and/or MRI findings estimating T1-weighted hypointense lesions and central nervous system (CNS) atrophy measures (Grossman et al., 2007a). Even a phenotype as “simple” as dosing may be regarded by various experts in a host of different ways. For instance, dosing phenotypes relevant to treatment with anti-psychotics for the treatment of schizophrenia are often assessed by “maintenance dose,” which may reflect halving the drug dose that was proven effective during the acute phase (Gaszner and Makkos, 2004) or a titration regimen starting from minimum doses determined experimentally up to a recommended target dose on average (Perry and Hradek, 2004). When considering phenotypes of treatment safety recombinant protein therapeutics (including supplements to, or blockers of, endogenous proteins, immunomodulatory molecules and small, highly selective, targeted binding proteins), as opposed to
323
small-molecule drugs, introduce an additional dimension requiring consideration of their potential immunogenicity (Krejsa et al., 2006). Whatever the case may be, the phenotypes to be tested must be determined prior and blinded to acquisition of results and are required to be precise, consistent, obtainable in the current investigation, reproducible and clinically relevant. Tailor-Design of Your Study by Available Resources and Specific Goals Researching the genetic determinants underlying a phenotype of interest is a task largely pre-defined by the specific characteristics of the studied phenomena. The investigator has no control over attributes such as phenotype penetrance; frequency of the (unknown) causative polymorphism(s); number of involved and modifier genes; genotype relative risk; phenocopy and genocopy effects; linkage disequilibrium extent between the typed markers and the causative polymorphism(s); and so on. Notwithstanding, an array of tools are available at the disposal of the pharmacogenetic researcher aimed to better the chances of identifying genotype–phenotype associations by sophisticated tailor fitting each individual project (Figure 27.1): ●
PHARMACOGENETIC STUDIES: FROM CONCEPT TO PRACTICE
■
●
● ●
collection of a large enough sample size as reflected by power estimations (Chapman et al., 2003; Kelly et al., 2005; Singer et al., 2005); distinct and precise, yet simple and broad, phenotype definitions; skilled marker selection (as detailed later); and appropriate choice of study design.
Classic Case–Control Study Design The classic most “simplistic” study design, which is still considered the basic and fundamental association test pursued in genetic literature, is the case-control study. Originally, this analysis was used for testing significant differences in exposure without attempting to quantify the risk associated with exposure: Statistically, do more lung cancer patients have a history of smoking than controls? Rather than by how many times does smoking increases the risk of lung cancer? Case-control studies contain only group-level information and may determine the association between potential cause and effect on an individual basis. In pharmacogenetic investigations, cases and controls do not refer to health states or existence of drug exposure (affected versus unaffected, treated versus untreated), but rather to subpopulations within a source population of affected and treated individuals: positive responders versus non-responders, or patients manifesting ADRs versus safely treated individuals. Investigation of quantitative traits (as opposed to qualitative traits) is traditionally pursued by linear regression methods assuming that the Y-axis variable is a continuous measurement (e.g., optimized drug doses prescribed to schizophrenia patients [Grossman et al., 2007b]). In order to avoid bias due to confounding effects inherent in the study populations while employing qualitative, as well as quantitative, phenotype definitions, and so as to allow follow-up replication analysis within the total or the remaining trialed population (Figure 27.2), it is
324
CHAPTER 27
■
Pharmacogenetics and Pharmacogenomics
Study population Study population: Affected
Study population: Healthy Clinical trial blinded to genotype
Genotype
Genotype AA
Study population: All
AA Aa
Aa aa
aa Placebo
Drug dose 1
Drug dose 2
Clinical challenge
aa Aa AA
Responder
Nonresponder
Select equisize subgroups defined by “aa” carriers
Continuous phenotype
Training set all individuals in the current study
Statistical association analysis for unrelated/related individuals
Replicate in an independent population
Validation Training set 5% extremities of phenotype distributions in the current study
Replicate in 10/25/45% remaining individuals
Figure 27.1 Pharmacogenetic study design alternatives in a simplified case, genotyping a single nucleotide polymorphism in a study population. In research settings pharmacogenetic study populations may either be healthy or affected subjects. A priori genotyping may be used as inclusion criteria and in several models may be used to reduce study population, when equisize groups of genotype carriers (aa, aA, and AA) are randomized to treatment and subsequent analysis. In clinical trial settings, individuals are randomized to treatment groups (active drug, parallel dose comparison arms, gold standard comparator drug, and/or placebo) blinded to genotype. Only once trial is completed and efficacy and safety end-points are clinically analyzed does statistical analysis take into account genetic underlying effects. Similar approach can be utilized in adaptive trial design for interim analysis. It is expected that some genetic associations will reach statistical significance at random and thus it is vital to validate any association results in either the replication set within the current cohort, or in an independent population.
becoming acceptable to employ case–control designs in which cases are chosen as patients exhibiting the lowest 5% of a phenotype distribution, while controls represent the 95th percentile of this distribution. Contrasting these two subgroups in the initial analysis enables focusing on the most extreme representations of the phenotype of interest with a lenient significance level, accentuating the roles that causative factors (genetic and environmental) play in exhibition of these phenotype states, and expecting most of the findings to reflect false-positive associations. In the second stage, designed to replicate and validate the obtained results, individuals represented by the remainder of the distribution space are split into cases and controls. Analysis can be performed at this level by using a stringent Bonferroni-corrected significance level as a “replication-based analysis” or by combining
test statistics from both stages and applying significance level corrected only for the number of markers tested, titled “jointanalysis” (Skol et al., 2006). The above-described alternatives of one-stage versus two-stage association study designs are summarized in Figure 27.2. Familial Study Design Familial and twin genetic association studies allow refined elucidation of genetic associations while inherently correcting for environmental effects and risk factors (Endrenyi et al., 1976; O’Reilly et al., 1994). The task of collecting appropriate related cohorts suffering from the same conditions and exposed to the same pharmacological agents is often not trivial (although has been demonstrated to be helpful in the case of nicotine
Marker Selection – Strategy and Application
Phenotype Continuous
Dichotomous
H 5
ADRs
No ADRs
Cases
Controls
95
%
e.g., biomarker blood levels 5%
5%
Stage I
Cases
Controls Stage II Replicationbased analysis
Joint analysis Combining test statistics from both stages at significance level of α /M
5
Cases
50
95%
Controls
Significance level of α/(πmarkers M )
Figure 27.2 One-stage versus two-stage case–control genetic association analyses. The phenotype of interest in pharmacogenetic studies can be categorized as either continuous or dichotomous. When occurrence of adverse drug reactions is analyzed, the natural statistical model calls for a case-control design. However, efficacy end-points tend to display a continuous distribution, such as measurements of biomarker blood levels. A two-stage design would split the study population into (Stage I) a case–control analysis contrasting the 5% extremities of the Normalized distribution; thereafter (Stage II) would employ a replication-based analysis of the remaining 90% of the distribution with adjusted requirements for significance level. A joint analysis will then ensue, combining test statistics from both stages at a corrected significance level (Skol et al., 2006). ADR, adverse drug reaction; , study false-positive rate (type I error rate); M, number of genotyped markers; markers, the proportion of markers found to be associated in Stage I and followed up in Stage II.
metabolism [Swan et al., 2004] and alcohol exposure [Heath and Martin, 1994]). However, implementation of healthy volunteer challenge studies (see later) within related cohorts may be highly effective (Kendler et al., 1988). Healthy Volunteer Challenge Design Pharmacogenetic investigations, as opposed to disease predisposition genetic analyses, bear the extraordinary advantage of inducing the phenotype of interest in any given individual – including in healthy volunteers. In other words, in order to study the effects of drug intake in a genetic context, one can either investigate affected individuals prescribed the drug in question
■
325
for the therapeutic purpose of controlling their condition (the traditional approach) or administer the drug to healthy consenting individuals specifically for the purpose of PGx research. This unique option allows not only the rapid recruitment of study population to be screened for genetic variation but also enables specific tailoring of the study design to address PGx questions, rather than clinical trial end-points. In practice, researchers assess the impact of pre-selected polymorphisms on the PGx phenotypes (e.g., efficacy, safety, dosing, and pharmacokinetic properties) of the studied drug, by prospective clinical analysis of healthy volunteers chosen conditioned on their genotype profile. In this manner, Kirchheiner et al. (2004) identified a total of 25 healthy males carrying one of three genotypes of interest (CYP2D6 ultra, extensive, and poor metabolizers) out of more than 1000 healthy volunteers in a PGx study of Mitrazapine and clinically challenged this miniature group of interest for clinical measurements. This cohort size matched power estimates for detecting at least a 1.5-fold effect on drug clearance differences between subgroups. Similar designs have been published for a variety of pharmaceutical applications and bear a great potential for simplification of PGx research in agents with favorable safety profile. Placebo and Multiple Dosing Cohort Comparisons PGx association studies in general, and those of medications for psychiatric and immunological diseases in particular, are characterized by a poor signal-to-noise ratio: approximately one-third of the patients enrolled in efficacy trials may respond to placebo treatment. The placebo “response” in randomized clinical trials includes such statistical artifacts as regression to the mean, drift in measurement of the response over time, and bias of expectations by both patients and evaluators, as well as real effects such as spontaneous recovery, a tendency to seek treatment outside the study, and the response to additional attention and concern involved with participating in clinical trials (Singer et al., 2005). It has been shown that placebo “response” strongly affects statistical power of association studies and that employing a placebotreated control group in the PGx analysis enables elucidation of specific drug-induced genetic effects, differentiating these from confounding effects related to underlying disease progression and severity (Grossman et al., 2007a). Similarly, employment of trial arms administering different drug doses may shed light on the genetic underlying mechanisms and their magnitude (Risner et al., 2006).
MARKER SELECTION – STRATEGY AND APPLICATION Candidate Genes Strategy Versus Whole-Genome Scans A comprehensive search for genetic influences on drug response would involve examining all genetic differences in a large number of affected individuals and controls. This will eventually
326
CHAPTER 27
■
Pharmacogenetics and Pharmacogenomics
(and probably not too far ahead) be achievable by complete genome re-sequencing. However, current practice is limited to a systematical test of common genetic variants (those with 2–5% minor allele frequency [MAF] and higher). These variants explain most of the genetic diversity in human populations and may be sampled taking advantage of the genetic property termed linkage disequilibrium (LD) (see also Chapter 2). LD reflects shared ancestry of contemporary chromosomes, measuring the degree of non-random co-occurrence of two (or more) polymorphic sites along a stretch of genomic sequence. Polymorphisms showing a high degree of LD may be considered redundant, eliminating the necessity to type all SNPs in this correlated group. It suffices to genotype a representative SNP and virtually represent the entire variation captured by the whole group, given the LD threshold imposed on the model. Traditionally, studies of common traits have been studied by two different strategies: family-based linkage studies across the entire genome and population-based association studies of individual candidate genes (usually only a handful). Although there have been notable successes, progress has been slow due to the inherent limitations of both methods: linkage analysis has low power except when a single locus explains a substantial fraction of disease, and association studies of one or a few candidate genes examine only a small fraction of the “universe” of sequence variation in each patient. Progress in technology and the development of sophisticated statistical tools have paved the way toward the application of Whole Genome Association Studies (WGAS) in a population-based cohort (see Chapter 8). This progress makes it possible to study the full scope of possible genetic determinants involved with phenotypes of drug response in a hypothesis-free fashion (i.e., unbiased approach). However, the quest for the Holy Grail, that is, the identification of causal polymorphisms underlining the phenotype of interest, remains a key unanswered question even when interrogating the entire genome. While many approaches have been suggested in the literature as to how to decipher between true associations and falsepositive results, no consensus is likely to be established in the next few years (for recommendations on replication design in large-scale association studies see the NCI-NHGRI Working Group on Replication in Association Studies [2007]). Expert genetic, biological, and pharmaceutical skills will be needed to establish a prioritization scheme by which result interpretation will reflect not only statistical plausibility but also scientific rationale. Independent comparative studies will eventually give rise to the most successful strategies that should be pursued. Hence, the strategy of choice for years to come will progressively rely on a single universal tagging set that will be used in all whole-genome association studies. This would not only ensure comparability for replication efforts within complex traits but also allow direct comparison of the role of the same variants as risk factors for different conditions. Led by this vision, The National Center for Biotechnology Information (NCBI) is currently managing the submission, storage, and access for clinical phenotype measures and associated whole-genome genotype data for several different programs, making this data as widely
available to the research community as possible while protecting the privacy of the participants as defined by consent agreements for individual projects. Application of whole-genome scans for PGx phenotypes is currently being undertaken by various academic and pharmaceutical researchers. One pioneering example depicts a WGAS investigating drug-induced liver injury ADRs and represents the incorporation of PGx in pharmaceutical R&D (Kindmark et al., 2007). Further examples are discussed in Chapter 28. Pharmacokinetics- Versus Pharmacodynamics-Based Gene Selection Once administered, a drug goes through a series of interactions in the body that are responsible not only for the therapeutic effect conveyed by this agent, but also for side effects and bystander alterations to other compounds processed in conjunction. As a whole, all interactions related to modifications acted upon the drug by the host body are termed “drug pharmacokinetics” and include all aspects of drug absorption, distribution, metabolism, and elimination (ADME). The most prominently studied molecule group within the sphere of pharmacokinetics has indisputably been drug-metabolizing enzymes (DMEs). These enzymes exist in every eukaryotic cell and in most, if not all, prokaryotes and are thought to have evolved in humans to neutralize xenotoxins and/or to control concentrations of signaling molecules in endogenous pathways (Nebert and Dieter, 2000). The fact that modern drugs have been introduced to the in vivo environment very recently in evolution uniquely excludes selection forces from having impacted the functional genetic properties of DMEs, thus potentially permitting whopping effects on drug-response phenotypes (Nebert and Dieter, 2000). In the early years of PGx research and up until very recently, genes encoding pharmacokinetic properties of drugs have been the focus of research (for a comprehensive table of drug–drug interactions based on cytochrome P450 enzymatic activity see [Flockhart, 2007]). This has been due to the scientific expectation of marked clinical effects underlined by diminished or absent capacity of the body to obliterate drugs or transform prodrugs into active compounds, but also due to the pragmatic constraint posed by considerable gaps in our understanding of, and technological methods for, investigating genes encoding drug pharmacodynamic features. Pharmacodynamics relates to any modifications acted upon the host body by the drug, that is, drug targets (receptors or enzymes) and their downstream signal pathways. With the advent of the Human Genome Project, the HapMap project (see below) and incredible progress in genotyping techniques and costs, it is currently feasible to interrogate known functional, as well functionally as-yet-unknown, genetic variants in pharmacodynamic candidate genes. It is now becoming increasingly evident that polymorphisms within genes involved in the pharmacodynamic properties of drugs may in fact be more important and clinically relevant than variants of ADME genes for non-Mendelian drug-response phenotypes. Examples supporting this notion include reports on the relative contribution of variants in the VKORC1 and CYP2C9 genes to warfarin
Marker Selection – Strategy and Application
response (Aquilante et al., 2006); SCN1A and CYP2C9 to phenytoin response (Tate et al., 2005); and 5-HTR2A and CYP2D6 to paroxetine response (Murphy et al., 2003). Pathway-Based Analyses and Search for Epistatic Associations Advances in haplotype tagging strategies have made it possible to represent variation in entire pathways relevant to drug response economically. The focus in pathway-based PGx investigation is further facilitated by the accumulation of knowledge regarding pathways relevant to drug response (e.g. PharmGKB Pathways). Although promising, pathway-based analyses face considerable challenges. Most fundamentally, the huge number of tests conducted when analyzing large SNP datasets, multiple phenotype definitions, and application of a diverse spectrum of parametric and non-parametric statistical methodologies pose considerable barriers on power, statistical design, and interpretation. Increasing the population sample size, applying permutation-based multiple correction methods, and seeking independent replication are obvious solutions dealing with these difficulties. A further complication is that all components of all pathways that a drug acts on are rarely understood completely (Need et al., 2005). As PGx investigations rapidly progress toward a pathwaycentered approach, considering dozens and hundreds of genes in each instance, or assigning prior probability scores of wholegenome scans founded on the biological processes involved with a given drug’s PGx, it is progressively evident that research should embrace methods that take into account possible interactions between genetic variants along a pathway of interest. It is very likely that, for instance, genetic-based conformational changes brought forth in one molecule along a drug-related pathway will require complementary structural, quantitative, or functional changes in a downstream (or upstream) molecule to produce a detectable sum total effect on phenotype measurements. This genetic interaction is termed “epistasis” and it is yet unclear how best, computationally and statistically, to surmount the complexity arising from the combinatorial magnitude of all the possible interactions involved within a pathway. Moreover, the high dimensionality of the resulting analysis overwhelms traditional analysis methods, a problem exponentially prohibitive for large candidate gene studies and whole-genome investigations. The Multifactor Dimensionality Reduction (MDR) approach (Wilke et al., 2005) has been suggested as one potential solution for this problem, although sophisticated analytical approaches capable of modeling high order interactions and sizable datasets will have to be developed in order to handle genome-wide and large-scale candidate gene studies. Gene-expression and proteomic profiling may in the future provide additional knowledge and complementary diagnostics associated with pathways relevant to drug response, although at this time it is difficult to differentiate incidental differences in expression levels from those that are causally relevant to drug response (Need et al., 2005). In conclusion, thanks to technological and computational advances, it is no longer justifiable to look at the effects of single genes in isolation, but rather
■
327
incorporate the milieu of potential effects more comprehensively describing the sum total of a studied phenotype. HapMap – A Public Resource for Marker Selection The International HapMap Project was launched in October 2002 to create a public, genome-wide database of common human sequence variation, providing information needed as a guide to genetic studies of clinical phenotypes (The International HapMap Project, 2003). Four different ethnic populations were systematically genotyped for each SNP variant as follows: (i) 90 individuals (30 parent–offspring trios) from Yoruba, Ibadan, Nigeria; (ii) 90 individuals (30 trios) from Utah, USA (the Centre d’Etude du Polymorphisme Humain collection); (iii) 45 non-related Han Chinese from Beijing, China; and (iv) 44 nonrelated Japanese from Tokyo, Japan. The HapMap project set out to describe patterns of association between typed polymorphisms (LD) in the above-described four population samples in order to identify a minimum set of SNPs (“tagging SNPs”) sufficient to represent all common variants in the human genome. A publicly available software for tagging has been incorporated into the HapMap website (description and a more detailed version are available in de Bakker et al. [2005]), enabling any given user to select a desirable set of tag SNPs based on a variety of criteria (such as population ethnicity, force inclusion of SNPs with known functional effects or which have previously been described in the literature, and so on). The chief criterion for tagging capacity employed is the r2 threshold. This is a study-independent measure of SNP utility that has become a leading standard for evaluating performance of marker sets. It represents the correlation coefficient between any observed marker and a putative causal allele, reflecting the expected drop in non-centrality of an association test statistic under specified conditions (Pe’er et al., 2006). It is customary to enforce a minimum pair-wise r2 threshold of 0.7–0.8. Analyses increasingly attest to the remarkable performance of tagging (Conrad et al., 2006; de Bakker et al., 2006), indicating that the HapMap samples provide an appropriate resource for selecting globally useful tags (although African populations remain partially unrepresented by tagging approaches due to their high diversity). Thanks to the fact that all phases of the HapMap database have been publicly available ever since the projects’ inception, both commercial companies and academic laboratories routinely implement HapMap data in their drug research and development programs. In the realm of pharmacogenetics, for instance, elucidation of differential response to, and dosing of, warfarin anti-coagulation treatment as governed by VKORC1 was performed using the HapMap data (Rieder et al., 2005). Another example, specifically designed to examine the capacity of HapMap tools to discover and validate PGx phenotypes, was recently published (Jones et al., 2007). In this study, HapMap SNPs in the thiopurine methyltransferase (TPMT) gene region (tags) were tested for association with cell line absolute levels of TPMT activity. It was expected that a variant of biochemically and clinically validated relevance to thiopurine plasma levels and induced toxicity will rank as most significantly associated in all analyses. Indeed, this SNP and others in high LD with it tested positively, out of a total of 66 HapMap
328
CHAPTER 27
■
Pharmacogenetics and Pharmacogenomics
SNPs in the TPMT gene region. However, an agnostic genomewide approach to assess genomic predictors of TPMT phenotype in this dataset ranked TPMT haplotypes at 97th place out of the 17,542 genes tested. The latter result demonstrates that wholegenome tagging approaches are useful screening tools for the highest few percent of associated genes; nonetheless, ultimately biology, biochemistry, and clinical expertise are required to decipher true associations from false-positives. Validation of Association Results Any results detected in “first stage” genetic analyses require further validation. This is usually accomplished by two lines of investigation: functional analysis and replication studies. Functional studies are aimed at revealing direct causality between the reported genetic variant and measurable functional modifications elicited in the encoded molecule by employing animal models (e.g., knockout mice) or in vitro investigations. Replication studies are performed in independent patient populations mimicking as closely as possible the study design settings and testing only the top priority fraction of SNP associations detected in the first stage. Replication analysis serves not only for confirmation of true association between a genetic marker and a phenotype of interest, but it is also a means for better assessing the true magnitude of the detected effect, as first publications tend to report inflated effects due to the phenomenon termed “the winner’s curse” (Ioannidis et al., 2001). Prospective pharmacogenetic analyses compromise an important step toward the validation of genetic markers as predictors of clinical management in “real-life” settings (Grossman, 2007). Although few such studies have been conducted thus far, it is becoming generally accepted that prospective analyses comprise necessary and integral evidence of the potential utility of genetic tests in routine health care management. Initiatives already prospectively monitoring patients for PGx outcomes include several studies aimed at analysis of cost-effectiveness of HLA-B*5701 genotyping in abacavir-treated HIV-positive patients prone to suffer from hypersensitivity reaction (Hughes et al., 2004, 2007; Rauch et al., 2006; Zucman et al., 2007); a genotype-guided clinical cancer trial for management of rectal cancer by irinotecan therapy based on analysis of a polymorphic tandem repeat in the 5-untranslated region of the TYMS enhancer region, resulting in 2–9 copies of a 28-bp repeated sequence (McLeod et al., 2005); and genotype-guided dosing analyses of warfarin therapy (Hillman et al., 2005).
FROM BENCH TO BEDSIDE: INTEGRATION OF PHARMACOGENETIC TESTING INTO CLINICAL PRACTICE Analytic Assessment of Pharmacogenetic Tests and Their Utility in Clinical Practice Demonstration of unambiguous association between genotype(s) and drug-response features in controlled retrospective datasets,
albeit promising, does not guarantee the usefulness of routine genetic testing for the clinical decision-making process in the daily clinical environment. A whole host of factors need to be further assessed in order to predict and determine the costeffectiveness and practical utility of a pharmacogenetic diagnostic test. This evaluation should follow recommendations such as those devised by the Centers for Disease Control (CDC, 2006), including four components: Analytic Validity; Clinical Validity; Clinical Utility; and Ethical, Legal, and Social implications. Each of these components focuses on a different aspect of test performance and applicability in real-life settings, starting at the laboratory performing the test, through clinical studies proving its benefits, to more subjective evaluation of managed care burden associated with introduction of the test to routine practice and reimbursement frameworks. Each of these components can be empirically measured to reflect high validity and clear utility standards: 1. Analytic validity of a genetic test defines its ability to accurately and reliably measure the genotype of interest. Empirically, analytic validity includes analytic sensitivity (or the analytic detection rate, number of polymorphism carriers detected by the test divided by the total number of polymorphism carriers in the population); analytic specificity (number of normal genotype carriers detected by the test divided by the total number of individuals who do not carry the polymorphism); laboratory quality control (including measurements such as inter- and intra-assay variability, repeatability, and so on); and assay robustness (measuring how resistant the assay is to changes in pre-analytic and analytic variables, such as inter-laboratory variability, DNA extraction methodology, and so on). 2. Clinical validity of a genetic test defines its ability to detect or predict the associated phenotype. Empirically, clinical validity includes clinical sensitivity (or the clinical detection rate, the proportion of individuals who have a well-defined clinical phenotype and whose test values are positive); clinical specificity (the proportion of individuals who do not have the well-defined clinical phenotype and whose test results are negative); prevalence of the specific phenotype (the proportion of individuals in the selected setting who have, or who will develop, the phenotype); positive predictive value (the proportion of individuals with a positive test result who have, or will develop, an unwanted response when the drug is administered, PPV); negative predictive value (the proportion of individuals with a negative test result who will not have, or will not develop, an unwanted response when the drug is administered, NPV); penetrance (the frequency, under given environmental conditions, with which a specific phenotype is expressed by those individuals with a positive test result); and modifiers (genetic or environmental). 3. Clinical utility of a genetic test defines the elements that need to be considered when evaluating the risks and benefits
Examples of PGx Tests: Promising New Developments and Marketed Products
associated with its introduction into routine practice. Empirically, clinical utility includes administration settings recommendations (such as optimal time line for administration of the test, either prior to treatment initiation, during the first phase of dosing adjustments, or else); availability and effectiveness of counter-interventions; quality assurance (procedures in place for controlling pre-analytic, analytic, and post-analytic factors that could influence the risks and benefits of testing); pilot trials (assessing the performance of testing under real-world conditions, where inclusion criteria allow for individuals to experience co-morbidities, take concomitant medications, and so on); health risks (adverse consequences of testing or interventions in individuals with either positive or negative test results); economic evaluation (financial costs and benefits of testing); facilities (assess the capacity of existing resources to manage all aspects of the service); education (quality and availability of informational materials and expertise for all aspects of a screening service); and monitoring (assess a program’s ability to maintain surveillance over its activities and make adjustments). 4. Ethical, legal, and social implications surrounding a genetic test represent the safeguards and impediments that should be considered in the context of all the other components. FDA Regulation Policy and Its Impact on PGx Testing Application in Real-Life Health Care The FDA has recognized the importance of PGx and encourages its use in drug development. Recognizing that the clinical component of the overall cost of successful new drug development is ~58% of total costs (DiMasi et al., 2003) and that ~50% of Phase III trials fail, the FDA is promoting identification of PGx biomarkers at early drug development stages that would predict which drugs are likely to either fail or succeed in Phase III trials. This approach is reflected in the FDA’s 2004 white paper (FDA, 2004a), which identifies PGx as a key opportunity for the Critical Path. In addition, the FDA believes that there is value in applying long-established PGx testing to older, marketed drugs in the post-marketing period to improve their risk/benefit ratio by optimizing or individualizing dosing (Lesko and Woodcock, 2004). While multiple enthusiastic statements supporting the incorporation of PGx analysis to both new drug development and monitoring of marketed drugs have been made by FDA officials (Frueh et al., 2005; Lesko and Woodcock, 2002, 2004), there has not been a proactive regulatory action enforcing genomic data submission. In fact, ever since the publication of Guidance for Industry: Pharmacogenomic Data Submissions in 2004 (FDA, 2004b), relatively few drug labels have been updated to include recommendation for PGx testing (Table 27.2). The current policy refrains from enforcement of submissions, while promoting Voluntary Genomic Data Submission (VGDS). It is remarkable to note that despite the non-compulsory nature of this guiding principle, VGDS submissions are rapidly amassing and were filed for over 30 different compounds in the first year, exhibiting a growth trend going forward (Frueh, 2006).
■
329
EXAMPLES OF PGx TESTS: PROMISING NEW DEVELOPMENTS AND MARKETED PRODUCTS Warfarin Perhaps the best example of a successful pharmacogenetic association, for which the clinical relevance is clear, features management of treatment with the oral anti-coagulant warfarin, Warfarin, a derivative of coumarin, is prescribed for the longterm treatment and prevention of thromboembolic events, with more than 21 million prescriptions annually in the United States alone (Rieder et al., 2005). However, due to the drug’s narrow therapeutic index, a variety of complications are associated with its treatment, even after dose adjustment according to age, sex, weight, disease state, diet, and concomitant medications. Investigation of pharmacokinetic and pharmacodynamic drug properties indicated the additive involvement of two genes in determination of warfarin maintenance dose. The first gene to be identified encodes CYP2C9, which is responsible for most of the metabolic clearance (~80%) of the more pharmacologically potent S-enantiomer of warfarin (Rettie et al., 1992). Both CYP2C9*2 and *3 cause a reduction in S-warfarin clearance, with 10-fold variation seen from the genotype linked with the highest (CYP2C9*1/*1) to lowest (CYP2C9*3/*3) activity. Numerous studies have associated these genotypes with initial dose sensitivity, delayed stabilization of maintenance dose, delays in hospital discharge, and increased bleeding complications (Gardiner and Begg, 2006). However, it is estimated that CYP2C9 variants account for only 10–20% of the total variation in warfarin dose, with additional genetic and environmental factors playing larger roles in dose determination. The second gene identified as a predictor of dosing encodes the vitamin K epoxide reductase complex protein 1 (VKORC1), targeted by warfarin. Consideration of the VKORC1 genotype or haplotype, together with the CYP2C9 genotype, and factors such as age and body size are estimated to account for 35–60% of the variability in warfarin dosing requirements. Despite the fact that these data have been reproduced by multiple independent groups, it remains to be tested in large, prospective, clinical studies whether initial dose may be tailored to patients by CYP2C9 and VKORC1 genotyping, coupled with known clinical variables. While these studies are currently under way, the clinical pharmacology advisory panel to the FDA acknowledged (November 2005) the importance and potential for genotyping of CYP2C9 and VKORC1 during the early phase of warfarin therapy, and the drug label was amended accordingly in August 2007. A month later, the FDA cleared for marketing the Nanosphere Verigene Warfarin Metabolism Nucleic Acid Test to aid physicians manage warfarin therapy. Trastuzumab Trastuzumab therapy for the treatment of breast cancer (involving a monoclonal antibody specifically targeting HER2/neu
TABLE 27.2
FDA valid pharmacogenetic biomarkers – associated dosage guidelines in drug labels and availability of approved genetic tests (Summer, 2007)
330
Gene (variant)
Gene function/ significance
Dosage guidelines
Label section
PGx test?
FDA-approved test available?
Voriconazole [6]
CYP2C19 (poor metabolizers)
Major metabolizer
1. Contraindicated concomitantly with CYP450 inducers; 2. Increased maintenance dose regimen outlined for Phenytoin; 3. No adjustmentsa for oral contraceptives inhibiting CYP2C19 or HIV protease inhibitors that inhibit CYP3A4; 4. Reduce Cyclosporine and Omeprazole by half; 5. Reduce Tacrolimus to one third
Clinical pharmacology; Drug Interactions;
No
AmpliChip/Roche
Celecoxib [1]
CYP2C9 (poor metabolizers)
Major metabolizer
No
Clinical pharmacology; Drug interactions;
No
No
Atomoxetine [17]
CYP2D6 (poor metabolizers)
Major metabolizer
Specific regimen suggested when strong CYP2D6 inhibitor co-administered
Clinical pharmacology; Drug interactions; CYP2D6 metabolism; General dosing information
No
AmpliChip®/Roche®
Capecitabine [2]
DPD (deficiency)
Rate-limiting metabolizer
Contraindicated
Contraindications; Clinical pharmacology; Precautions
No
No
Rusburicase [3]
G6PD (deficiency)
Severe hemolysis caused by hydrogen peroxide by-product
Contraindicated
Boxed warning; Contraindications; Warnings
Recommended
No
Rifampin, isoniazid, and pyrazinamide [1]
NAT (slow acetylator)
Isoniazid major metabolizer
No
Clinical pharmacology; Adverse reactions; Drug interactions
No
No
Azathioprine [2]
TPMT (*2, *3A and *3C)
Metabolizer
Recommended altered regimen, but not specified
Warnings; Lab tests; Dosage and administrations; Drug interactions
Recommended
No
Irinotecan
UGT1A1 (*28)
Conjugation of active metabolite
Consider reduction of dose in homozygotes
Clinical pharmacology; Warnings; Dosage and administration
No
Invader assay/Third Wave
■
Most adjustments relate to CYP3A4 drug–drug interactions, although it is a minor metabolizer of Voriconazole.
Pharmacogenetics and Pharmacogenomics
This table is adopted from Grossman (2007). Details in this table are derived from package inserts of the listed drugs and the table of valid genomic biomarkers in the context of approved drug labels http://www.fda.gov/cder/genomics/genomic_biomarkers_table.htm. Note – A drug label change was introduced in August 2007 VKORC1 (in addition to CYP2C9) genotyping in patients requiring warfarin therapy. An accompanying genetic test (Nanosphere Verigene Warfarin Metabolism Nucleic Acid Test) was FDA-approved in late September 2007. a
CHAPTER 27
Drug [# drugs with similar label references]
Future Developments Required for the Field to Fully Meet its Expectations
over-expressing tumors) is not only an example of a protein therapeutic for which an obligatory biomarker assay has been issued, but it is also an example of the utilization of PGx research for drug rescue: trastuzumab is marketed solely for the subset of patients (~10–25%) who over-express HER2/neu, providing care to eligible patients and returning the investment to the developing manufacturer. Interestingly enough, although in practice the obligatory biomarker assay has been set to measure protein over-expression by immunohistochemistry, studies have shown stronger association when patient subsets were determined by fluorescence in situ hybridization, reflecting gene copy number (Krejsa et al., 2006). 6-Mercaptopurine In 1953, the drug 6-mercaptopurine (6-MP) was marketed in the United States for the treatment of leukemia. Despite great expectations in the medical world, about 20 years ago fatal toxicity was discovered in 0.3% of treated children. A similar scenario was repeated for azathioprine marketed in 1968. It was later discovered that polymorphisms within the TPMT gene underlie the large inter-individual differences in the enzyme’s activity, leading to a high risk of thiopurine-induced toxicity in homozygotes for the defective alleles, and inadequate therapeutic efficacy in patients with high activity TPMT. Tests for TPMT activity (genotype, enzyme activity, and metabolite screening) are available in the United States and throughout Europe; however, clinical implementation of these tests is very low. This is contrary to expectations based on cost-effectiveness analysis of TPMT testing in children with acute lymphocytic leukemia (ALL) showing high savings per life-year (Zika et al., 2006). Current clinical practice for the management of leukemia via thiopurine medications dictates careful monitoring of white blood cell counts and clinical outcomes in a fashion that negates the necessity of incorporating the genetic test for prevention of serious adverse events. It is expected, however, that as genetic tests become generally accepted for a variety of conditions, it will become progressively acceptable both by physicians and health systems, as well as by the general public, to use TPMT genetic tests as prognostic tools. AmpliChip The first microarray-based gene-chip, approved both in the United States and EU, was released to the market by F. Hoffmann-La Roche Ltd (Switzerland) in 2003 as the AmpliChip CYP450®. The product was designed to identify key genetic polymorphisms in two CYP450 genes, CYP2D6 and CYP2C19, the products of which are cumulatively responsible for much of the first pass metabolism of many pharmaceutical compounds. As highlighted by insurance companies who refuse to cover the costs ($600–$1300), the test has not been clearly demonstrated to convey clinical utility and sufficient cost-effectiveness (Grossman et al., 2007b; Thakur et al., 2007) and is deemed “experimental, investigational or unproven.” The regulatory agencies cleared this test based solely on analytical performance and validity information, but indicated its utility
■
331
specifically for clinical application. Thus clinicians, as well as patients, were misled to expect firm impact on clinical decisionmaking guidelines. Until unambiguous evidence proves the clinical use of this and other genetic tests, caution is advised in their interpretation and application in health care management.
FUTURE DEVELOPMENTS REQUIRED FOR THE FIELD TO FULLY MEET ITS EXPECTATIONS Table 27.3 summarizes steps still needed for PGx testing to materialize into practical tools used in clinical decisionmaking. First, in order to capture the entirety of processes involved in, and affected by, any given drug, both in vivo and in interaction with environmental factors, open communications between researchers, practitioners, regulatory agencies, and pharmaceutical industries must be established. Launching PGx educational programs within the academic curriculum will facilitate acceptance of genetic testing both by health care providers and by the general public. Next, facilitated by the latter, large prospective studies must be conducted in order to truly evaluate the utility of genetic tests for specific indications in “real-life” clinical setting.Tailoring the research to specifically target PGx-based end-points, rather than adopting and artificially manipulating retrospective measurements from treatment efficacy clinical trials, will assist in elucidating “true” PGx effects. In this fashion, employing placebo, control, and multiple dosing treatment groups may act as means for detection of specific drug-induced genetic mechanisms. As research progresses and technology develops the need for investigation of gene–gene and gene–environment effects in a computationally reasonable manner will become more and more pressing. Such lines of investigation will require large, often combined, datasets, necessitating the standardization of phenotype definitions by the research community. Lastly, incorporation of genetic testing in clinical practice will only become possible if regulatory and funding agencies acknowledge its potential and promote its development by changing their policy. PGx is a field uniquely attractive to both academic medical centers and the pharmaceutical and biotechnological industries. As such, governmental agencies, such as the US FDA and the EMA, regulate the development, marketing, and clinical applications associated with maturation of research advances. In order for PGx to wholly fulfill its potential and expectations a structured framework of incentives, priorities, and policies must be devised that would promote allocation of resources by each of the concerned parties. A potential solution to improve surveillance might be sampling and privacyprotected DNA banking from the first 250,000 patients treated with a newly marketed drug, or a similar risk-management system coordinated with regulatory agencies (Roses, 2004). The most pronounced deficiency identified in PGx research relates to the need to improve drug safety and efficacy profiles of generic, off-patent drugs. In this aspect it is important to mention
332
CHAPTER 27
TABLE 27.3
■
Pharmacogenetics and Pharmacogenomics
Steps required for PGx to fully meet its potential as “the right drug for the right patient”
Future challenges ● ● ● ● ● ● ● ● ●
Integrative approach: Cross-talk between geneticists, statisticians, pharmacologists, molecular biologists, physicians, regulatory agencies, and pharmaceutical companies. Education programs should be incorporated into the academic curriculum. Prospective studies have to be conducted to evaluate the potential predictive power and clinical utility of PGx testing in clinical settings. Possibly employ placebo and control treatment groups in analysis to elucidate specific drug-induced genetic effects. Focus on gene–gene interactions and the net genetic effect. Integrate environmental effects and gene – environment interactions into the statistical models. Standardize unequivocal and reproducible phenotype definitions. Genetics will get more and more acceptable as a tool for clinical practice as policy will catch up with scientific developments. Surveillance of ADRs in the post-marketing phase through risk-management systems coordinated by both manufacturers and regulatory agencies. Establishment of incentive structure for the industry to improve drug safety and efficacy beyond the terms of current patent protection.
that withdrawn medications lose their patent protection, and thus commercially driven pharmaceutical companies lose interest in rescuing these formulations by conducting safety PGx experiments. To date, it has yet to be demonstrated whether PGx would be cost-effective in resurrecting failed marketed
drugs (Shah, 2006), but thorough investigations have yet to be conducted. Only controlled incentives, calling for prolongation of patent protection terms and similar motivations, will drive the pharmaceutical industry to invest the considerable costs and resources required.
REFERENCES Aquilante, C.L., Langaee,T.Y., Lopez, L.M.,Yarandi, H.N.,Tromberg, J.S., Mohuczy, D., Gaston, K.L., Waddell, C.D., Chirico, M.J. and Johnson, J.A. (2006). Influence of coagulation factor, vitamin K epoxide reductase complex subunit 1, and cytochrome P450 2C9 gene polymorphisms on warfarin dose requirements. Clin Pharmacol Ther 79, 291–302. Arnett, D.K., Claas, S.A. and Glasser, S.P. (2006). Pharmacogenetics of antihypertensive treatment. Vascul Pharmacol 44, 107–118. CDC (2006). Evaluation of genetic testing. ACCE: A CDC-Sponsored Project carried out by the Foundation of Blood Research. http:// www.cdc.gov/genomics/gtesting/ACCEhtm. Chapman, J.M., Cooper, J.D., Todd, J.A. and Clayton, D.G. (2003). Detecting disease associations due to linkage disequilibrium using haplotype tags a class of tests and the determinants of statistical power. Hum Hered 56, 18–31. Conrad, D.F., Andrews,T.D., Carter, N.P., Hurles, M.E. and Pritchard, J.K. (2006). A high-resolution survey of deletion polymorphism in the human genome. Nat Genet 38, 75–81. de Bakker, P.I., Yelensky, R., Pe’er, I., Gabriel, S.B., Daly, M.J. and Altshuler, D. (2005). Efficiency and power in genetic association studies. Nat Genet 37, 1217–1223. http://www.broad.mit.edu/ mpg/tagger/server.html. de Bakker, P.I., Burtt, N.P., Graham, R.R., Guiducci, C., Yelensky, R., Drake, J.A., Bersaglieri, T., Penney, K.L., Butler, J., Young, S. et al. (2006). Transferability of tag SNPs in genetic association studies in multiple populations. Nat Genet 38, 1298–1303.
DiMasi, J.A., Hansen, R.W. and Grabowski, H.G. (2003). The price of innovation new estimates of drug development costs. J Health Econ 22, 151–185. Endrenyi, L., Inaba, T. and Kalow, W. (1976). Genetic study of amobarbital elimination based on its kinetics in twins. Clin Pharmacol Ther 20, 701–714. FDA (2004a). Challenge and opportunity on the critical path to new medical products. http://www.fda.gov/oc/initiatives/criticalpath/ whitepaperhtml. FDA (2004b). Guidance for Industry: Pharmacogenomic Data Submissions http://www.fda.gov/cber/gdlns/pharmdtasubhtm. Flockhart, D.A. (2007) Drug interactions. http://medicine.iupui. edu/flockhart/. Frueh, F.W. (2006). Impact of microarray data quality on genomic data submissions to the FDA. Nat Biotechnol 24, 1105–1107. Frueh, F.W., Goodsaid, F., Rudman, A., Huang, S.M. and Lesko, L.J. (2005). The need for education in pharmacogenomics: A regulatory perspective. Pharmacogenomics J 5, 218–220. Gardiner, S.J. and Begg, E.J. (2006). Pharmacogenetics, drug-metabolizing enzymes, and clinical practice. Pharmacol Rev 58, 521–590. Gaszner, P. and Makkos, Z. (2004). Clozapine maintenance therapy in schizophrenia. Prog Neuropsychopharmacol Biol Psychiatry 28, 465–469. Goldstein, D.B., Tate, S.K. and Sisodiya, S.M. (2003). Pharmacogenetics goes genomic. Nat Rev Genet 4, 937–947. Grossman, I. (2007). Routine pharmacogenetic testing in clinical practice – dream or reality? Pharmacogenomics 8, 1449–1459.
References
Grossman, I.,Avidanb, N., Singer, C., Goldstaub, D., Hayardeny, L., Eyal, E., Ben-Asher, E., Paperna, T., Pe’er, I., Lancet, D. et al. (2007a). Pharmacogenetics of glatiramer acetate therapy for multiple sclerosis reveals drug-response markers. Pharmacogenet Genomics 17(8), 657–666. Grossman, I., Liu, Y., Walley, N., et al. (Oct 2006). Pharmacogenetic analysis of antipsychotics: Comprehensive analysis of pharmacokinetic variants. Presented at: 56th Annual Meeting of the American Society of Human Genetics (ASHG). New-Orleans, LA, USA. Heath, A.C. and Martin, N.G. (1994). Genetic influences on alcohol consumption patterns and problem drinking results from the Australian NH&MRC twin panel follow-up survey. Ann N Y Acad Sci 708, 72–85. Hillman, M.A., Wilke, R.A., Yale, S.H., Vidaillet, H.J., Caldwell, M.D., Glurich, I., Berg, R.L., Schmelzer, J. and Burmester, J.K. (2005). A prospective, randomized pilot trial of model-based warfarin dose initiation using CYP2C9 genotype and clinical data. Clin Med Res 3, 137–145. Hughes, D.A., Vilar, F.J., Ward, C.C., Alfirevic, A., Park, B.K. and Pirmohamed, M. (2004). Cost-effectiveness analysis of HLA B*5701 genotyping in preventing abacavir hypersensitivity. Pharmacogenetics 14(6), 335–342. Hughes, S., Hughes, A., Brothers, C., Spreen, W. and Thorborn, D. (2007). PREDICT-1 (CNA106030) the first powered, prospective trial of pharmacogenetic screening to reduce drug adverse events. Pharm Stat. Ioannidis, J.P., Ntzani, E.E., Trikalinos, T.A. and Contopoulos-Ioannidis, D.G. (2001). Replication validity of genetic association studies. Nat Genet 29, 306–309. Jones, T.S., Yang, W., Evans, W.E. and Relling, M.V. (2007). Using HapMap tools in pharmacogenomic discovery: The thiopurine methyltransferase polymorphism. Clin Pharmacol Ther 81, 729–734. Kelly, P.J., Stallard, N. and Whittaker, J.C. (2005). Statistical design and analysis of pharmacogenetic trials. Stat Med 24, 1495–1508. Kendler, K.S., Martin, N.G., Heath, A.C., Handelsman, D. and Eaves, L.J. (1988). A twin study of the psychiatric side effects of oral contraceptives. J Nerv Ment Dis 176, 153–160. Kindmark, A., Jawaid, A., Harborn, C.G., Barratt, B.J., Bengtsson, O.F., Andersson, T.B., Carlsson, S., Cederbrant, K.E. et al. (2007). Genome-wide pharmacogenetic investigation of a hepatic adverse event without clinical signs of immunopathology suggests an anderlying immune pathogenesis. Pharmacogenomics J, 1–10. Kirchheiner, J., Henckel, H.B., Meineke, I., Roots, I. and Brockmoller, J. (2004). Impact of the CYP2D6 ultrarapid metabolizer genotype on mirtazapine pharmacokinetics and adverse events in healthy volunteers. J Clin Psychopharmacol 24, 647–652. Krejsa, C., Rogge, M. and Sadee, W. (2006). Protein therapeutics new applications for pharmacogenetics. Nat Rev Drug Discov 5, 507–521. Kurtzke, J.F. (1983). Rating neurologic impairment in multiple sclerosis an expanded disability status scale (EDSS). Neurology 33, 1444–1452. Lesko, L.J. and Woodcock, J. (2002). Pharmacogenomic-guided drug development regulatory perspective. Pharmacogenomics J 2, 20–24. Lesko, L.J. and Woodcock, J. (2004). Translation of pharmacogenomics and pharmacogenetics a regulatory perspective. Nat Rev Drug Discov 3, 763–769. McLeod, H.L., Tan, B., Malyapa, R., Abbey, E., Picus, J., Myerson, R., Zehnbauer, B., Mutch, M., Dietz, D. and Fleshman, J. (2005).
■
333
Genotype-guided neoadjuvant therapy for rectal cancer. Proc Am Soc Clin Oncol 23, 197. Meadows, M. (2005). FDA approves heart drug for black patients. FDA Consum 39, 8–9. Murphy, G.M., Jr., Kremer, C., Rodrigues, H.E. and Schatzberg, A.F. (2003). Pharmacogenetics of antidepressant medication intolerance. Am J Psychiatry 160, 1830–1835. NCI-NHGRI Working Group on Replication in Association Studies (2007). Replicating genotype-phenotype associations. Nature 447, 655–660. Nebert, D.W. and Dieter, M.Z. (2000). The evolution of drug metabolism. Pharmacology 61, 124–135. Need, A.C., Motulsky, A.G. and Goldstein, D.B. (2005). Priorities and standards in pharmacogenetic research. Nat Genet 37, 671–681. Nevins, J.R. and Potti, A. (2007). Mining gene expression profiles expression signatures as cancer phenotypes. Nat Rev 8, 601–609. O’Reilly, R.L., Bogue, L. and Singh, S.M. (1994). Pharmacogenetic response to antidepressants in a multicase family with affective disorder. Biol Psychiatry 36, 467–471. Pe’er, I., de Bakker, P.I., Maller, J.,Yelensky, R., Altshuler, D. and Daly, M.J. (2006). Evaluating and improving power in whole-genome association studies using fixed marker sets. Nat Genet 38, 663–667. Perry, P. and Hradek, K. (2004). Antipsychotic maintenance therapy. In Clinical Psychopharmacology Seminar. PharmGKB Pathways. http://www.pharmgkborg/search/pathway/ pathwayjsp. Rauch,A., Nolan, D., Martin,A., McKinnon, E.,Almeida, C. and Mallal, S. (2006). Prospective genetic screening decreases the incidence of abacavir hypersensitivity reactions in the Western Australian HIV cohort study. Clin Infect Dis 43, 99–102. Rettie, A.E., Korzekwa, K.R., Kunze, K.L., Lawrence, R.F., Eddy, A.C., Aoyama, T., Gelboin, H.V., Gonzalez, F.J. and Trager, W.F. (1992). Hydroxylation of warfarin by human cDNA-expressed cytochrome P-450: a role for P-4502C9 in the etiology of (S)-warfarindrug interactions. Chem Res Toxicol 5, 54–59. Rieder, M.J., Reiner, A.P., Gage, B.F., Nickerson, D.A., Eby, C.S., McLeod, H.L., Blough, D.K., Thummel, K.E., Veenstra, D.L. and Rettie, A.E. (2005). Effect of VKORC1 haplotypes on transcriptional regulation and warfarin dose. N Engl J Med 352, 2285–2293. Risner, M.E., Saunders, A.M., Altman, J.F., Ormandy, G.C., Craft, S., Foley, I.M., Zvartau-Hind, M.E., Hosford, D.A. and Roses, A.D. (2006). Efficacy of rosiglitazone in a genetically defined population with mild-to-moderate Alzheimer’s disease. Pharmacogenomics J 6, 246–254. Roses, A.D. (2004). Pharmacogenetics and drug development the path to safer and more effective drugs. Nat Rev Genet 5, 645–656. Roses, A.D., Saunders, A.M., Huang, Y., Strum, J., Weisgraber, K.H. and Mahley, R.W. (Feb 2007). Complex disease-associated pharmacogenetics: Drug efficacy, drug safety, and confirmation of a pathogenetic hypothesis (Alzheimer’s disease). Pharmacogenomics J 7(1): 10–28. Singer, C., Grossman, I., Avidan, N., Beckmann, J.S. and Pe’er, I. (2005). Trick or treat the effect of placebo on the power of pharmacogenetic association studies. Hum Genomics 2, 28–38. Shah, R.R. (2006). Can pharmacogenetics help rescue drugs withdrawn from the market? Pharmacogenomics 7(6), 889–908. Skol, A.D., Scott, L.J., Abecasis, G.R. and Boehnke, M. (2006). Joint analysis is more efficient than replication-based analysis for twostage genome-wide association studies. Nat Genet 38, 209–213.
334
CHAPTER 27
■
Pharmacogenetics and Pharmacogenomics
Spraggs, C.F., Pillai, S.G., Dow, D., Douglas, C., McCarthy, L., Manasco, P.K., Stubbins, M. and Roses, A.D. (2005). Pharmacogenetics and obesity common gene variants influence weight loss response of the norepinephrine/dopamine transporter inhibitor GW320659 in obese subjects. Pharmacogenet Genomics 15, 883–889. Swan, G.E., Benowitz, N.L., Jacob, P., 3rd, Lessov, C.N., Tyndale, R.F., Wilhelmsen, K., Krasnow, R.E., McElroy, M.R., Moore, S.E. and Wambach, M. (2004). Pharmacogenetics of nicotine metabolism in twins methods and procedures. Twin Res 7, 435–448. Tate, S.K., Depondt, C., Sisodiya, S.M., Cavalleri, G.L., Schorge, S., Soranzo, N., Thom, M., Sen, A., Shorvon, S.D., Sander, J.W. et al. (2005). Genetic predictors of the maximum doses patients receive during clinical use of the anti-epileptic drugs carbamazepine and phenytoin. Proc Natl Acad Sci USA 102, 5507–5512. Telenti, A. and Goldstein, D.B. (2006). Genomics meets HIV-1. Nat Rev Microbiol 4, 865–873. The International HapMap Project (2003). Nature 426, 789–796. Thakur, M., Grossman, I., McCrory, D.C., Orlando, L.A., Steffens, D.C., Cline, K.E., Gray, R.N., Farmer, J., DeJesus, G., O’Brien, C., Samsa, G., Goldstein, D.B., Matchar, D.B. (2007). Review of evidence for genetic testing for CYP450 polymorphisms in management of patients with non psychotic depression with selective serotonin reuptake inhibitors (SSRIs). Gen Med 9, 826–835.
van den Akker-van Marle, M.E., Gurwitz, D., Detmar, S.B., Enzing, C.M., Hopkins, M.M., Gutierrez de Mesa, E. and Ibarreta, D. (2006). Cost-effectiveness of pharmacogenomics in clinical practice a case study of thiopurine methyltransferase genotyping in acute lymphoblastic leukemia in Europe. Pharmacogenomics 7, 783–792. Warren, L.L., Hughes, A.R., Lai, E.H., Zaykin, D.V., Haneline, S.A., Bansal, A.T., Wooster, A.W., Spreen, W.R., Hernandez, J.E., Scott, T.R., Roses, A.D., Mosteller, M. (2007). Use of pairwise marker combination and recursive partitioning in a pharmacogenetic genome-wide scan. Pharmacogenomics J 7, 180–189. Weiss, S.T., Litonjua, A.A., Lange, C., Lazarus, R., Liggett, S.B., Bleecker, E.R. and Tantisira, K.G. (2006). Overview of the pharmacogenetics of asthma treatment. Pharmacogenomics J 6, 311–326. Wilke, R.A., Reif, D.M. and Moore, J.H. (2005). Combinatorial pharmacogenetics. Nat Rev Drug Discov 4, 911–918. Wilkinson, G.R. (2005). Drug metabolism and variability among patients in drug response. N Engl J Med 352, 2211–2221. Zika, E., Gurwitz, D. and Ibearreta, D. (2006). Pharmacogenetics and Pharmacogenomics: State-of-the-art and potential socio-economic impact in the EU. European Communities. Zucman, D.,Truchis, P., Majerholc, C., Stegman, S. and Caillat-Zucman, S. (2007). Prospective screening for human leukocyte antigen-B*5701 avoids abacavir hypersensitivity reaction in the ethnically mixed French HIV population. J Acquir Immune Defic Syndr 45, 1–3.
RECOMMENDED RESOURCES Journals
Websites
Ioannidis, J.P., Ntzani, E.E., Trikalinos, T.A., and Contopoulos-Ioannidis, D.G. (2001). Replication validity of genetic association studies. Nat Genet 29, 306–309. NCI-NHGRI Working Group on Replication in Association Studies (2007). Replicating genotype-phenotype associations, Nature 447, 655–660. Roses, A.D. (2004). Pharmacogenetics and drug development: The path to safer and more effective drugs. Nat Rev Genet 5, 645–656.
http://www.hapmap.org/International Hap Map Project web site http://www.cdc.gov/genomics/gtesting/ACCE.htm ACCE: A CDCsponsored project for evaluation of genetic testing http://www.fda.gov/cber/gdlns/pharmdtasub.htm US Food and Drug Administration site for pharmacogenetic data submissions http://www.pharmgkb.org/search/pathway/pathway.jsp PharmGKB Pathways
CHAPTER
28 The Role of Genomics and Genetics in Drug Discovery and Development Robert I. Tepper and Ronenn Roubenoff
INTRODUCTION The elucidation of the structure and the increasing knowledge of the function of human genes and their interactions are transforming the process by which novel therapeutic agents are identified. Prior to 1990, the collection of therapeutic agents commonly used in medicine targeted less than 400 gene products (Overington et al., 2006). These therapeutic agents, mostly low molecular weight (e.g., 800 d) chemical entities, were typically identified based on the pharmacological activity of natural product extracts or synthesized chemical molecules screened in complex in vitro or in vivo assays. The identification of the target proteins for these drugs typically came after the identification of a pharmacological effect and involved years of complex research involving biochemical purification and characterization. Using this labor-intensive approach, most of the gene products of the human genome associated with important physiological and pathophysiological states could not be identified. The genomic revolution of the 1990s allowed for the rapid identification of the vast majority of human genes and thereby provided a means to identify most human gene products. The potential set of targets for which drugs could be identified therefore increased dramatically. In addition, many new approaches were conceived that allowed for the functional correlation of genes with each other into functional “pathways” (Boehm et al., 2007). The association of individual genes, or
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
pathways, with disease states has led to a dramatic advance in our understanding of the genetic deregulation that characterizes human disease. This fundamental molecular understanding of disease has provided many novel targets for drug discovery. In addition, the subclassification of common diseases on the bases of the genetic pathways that are altered has led to a more precise way of identifying which patients are likely to benefit from a particular therapeutic agent. The drug discovery process itself has also taken advantage of new methodologies for screening drugs, using “whole-genome” approaches (Kramer and Cohen, 2004). These advances in drug discovery and development resulting from genomics are described in more detail in this chapter.
THE DRUG DISCOVERY PROCESS The steps involved in the discovery and development of novel drugs are depicted in Figure 28.1. The time period from the start of novel target identification to the entry of a new drug into the marketplace is a lengthy one, typically over 12 years. Moreover, the average cost for this process is very high and is increasing. Current estimates for the research and development cost to deliver one successful drug into the market is in excess of one billion dollars, when one includes in the average cost of drugs that have failed during the discovery and development process (Munos, 2006). Approximately 10–12 drug candidates are generated and tested
Copyright © 2009, Elsevier Inc. All rights reserved. 335
336
CHAPTER 28
■
Role of Genomics in Drug Discovery and Development
The drug discovery and development process Target identification and validation
Chemical screening
Lead identification and optimization
Discovery 3–5 years Preclinical animal studies
Clinical trials
Regulatory approval and marketing
Development 5–7 years
Figure 28.1 Basic steps in the drug discovery and development process.
for one successful drug to be developed. Taken together, the time interval, cost, and 90% failure rate make the pharmaceutical process an extremely risky endeavor, particularly for drug candidates that are “first in class.” The development of genomic and associated proteomic strategies has provided a wealth of opportunities for the development of completely novel drugs. While the prospects for major breakthroughs in our understanding of disease pathophysiology remain great, genomic strategies in their first decade of incorporation into the drug discovery process have not resulted in an increase in productivity (decreased cost and/or time) (Betz, 2005; Brown and Superti-Furga, 2003). This is in part due to the need to create the requisite informatics processes and the biological material (e.g., patient tissues for DNA, RNA, and protein analyses) and correlative clinical information to make genomic information optimally useful. In addition, experience with genomics techniques is rapidly increasing, and this is also expected to impact positively on the productivity of processes that utilize genomic information.
GENOMICS IN TARGET DISCOVERY Despite the relatively nascent use of genomics in drug discovery, there are already many aspects of the process that have been improved through the use of genomic information. Target discovery, the process by which novel drug targets (i.e., the gene product, typically a protein to which the drug is directed) are identified and biologically validated, has perhaps been the most mature focus of genomics applications to date. Extensive analyses of the human genome have identified many of the 20,000 protein gene products in man (International Human Genome Sequencing Consortium, 2004). This information has allowed scientists to comprehensively identify major classes of proteins in the human genome. Take for example the case of protein kinases, of which there are about 1000 representatives in the human genome (Kostich et al., 2002). Kinases are “druggable,” meaning that there are well-established methods for developing chemical
modulators of these targets (most often inhibitors) bearing drug-like properties. They are also known to play important roles in human physiology, for example, by regulating the process of intracellular signal transduction which controls fundamental cellular processes including cell division, programmed cell death (apoptosis), and cellular differentiation in various cell types (Kostich et al., 2002). An estimated seventy or more novel kinase targets have entered the drug discovery process over the past 10 years, and the associated chemical antagonists generated have been studied in a number of pathophysiologic states, including cancer, various inflammatory states, and metabolic disorders. A list of major “druggable” protein classes that have been extensively expanded and studied using genomic approaches are provided by Zheng et al. (2006), with an estimate of the number of drug targets in 2006 being over 1500, compared to less than 500 in 1996. Some examples of drug candidates that have been developed as a result of genetic or genomic screening include the Janus kinase and p38 mitogen-activated protein (MAP) kinase inhibitors for inflammation; sunitinib, a multiple receptor tyrosine kinase inhibitor approved for renal cell carcinoma; and gefitinib, an epidermal growth factor receptor tyrosine kinase inhibitor approved for lung cancer. Another fundamental genomics technology used extensively in the target validation process is that of expression profiling. Expression profiling refers to the determination of the transcriptional activity of multiple genes (often the entire human genome) in one or more tissue or cell types. Expression profiling studies have been used extensively to determine the differences in gene expression between normal and diseased tissues (Golub et al., 1999; Hedenfalk et al., 2001; Khan et al., 2001); for example, the comparison of normal lung tissue with lung cancer of various types or the comparison of normal and inflamed organs. In this way, genes that are candidates for involvement in pathologic states may be identified. Further biological studies may then be performed to determine whether any of these candidate genes correlated with disease activity may be causally involved in the disease process. If so validated, these gene products, if chemically tractable, may be novel targets for drug discovery. Comparative genomics has also played an important role in drug discovery (Youngman et al., 2001). Comparative genomics refers to the identification of the DNA sequence variability in gene orthologs. Orthologs are genes in different species that evolved from a common ancestral gene by speciation. Normally, orthologs retain the same function in the course of evolution. Identification of orthologs is critical for reliable prediction of gene function, as invariant regions of orthologous genes are typically essential for normal gene function. By identifying orthologs from a number of species, drug discovery scientists often obtain valuable information on how to inhibit the function of a given gene product. Moreover, identification and testing of multiple orthologs also ensures that chemical inhibitors that are being developed against one species (e.g., humans) are active in others. As the process of drug efficacy and safety testing makes extensive use of non-human organisms (e.g., rodents and monkeys), ortholog testing has been an important advance
Pharmacogenomics and Drug Development
Identify/Validate Identify Lead Gene Targets Compounds
Target Identification
Target Validation
Lead Development
Eliminate Targets Genomics
Improve Efficacy and Safety
Preclinical
Validate Leads
Clinical
Target Patients
Market
Select Patients Genomic Medicine
Figure 28.2 Genetic and genomic methods have impacted nearly every phase of drug development.
in the generation of new drugs. The ability to rapidly determine sequence variation in a target gene from one human to another has also been a valuable tool in the drug discovery process. In this way, one can test whether any sequence variation is present in the target gene within the population and at what frequency and whether or not the variation affects the ability of the drug to act on the target. Sequence variation in disease-associated genes has also shed light on the susceptibility of individuals to a given disease (e.g., see Duerr et al., 2006).
GENOMIC APPROACHES TO DRUG IDENTIFICATION The advent of genetic and genomic capability has modified how drug candidates are discovered and developed (Figure 28.2). The process of high-throughput screening (HTS), by which thousands of chemical compounds are rapidly assayed for a biological effect using automated methods, has become a routine part of modern drug discovery (Lang et al., 2006; Sams-Dodd, 2005). Compounds that score positive in high-throughput screening assays (commonly known in the drug discovery vernacular as “hits”) are then followed up with secondary assays that validate the specificity of the primary screen. Validated hits are then prioritized based on their chemical tractability and further refined using medicinal chemistry to generate lead compounds (or simply “leads”). An iterative process known as lead optimization then follows, consisting of repeated cycles of chemical modifications and biological assays, the latter to define the lead molecule’s potency against the target and a number of pharmacological properties including predicted solubility, distribution, and metabolism. While the drug screening, hit validation, and “hit to lead” processes can be performed quite efficiently (often less than 6 months for all three), it is not unusual for the lead optimization process to require more than a year of effort on the part of 15–20 biologists and chemists. Genomics has already contributed quite extensively to the drug screening and identification process. Completely novel
■
337
screening methodologies that rely on “whole-genome” screening approaches have been designed. An example of this is RNAi-based screens (Chen et al., 2007). RNAi, or RNA interference, is a biological process of gene-expression regulation in which double-stranded RNA inhibits the expression of individual genes with complementary nucleotide sequences. Using this approach, genes can be inactivated, one at a time, and the phenotypic effect can be studied in a cell-based screen. If one is interested, for example, in understanding how the effects of a cytotoxic anti-cancer drug can be augmented, one can devise a high-throughput RNAi screen in which individual genes are inactivated in a cell in the presence of the drug (Whitehurst et al., 2007). In individual assay wells where cytotoxicity has been augmented, one can hypothesize that inactivation of the specific gene product complements the anti-cancer effect of the drug. Conversely, inhibition of the drug’s cytotoxicity would suggest that inhibition of the specific gene product confers a degree of resistance of the drug’s cytotoxic effect on the cancer cell. In this way, novel targets for anti-cancer agents and/or an understanding of cancer drug resistance can be studied. Innovative databases and knowledge management tools are also being devised to facilitate the link between drug action, functional genomic information and disease biology. Representative of this class of tools is the “Connectivity Map” devised by Justin Lamb, Todd Golub, and colleagues (Lamb et al., 2006). The connectivity map is a reference collection of geneexpression profiles from cultured human cells treated with bioactive small molecules, together with software that allows for pattern-matching. This map has demonstrated utility in finding connections among small molecules sharing a common mechanism of action, as well as chemicals and biological processes. As such, it serves as an important tool to link diseases, known drugs, and drug candidates. Genomic signatures from patient specimens (e.g., gene-expression patterns and DNA polymorphisms), for example in cancer, have also been studied as a guide in selecting specific chemotherapeutic agents (Dressman et al., 2007; Potti et al., 2006a, b), demonstrating the relevance of novel genomics approaches to clinical medicine.
PHARMACOGENOMICS AND DRUG DEVELOPMENT There is a growing concern in the pharmaceutical industry that the “Blockbuster Model” of doing business is coming to an end. There are many reasons for this, which are beyond the scope of this discussion. However, one important reason is that the notion of a single drug being adequate for all patients with a disease is no longer tenable in the post-genomic era, when it is possible to identify genetic markers of response or non-response to a drug even before it is given. The reluctance of drug companies to pursue this line of research in the hopes that a drug will be given to everyone even if it only works in a small subgroup of patients was dealt a severe blow by the experience of Astra Zeneca with gefitinib, a first-in-class tyrosine kinase inhibitor
338
CHAPTER 28
■
Role of Genomics in Drug Discovery and Development
for non–small-cell lung cancer. The drug was approved by Food and Drug Administration (FDA) with an approximately 10% response rate on the basis of two Phase 2 studies and an ongoing Phase 3 study. At the time of approval, in May 2003, the sponsor had not been able to discover a pharmacogenomic marker that would identify responders. However, such a biomarker was actually published only a year later, on May 20, 2004, by an academic group (Lynch et al., 2004). Within months, two academic labs were providing research-grade tests to clinicians, and a validated, commercial test was available shortly thereafter. As a result, gefitinib sales declined precipitously because physicians could identify the 10% or so of patients who were likely to respond before prescribing the drug. Subsequently, Phase 3 data showed no survival benefit for patients treated with gefitinib of and the drug was abandoned in favor of erlotinib (Tarceva, Genentech/ OSIP), a more efficacious follow-on medication. In contrast to this unhappy experience, there is a growing list of drugs for which pharmacogenomic biomarkers play a key role in development, approval, and subsequent use (see Table 28.1). However, failures in the development of genomically based drugs are still achingly common, although the success stories are often triumphs of science and drug development, along with a modicum of luck. The first great success, which remains in many ways the hallmark, is trastuzumab (Herceptin, Genentech), a drug that has become a mainstay of breast cancer therapy. Trastuzumab is indicated for the treatment of breast cancer that overexpresses a cell surface protein called HER2/ neu, which happens in about 25% of patients. If this protein is present, the disease is more likely to metastasize, and trastuzumab has been shown to prolong survival significantly. However, in the absence of the protein, which is the case in the majority of patients, there would be no benefit to treatment, and the patient would only incur the expense and potential side effects, including congestive heart failure. Therefore, the availability of a diagnostic test for HER2/neu is as important as the drug itself in both drug development and therapeutics. In 1998, trastuzumab was the first drug to be approved jointly with a paired diagnostic test, based on immunohistochemistry. An improved assay, using fluorescence in situ hybridization (FISH) was approved in 2001. The new assay was a major improvement because it was more precise and accurate than the original, which misclassified as many as 1/4 of patients and limited clinician’s enthusiasm for the drug. After the new assay appeared, trastuzumab went on to become a blockbuster therapeutic. The two examples cited above are both based on pharmacogenomic variability in the drug target, which in turn alters the pharmacodynamics of the drugs. However, this is not the only type of pharmacogenomics that can be useful in drug development. Variability in genes encoding drug-metabolizing enzymes can markedly alter the pharmacokinetics of drugs, as can variations in drug transporters. Some examples of how genetic variations in both target genes and metabolic/transporter genes can affect drug action are shown in Table 28.2 (see Chapter 27). Variation in metabolism of purine analogs such as azathioprine (Imuran) and mercaptopurine (6-MP) has been
T A B L E 2 8 . 1 Examples of drugs developed based on genetic or genomic information Drug
Year approved
Target
Biomarker
Imatinib
2001
Bcr-abl mutation
Philadelphia chromosome
Trastuzumab
1998
HER2/neu mutation
FISHa assay
Mirostipen
Not approved; withdrawn by sponsor in 2002
Myeloid progenitor inhibitory factor-1
NA
MLN518
In development
FLT-3 tyrosine kinase
FLT-3 mutation by PCRa
Cetuximab
2004
Epidermal growth factor receptor
EGFR expression
Gefitinib
2003–2005b
Erlotinib
2005
Panitumumab
2006
EGFR mutations EGFR mutations K-RAS wild type by PCR
a
FISH, fluorescence in situ hybridization; PCR, polymerase chain reaction. b Withdrawn after failing to demonstrate survival benefit in phase 3 studies.
demonstrated by Weinshilboum and colleagues (Weinshilboum, 2003). These drugs are metabolized by thiopurine methyltransferase (TPMT). People who are homozygous for loss of function variants in this gene – about 1% of Caucasians – are at risk for bone marrow aplasia if given standard doses of these drugs for treatment of cancer or autoimmune disease. In addition, about 10% of patients who are heterozygous for loss of TPMT function are at risk for bone marrow toxicity. Conversely, activity of TPMT can interfere with the success of chemotherapy for leukemia with these purine analogs, and testing of genetic variants has now become commonplace and is recommended in the drug label for 6-MP (Lennard et al., 1990).
PHARMACODYNAMIC MARKERS AND THEIR ROLE IN DRUG DISCOVERY AND DEVELOPMENT It is virtually impossible to develop a successful drug without a robust pharmacodynamic (PD) marker. In early stages of
Pharmacodynamic Markers and their Role in Drug Discovery and Development
■
339
TABLE 28.2 Examples of medications that are affected by pharmacogenetic variation in drug target or metabolism or both (from Roden et al., 2006) Drug
Variable clinical effect
Genes with associated variants
Possible mechanism
Azathioprine and Mercaptopurine
Bone marrow aplasia Reduced therapeutic effect at standard doses
TPMT
Hypofunctional alleles Wild-type alleles
Some antidepressants and -blockers
Increased side-effect risk Decreased efficacy
CYP2D6
Hypofunctional alleles Gene duplication
Omeprazole
Helicobacter pylori cure rate
CYP2C19
Hypofunctional alleles
Irinotecan
Neutropenia
UGT1A1
Decreased expression due to regulatory polymorphism
HIV protease inhibitors
Central nervous system levels
MDR1
Altered P-glycoprotein function
-blockers
Blood pressure lowering and heart rate slowing
ADRB1
Altered receptor function or number
Inhaled 2-agonists
Bronchodilation
ADRB2
Altered receptor function or number
Diuretics
Blood pressure lowering
ADD1
Altered cytoskeletal function by adducin variants
Warfarin
Anticoagulation
VKORC1 CYO2C9
Variant haplotypes in regulatory regions leading to variable expression Coding region variants causing reduced S-warfarin clearance
Abacavir
Immunological reactions
HLA variants
Altered immunologic responses
QT-prolonging antiarrhythmics
Drug-induced arrhythmia
Ion-channel genes
Exposure of subclinical reduction in repolarizing currents by drugs
General anesthetics
Malignant hyperthermia
RYR1
Anesthetic-induced increased release of sarcoplasmic reticulum calcium by mutant channels
Inhaled steroids
Bronchodilation
CRHR1
Unknown
HMG-CoA reductase inhibitors (statins)
Low-density lipoprotein cholesterol lowering
HMGCR
Altered HMG-CoA reductase activity
ADD1, the gene encoding-adducin; ADRB1, gene encoding the 1-adrenergic receptor; ADRB2, the gene encoding the 2-adrenergic receptor; CRHR1, the gene encoding corticotrophin-releasing hormone receptor-1; CY2C19, the gene encoding the 2C19 cytochrome P450 isoform; CYP2C9, the gene encoding the 2C9 cytochrome P450 isoform; CYP2D6, the gene encoding the 2D6 cytochrome P450 isoform; HMG-CoA, 3 hydroxy-3-methylglutaryl coenzyme A; HMGCR, the gene encoding HMG-CoA reductase; MDR1, the gene encoding P-glycoprotein; RYR1, the gene encoding the skeletal muscle calcium-release channel; TPMT, the gene encoding thiopurine methyltransferase; UGT1A1, the gene encoding uridine diphosphate glycosyltransferease 1 family, polypeptide A1; VKORC1, the gene encoding vitamin K epoxide reductase complex, subunit 1.
pre-clinical development, the biomarker is needed to select among competing drug candidates, usually using rodents to compare the pharmacokinetic (PK), PD, and toxicological profiles of leading candidate molecules to select the best ones to take forward into more extensive (and expensive) evaluation. Once the drug enters human trials, a PD marker is crucial for understanding dose selection. If a drug fails to elicit the desired
clinical response, is this because there was not enough drug, or because the drug did not reach its target, or because the target itself is not an important cause of the disease? Without having both PK and PD information, this question cannot be answered. Instead, the program must rely on toxic effects to determine the highest possible dose, and then reduce the dose sufficiently to achieve tolerability and hope that there will still be efficacy.
340
CHAPTER 28
■
Role of Genomics in Drug Discovery and Development
Such an approach can work, but it is obviously suboptimal, more hazardous, and carries a higher risk of ultimately failing in latestage trials. Thus, the development and validation of a robust PD marker is an essential step in any drug development program. Genetic information and genomic methods are playing an increasingly prominent role in the discovery and validation of PD assays. Understanding of the pathway in which a drug target operates is crucial to developing potential PD assays against it. It has become commonplace to use transcriptional profiling to measure changes in expression of genes up and downstream from the drug’s target in order to understand which points in the pathway are most sensitive measures of the drug’s effects. From this rapid and rather non-specific method, more specific assays can be planned. Methods used for PD assays include flow cytometry, Western blotting, plate-based immunoassays (ELISAs), single- and multiplexed sandwich immunoassays, and immunohistochemistry. In the final assay, the ability to create a Good Laboratory Practice (GLP) acceptable test becomes an important consideration. However, for interim use and internal decision-making, GLP assays are generally not required. For example, anti-tumor drugs that affect molecular chaperones such as heat shock protein (HSP)-90 are known to affect the conformation, stability, and function of various “client” proteins, including ERBB2, AKT, and ERK. Thus, drugs that inhibit HSP-90 are expected to cause declines in these proteins, as well as an increase in HSP-70 (Powers and Workman, 2007). Working from this knowledge, it is possible to measure declines in ERBB2 or phosphorylated ERK in circulating white blood cells as PD markers of exposure to inhibitors of HSP-90. These assays, although derived from genetic and genomic understanding of the target pathway, can be converted to proteomic assays and used in GLP-compliant fashion to facilitate regulatory acceptance of the results.
TOXICOGENOMICS Toxicogenomics is the use of genetic and genomic assays to perform tests of the safety of a drug or drug candidate. The goal is to identify toxic compounds earlier in the discovery process than is currently the norm. Traditionally, toxicological evaluation has been done in laboratory animals, beginning with rodents and rabbits. Because these studies are laborious and expensive, they do not lend themselves to high throughput and are generally performed only after a lead compound has been identified. As a result, toxicity profiles of molecules are not generated until relatively late in drug discovery, and such findings can stop a molecule’s development only after a high initial expenditure. With new genetic and genomic toxicology, the goal is to develop high-throughput toxicity profiling that can be applied earlier in the drug discovery process to more compounds, effectively weeding out potential toxins early and at a lower price. Several companies, such as Iconix Biosciences and GenoLogic, have developed platforms for gene-expression toxicity studies. Not only can these methods be used to do experiments faster, but they can also give additional
insight into the pathogenesis of the toxicity, since expression profiling can give clues to the biology far more comprehensively than traditional histology or clinical testing can. Once the cause of the toxicity is understood, pharmacologists can develop screening tests for identifying these problems in back-up molecules and do this even earlier in the drug development process.
GENETICS AND GENOMICS IN CLINICAL TRIAL DESIGN It has now become feasible to obtain DNA results within a few days of drawing a sample, so that clinical trials can use genetic results to either screen potential subjects or stratify them. The most commonly used example is testing for Her2/neu mutations (using immunohistochemistry or in situ hybridization on clinical biopsy samples) before treatment for breast cancer. Sometimes, tight linkage between a genetic haplotype and serum protein marker allows the latter to be used as an index of the former without necessarily needing to do the genetic test. This is the case in rheumatoid arthritis, where anti-citrullinated cyclic peptide (CCP) can be used as an index for the presence of the human leukocyte antigen (HLA) shared epitope (SE). Recently, the Committee for Medicinal Products for Human Use (CHMP) recommended approval of panatimumab, an EGFR inhibitor, for colorectal cancer, only in patients with the unmutated KRAS allele (see www.emea.europa.eu/pdfs/human/ opinion/40511307en.pdf dated September 20, 2007). In the near future, it is likely that gene-expression profiling (using whole blood RNA profiling, for example) will also be feasible in “real time”, could be used to identify patients in whom targeted pathways are activated, and should thus be more amenable to pharmacological intervention.
GENOMICS IN DRUG APPROVAL AND REGULATION Drug development is a regulated process, and success is measured by receiving regulatory approval to manufacture and sell the resulting product. Because health care is so expensive, and new medications are an appreciable and growing part of that cost, regulators are very interested in any information that clarifies the risk-benefit and cost-benefit of a new drug. Thus, both the FDA in the United States and the European Medicines Evaluation Agency (EMEA) have been very involved in supporting and even demanding pharmacogenomic and genetic analyses. Both groups have developed guidelines for submitting genomic and genetic data, and FDA has offered a “safe harbor” mechanism under which such data can be submitted without risk to the critical path for drug approval. In 2008, FDA is expected to follow up with new guidance for use of biomarkers in early drug development, again with genomic and genetic testing being prominently involved.
References
CONCLUSION The comprehensive nature of genomic technologies and the progress to date underscore their importance to drug discovery and development in the future. Genomics has already had a major impact on the identification of novel targets for drug discovery, the development of novel drug-screening methodologies, the link between disease pathways and pharmacological effects, and the selection of drug treatment options for individual patients. The vast amount of data served up by the genomics revolution has required a “lag phase” of growth, as tools and biological
■
341
and computational resources and reagents have been developed to complement the genomic information. However, it is now likely that critical mass has been reached and that the benefits of genomic and genetic methods will become evident at an increasing pace. The wealth of information emerging in the last several years implies that we are emerging from the lag phase into a log phase of growth in the application of genomics to important disease-relevant biological inquiries, including novel drug discovery. The investment in genomics over the past two decades is providing a wealth of important information which will be essential in providing therapeutic solutions for major human diseases.
REFERENCES Betz, U.A. (2005). How many genomics targets can a portfolio afford? Drug Discov Today 10, 1057–1063. Boehm, J.S., Zhao, J.J., Yao, J., Kim, S.Y., Firestein, R., Dunn, I.F., Sjostrom, S.K., Garraway, L.A., Weremowicz, S., Richardson, A.L. et al. (2007). Integrative genomic approaches identify IKBKE as a breast cancer oncogene. Cell 129, 1065–1079. Brown, D. and Superti-Furga, G. (2003). Rediscovering the sweet spot in drug discovery. Drug Discov Today 8, 1067–1077. Chen, M., Du, Q., Zhang, H.Y., Wang, X. and Liang, Z. (2007). Highthroughput screening using siRNA (RNAi) libraries. Expert Rev Mol Diagn 7, 281–291. Dressman, H.K., Berchuck, A., Chan, G., Zhai, J., Bild, A., Sayer, R., Cragun, J., Clarke, J., Whitaker, R.S., Li, L. et al. (2007). An integrated genomic-based approach to individualized treatment of patients with advanced-stage ovarian cancer. J Clin Oncol 25, 517–525. Duerr, R.H., Taylor, K.D., Brant, S.R., Rioux, J.D., Silverberg, M.S., Daly, M.J., Steinhart, A.H., Abraham, C., Regueiro, M., Griffiths, A. et al. (2006). A genome-wide association study identifies IL23R as an inflammatory bowel disease gene. Science 314, 1461–1463. Golub, T.R., Slonim, D.K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov, J.P., Coller, H., Loh, M.L., Downing, J.R., Caligiuri, M.A. et al. (1999). Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring. Science 286, 531–537. Hedenfalk, I., Duggan, D., Chen, Y., Radmacher, M., Bittner, M., Simon, R., Meltzer, P., Gusterson, B., Esteller, M., Kallioniemi, O.P. et al. (2001). Gene-expression profiles in hereditary breast cancer. N Engl J Med 344, 539–548. International Human Genome Sequencing Consortium (2004). Finishing the euchromatic sequence of the human genome. Nature 431, 931–945. Khan, J.,Wei, J.S., Ringnér, M., Saal, L.H., Ladanyi, M., Westermann, F., Berthold, F., Schwab, M., Antonescu, C.R., Peterson, C. et al. (2001). Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks. Nat Med 7, 673–679. Kostich, M., English, J., Madison,V., Gheyas, F.,Wang, L., Qiu, P., Greene, J. and Laz, T.M. (2002). Human members of the eukaryotic protein kinase family. Genome Biol 3. RESEARCH0043.1–0043.12 Kramer, R. and Cohen, D. (2004). Functional genomics to new drug targets. Nat Rev Drug Discov 3, 965–972.
Lamb, J., Crawford, E.D., Peck, D., Modell, J.W., Blat, I.C., Wrobel, M.J., Lerner, J., Brunet, J.P., Subramanian, A., Ross, K.N. et al. (2006). The Connectivity Map: Using gene-expression signatures to connect small molecules, genes, and disease. Science 313, 1929–1935. Lang, P.,Yeow, K., Nichols, A. and Scheer, A. (2006). Cellular imaging in drug discovery. Nat Rev Drug Discov 5, 343–356. Lennard, L., Lilleyman, J., Van Loon, J. and Weinshilboum, R. (1990). Genetic variation in response to 6-mercaptopurine for childhood acute lymphoblastic leukaemia. Lancet 336, 225–229. Lynch, T., Bell, D., Sordella, R. et al. (2004). Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefitinib. N Engl J Med 350, 2129–2139. Munos, B. (2006). Can open-source R&D reinvigorate drug research?. Nat Rev Drug Discov 5, 723–729. Overington, J.P., Al-Lazikani, B. and Hopkins, A.L. (2006). How many drug targets are there?. Nat Rev Drug Discov 5, 993–996. Potti, A., Dressman, H.K., Bild, A., Riedel, R.F., Chan, G., Sayer, R., Cragun, J., Cottrill, H., Kelley, M.J., Petersen, R. et al. (2006a). Genomic signatures to guide the use of chemotherapeutics. Nat Med 12, 1294–1300. Potti, A., Mukherjee, S., Petersen, R., Dressman, H.K., Bild, A., Koontz, J., Kratzke, R., Watson, M.A., Kelley, M., Ginsburg, G.S. et al. (2006b). A genomic strategy to refine prognosis in early-stage non-smallcell lung cancer. N Engl J Med 355, 570–580. Powers, M. and Workman, P. (2007). Inhibitors of the heat shock protein response: Biology and pharmacology. FEBS Lett 581, 3758–3769. Roden, D.M., Altman, R.B., Benowitz, N.L., Flockhart, D.A., Giacomini, K.M., Johnson, J.A., Krauss, R.M., McLeod, H.L., Ratain, M.J., Relling, M.V. et al. (2006). Pharmacogenomics: Challenges and opportunities. Ann Intern Med 145, 749–757. Sams-Dodd, F. (2005). Target-based drug discovery: Is something wrong?. Drug Discov Today 10, 139–147. Weinshilboum, R. (2003). Inheritance and drug response. N Engl J Med 348, 529–537. Whitehurst, A.W., Bodemann, B.O., Cardenas, J., Ferguson, D., Girard, L., Peyton, M., Minna, J.D., Michnoff, C., Hao, W., Roth, M.G. et al. (2007). Synthetic lethal screen identification of chemosensitizer loci in cancer cells. Nature 446, 815–819. Youngman, P., Tepper, R. and Moore, J. (2001). Genomics technologies and strategies for identification of novel anti-infective agents. Curr Clin Top Infect Dis 21, 366–390.
342
CHAPTER 28
■
Role of Genomics in Drug Discovery and Development
Zheng, C.J., Han, L.Y., Yap, C.W., Ji, Z.L., Cao, Z.W. and Chen, Y.Z. (2006). Therapeutic targets: Progress of their exploration and investigation of their characteristics. Pharmacol Rev 58, 259–279.
(2007). Rebuilding Big Pharma’s business model. http://www.trinity. edu/sbachrac/drugdesign/Drug%20Costs%20Articles/rebuilding_big_pharma.pdf.
RECOMMENDED RESOURCES European Medicines Evaluation Agency http://www.emea.europa.eu/ Food and Drug Administration Guidelines Page http://www.fda. gov/cder/guidance/#clinical%20medicine The Personalized Medicine Coalition http://www.personalizedmedicinecoalition.org/
US Centers for Disease Control Population Genomics Homepage http://www.cdc.gov/genomics/ US National Institutes of Health Animal Model Genomics Homepage http://www.nih.gov/science/models/rat/ US Clinical Trials Listing http://www.clinicaltrials.gov/
CHAPTER
29 Role of Pharmacogenomics in Drug Development Colin F. Spraggs, Beena T. Koshy, Mark R. Edbrooke and Allen D. Roses
INTRODUCTION Fast paced developments in biomedical science continue to increase understanding of the molecular complexities of biological pathways and offer new possibilities for the prevention, treatment and cure of common, serious diseases. These opportunities are accompanied by the challenges of applying these new discoveries to produce affordable, safe and efficacious new therapies for the combat of disease. In this chapter, we review current application of genomic strategies in drug development that are having an impact across the drug development continuum and discuss how these genetic approaches re-define clinical trial design to incorporate non-traditional end points and transform pharmaceutical research and development productivity. Pharmacogenomics describes genomic/genetic investigations of drug response using relationships between DNA and other genomic (RNA) analytes and drug response (Phillips and Van Bebber, 2005; Weinshilboum and Wang, 2004). In this chapter, we use the term pharmacogenetics to also describe specifically the relationship between inherited genetic (DNA) variation and drug response (see also Chapter 27). The past decade has witnessed a decline in pharmaceutical productivity, where despite year after year increases in research and development expenditure, there has been a decline in the submission and approval of new, innovative drugs to treat disease and benefit patients. Recent estimates place successful pharmaceutical Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
launch rates at 10% overall of those drugs entering clinical development, with reasons for failure ascribed to adverse drug reactions (10%), lack of efficacy (30%) and unfavorable pharmacokinetic profile and/or metabolic properties (39%) (Towse, 2003). The highest failure rates are observed during Phases II (65%) and III (40%) which correspond to the most expensive phases for clinical development cost (current estimates $200 M). Poor productivity has been attributed to the pharmaceutical industry’s fascination with blockbuster drugs; rather, the rising development costs are a consequence of pipeline attrition due to the termination of drugs during clinical development through failure to achieve safety and efficacy target profiles. This is a shift from the early 1990s, when pipeline attrition was in the early phase of drug development (mostly Phase I) chiefly to poor pharmacokinetics and bioavailability. Efficacy failures are in part due to poor predictive animal models for some diseases and also to the inability to measure clinical endpoints that are relevant to treating the disease, rather than endpoints that reporting the physiology or pharmacodynamics of drug treatment (Littman and Williams, 2005). For example in Chronic Obstructive Pulmonary Disease (COPD), a chronic progressive disease of lung inflammation and damage coupled with deterioration of lung function, forced expiratory volume in one second (FEV1) is an established clinical endpoint for the evaluation of new chemical entities (NCEs) in COPD. However, studies have demonstrated that patients with equivalent FEV1 values Copyright © 2009, Elsevier Inc. All rights reserved. 343
344
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
have different lung pathologies (emphysema severity) determined by high resolution CT scanning and different responsiveness to administration of the inhaled corticosteroid beclomethasone (Gelb et al., 1996). This suggests that FEV1 does not correlate with more sensitive disease markers and may obscure the efficacy of NCEs in COPD clinical trials, indicating that more sensitive, disease relevant measurements are required to document response phenotypes. Adverse drug reactions (ADRs) also contribute to attrition, occurring either during drug development or after approval and launch. A number of successfully launched products including Rezulin (troglitazone, Sankyo, GSK, Pfizer) Lotronex (alosetron, GSK), Baycol (cerivastatin, Bayer), Tysabri (natalizumab, Elan, Biogen) and Vioxx (rofecoxib, Merck) have been withdrawn from the market following serious side effects. Due to their low incidence, these side effects were not observed during extensive clinical development but required access to post-marketing patient populations to become manifest. Review data suggest that 51% of medicines can cause serious ADRs and result in 5% of all hospital admissions and 700,000 injuries or deaths per year (Einarson, 1993; Lazarou et al., 1998). Furthermore, of 499 drug approvals since 1980, 21% have required post-launch changes in dose, of which 79% have been safety-related (Cross et al., 2002).
DRUG DEVELOPMENT CRITICAL PATH Safety remains a prime concern for regulatory authorities. Recognizing this concern and the productivity decline in the provision of new medicines, regulatory authorities have produced white paper discussion documents: “Innovation-Stagnation: Challenge and Opportunity on the critical path to new medicines” from the FDA (FDA, 2004) and Roadmap to 2010 from the EMEA (2004) which outline opportunities to support the development of new medical products. Pharmacogenomics is identified as a key opportunity in these strategy documents. To facilitate incorporation of pharmacogenomic approaches, FDA has also developed regulatory and non-regulatory guidance (FDA, 2005a) to support formal and non-formal review of genomic data through Voluntary Genomic Data Submissions (VGDS) and established the Interdisciplinary Pharmacogenomic Review Group (IPRG). Similar activities are being conducted in Europe through the EMEA Pharmacogenomics Briefing Meetings and VGDS discussions are held jointly by the United States and EU Regulators. Regulatory authorities are establishing an environment conducive to the integration and application of genomic technologies in support of NCE development. Regulatory encouragement of the use genomic technology is validated by recently FDA approved integration of clinically relevant pharmacogenomic data into product labels (Table 29.1). This information and usage guidance incorporating pharmacogenomics has been incorporated into some product prescribing information. At this stage, there are few examples of clinical decision making that utilizes pharmacogenomics. Nevertheless,
pharmacogenomic investigation is occurring more rapidly upon the introduction of new medicines and examples where pharmacogenomic publications precede the registration and approval of new medicines, as shown in Figure 29.1 (Krejsa et al., 2006). This trend suggests pharmacogenomics will become and extensively applied tool to inform clinical decision-making and match the right medicine to the right patient.
DRUG DEVELOPMENT ECONOMICS The typical cost of research and development for a NCE is $880–900 million and an average of 8 years to develop (Rawlins, 2004). The costs of product development and marketing have more than doubled during the last decade and high attrition rates in expensive pipeline development and subsequent promotion, make failures expensive to fund. Long term, these factors will have healthcare impacts of a failing return on pharmaceutical investment and the delivery of fewer new medicines to patients. The current drug development paradigm is not able to keep pace with the opportunities of basic science innovation and changes are required to deliver affordable, novel and effective drug therapies that will improve outcomes in common serious diseases. The goals are clear: decrease development costs, focus on the patient and work with the regulatory authorities in managing expectation around co-development of genomic based drugs and diagnostic tests. The application of genomics for drug development will lead to enhanced benefit/risk predictions, dose selection and exposure–response understanding, as well as more accurate prediction of efficacy in early clinical development and improved management of the risk profiles of potential medicines. Genomics will reduce attrition and increase drug development productivity. The high potential of pharmacogenomics to improve medicine approval and usage and healthcare is exemplified in many reports, including this one; however, studies using established economics-based resource allocation frameworks have shown that corresponding data to measure the impact of pharmacogenomics on clinical practice and health outcomes has not been collected (Phillips and Van Bebber, 2005) and there is incomplete documentation of the associations of metabolism, drug response and clinical outcome for many marketed drugs (Evans and Relling, 2004). These shortcomings limit currently the measurement of the value of pharmacogenomic interventions in healthcare.
METHODS FOR IDENTIFICATION OF GENETIC CLASSIFIERS Pharmacogenomics (PGx) studies at any phase in drug development involve the generation of genetic classifiers or markers. These genetic classifiers are used to predict pharmacokinetic, efficacy and/or safety variability, depending on the questions being addressed in the PGx study. Genetic classifiers are generated through one of two approaches, a candidate gene approach
Methods for Identification of Genetic Classifiers
TABLE 29.1
■
345
Integration of pharmacogenomic data into product labels
Gene
Drug
Indication
Details
Comment
HER2
Herceptin (trastuzumab)
Metastatic breast cancer
Clinical benefit limited to patients with tumor HER2 overexpression (20–30% of breast tumors). Label requires that patients are tested for HER2 expression in tumor prior to receiving herceptin treatment.
Approved September 1998
CYP2D6
Strattera (atomoxetine)
Attention Deficit Hyperactivity Disorder (ADHD)
Correlation of CYP2D6 poor metabolizer genotypes with increased drug levels, adverse event incidence and efficacy.
Change to existing label, January 2003
TPMT
Thiopurines
Acute lymphoblastic leukemia
TPMT is the enzyme that inactivates 6-mercatopurine (6MP), the prodrug for the thioguanine nucleotide (TGN) active metabolites. TPMT deficiency results in increased 6MP availability and increased TGN formation, which may lead to increased incidence of Grade IV neutropenia and thrombocytopenia in affected patients. Three variant alleles, TPMT *2, 3A and 3C account for 95% of all dysfunctional alleles and have a high concordance with reduced TPMT enzyme activity (Schaeffeler et al., 2004; Yates et al., 1997). Furthermore, review of patients with documented 6MP toxicity demonstrated that TPMT genotyping could identify patients at risk and enable integration of TPMT genotyping into the clinical management of 6-MP therapy (Evans et al., 2001).
Change to existing label, July 2003
EGFR
Erbitux (cetuximab)
Metastatic colorectal cancer
Clinical benefit limited to patients with tumors shown to be positive for Epidermal Growth Factor Receptor (EGFR) expression. Immunohistochemical evidence of tumor EGFR positivity is required before treatment.
Approved February 2004
UGT1A1
Camptosar (irinotecan)
Colorectal cancer
Irinotecan is a prodrug that is converted by esterases to its active metabolite SN-38. Increased SN-38 levels are associated with increased efficacy and side effects incidence (Grade III-IV diarrhea and/or neutropenia). The predominant route of SN-38 elimination is glucuronidation by the UDP glucuronyltransferase, UGT1A1and patients with the highest ratios of SN-38 parent /SN-38 glucuronide have been shown to have the highest incidence of side effects (Gupta et al., 1994). Clinical studies have shown association of common UGT1A1 polymorphisms with the risk of irinotecan induced diarrhea and/or neutropaenia. A prospective study confirmed association of the UGT1A1*28 variant with these side effects, where all patients suffering neutropenia had at least one UGT1A1*28 allele (Innocenti et al., 2004). Camptosar label changed to include information that patients with reduced UGT1A1 activity (determined by UGT1A1*28 genotyping) have increased risk of neutropenia. Heterozygous patients may also have increased risk. A reduced initial dose should be considered. FDA have approved a UGT1A1 genotyping test (Third Wave Technologies, Invader assay) for the detection of *1(TA6 repeat) and *28 (TA7 repeat) alleles of the UGT1A1 gene as an aid in the identification of patients with greater risk for decreased UGT enzyme activity (FDA, 2005b).
Change to existing label, November 2004
346
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
20 Somatropin
Years to first pharmacogenetic report
18 16 14
β-agonist IFNα
12 Fluoxetine 10
IFNβ
8
Irinotecan
6
Alteplase Rituximab
4
Anakinra
2
Cetuximab. Gefitinib
Pravastatin Trastuzumab
0 1980
1985
1990
1995 Approval date
2000 Erlotinib. 2005 Panitumumab*
Figure 29.1 Pharmacogenomic investigation is occurring more rapidly upon the introduction of new medicines (from Kresja et al., 2006).*Pantiumumab was approved by the EMEA in December, 2007. IFN: interferon.
or through a whole genome scan (WGS) approach. In the candidate gene approach, a set of genes is derived based on the hypothesis in question. The candidate gene lists are generated ad hoc based on the question being addressed to include drug target and mechanism pathway genes. In addition, a comprehensive panel of genes may be developed to interrogate issues that are fairly common in drug development such as variable pharmacokinetics. A straightforward approach is to use a set of tagging SNPs (tSNPs) that will capture the preponderance of common genetic variation (5%) throughout each selected gene (Carlson et al., 2004). In addition, functional SNPs for each selected are added based on a literature survey. Following the application of these candidate gene SNP panels, simple analyses largely focused on individual variations are carried out to generate the genetic classifier. To address questions of PK variability that occur usually in Phase I studies, a standardized, systematic approach has been introduced in GSK clinical development programs. Absorption, distribution, metabolism and excretion (ADME) genes play a central role in drug pharmacokinetics and pharmacodynamics (PK-PD). For most compounds in phase I of drug development, factors contributing to the differences in pharmacokinetic phenotypes are not fully understood. A pilot study to examine the utility of a comprehensive panel of ADME genetic markers has been initiated in early Phase studies in the GSK drug development pipeline. This provides an opportunity to gain additional human genetic insight into the pharmacokinetics of NCEs. The panel consists of 134 ADME enzymes, transporters and other genes involved in drug ADME. From these selected genes, approximately 1500 polymorphisms have been identified
for the marker panel, which includes all known functional polymorphisms that have described ADME phenotypes and includes selected tagging SNPs (tSNPs) that capture all known common variation in the ADME genes. Application of an ADME panel to assets where variability in exposure is outside of the accepted range (similar to bioequivalence parameters) will build knowledge and support dose selection rationale, particularly where PK/exposure is influenced by multiple genes with ADME pathway redundancy. With ever improving technology, WGS are becoming feasible for pharmacogenetic studies. These WGS provide a hypotheses free approach and are carried out using the commercially available chips targeting either 100,000 SNPs or 500,000 SNPs or even 1000,000 SNPs coverage of the genome. Although these WGS are becoming more tractable, they are still a number of hurdles that need to be overcome, including problems with high throughput analysis of such a large number of markers, multiple testing and marker–marker interaction assessment and new methodologies are being developed to address these challenges (Bowman and Delrieu, 2005; Delrieu and Bowman, 2006; Ritchie and Motsinger, 2005) (see Chapter 8).
PHARMACOGENOMICS IN THE DRUG DEVELOPMENT PIPELINE Figure 29.2 summarizes the opportunities for pharmacogenomics applications to better understand drug performance and improve productivity in the drug development pipeline. This can occur across all of the phases of drug development. Preclinical
Pharmacogenomics in the Drug Development Pipeline
Pharmacogenomic applications in the drug discovery and development “pipeline” Responder population identification
Target selection pathway analysis
Risk management post marketing surveillance
Early risk discharge (tolerability)
Target lead
Candidate
Predictive toxicology
FTIH
PK/dosing
POC Phase 2
Phase 3
Regulatory submission
Launch and LCM
Responders: stratify/enrich
Figure 29.2 Opportunities for application of pharmacogenomics during the drug discovery and development process. PK: pharmacokinetics. FTIH: first time in humans. POC: proof of concept. LCM: lifecycle management.
opportunities include application of genomics (genetic and gene expression profiling) to enhance target selection, through disease understanding and susceptibility gene screening (Roses et al., 2005) and to predict potential toxicity of chemotypes. Pharmacokinetic profiling in human volunteers conducted in Phase I studies and subsequently in Phase II studies in disease patients, provides early information of drug exposure that may influence subsequent assessments of safety and efficacy and dose selection (see “Drug Exposure Pharmacogenetics to Tune Efficacy and Safety Profiles”). Phase II studies are conducted in patients and are concerned with demonstration of drug efficacy and building a database of drug safety in the relevant patient population. Efficacy Proof of Concept (POC) studies, usually conducted in Phase II, are key decision making clinical trials, where based on achievement of the target product profile (predefined efficacy level with acceptable tolerability and side effect profiles), the NCE is progressed to larger, more expensive pivotal Phase III studies, aimed to demonstrate the safety and efficacy consistent with a registerable product. In addition to demonstrating an acceptable safety profile, the decision to progress NCEs to Phase III pivotal studies are based on three trial outcomes: (i) molecules demonstrating robust efficacy, which are progressed, (ii) molecules with no discernable efficacy, which are not progressed, (iii) molecules where the efficacy target is not achieved in the study population, but with a segment of the treated study population showing efficacy. Without the ability to classify and differentiate the responding subjects and so enable further molecule evaluation in a subset of clinical patients, the outcome of the progression decision in (i) is the same as for outcome (ii); the NCE is not progressed. Recent examples using pharmacogenetics to address outcome (iii) and also to identify efficacy response hypotheses for evaluation in subsequent development studies have been
■
347
generated in the obesity area (Roses, 2004; Spraggs et al., 2005) and also for rosiglitazone in Alzheimer’s Disease (AD) (Risner et al., 2006), as described below (see “Efficacy Pharmacogenetics – Here and Now”). Phase III trials provide the clinical evidence of safety and efficacy determined in sufficiently large numbers of appropriate patients to enable submission for regulatory approval of the drug for use as therapy. In addition to providing increased sample size for pharmacogenetic analyses through larger patient numbers, Phase III studies also enable the testing of genetic hypotheses that have been determined from earlier studies. Prospective Phase III study designs may accommodate pharmacogenetic considerations through genotype based stratification or enrichment of patient subgroups (see “Efficacy Pharmacogenetics – Here and Now”). Drug exposure to large numbers of patients for extended periods occurs in many Phase III trials, and low incidence safety signals may become manifest in sufficient numbers to enable retrospective pharmacogenetic investigation, leading to additional understanding and possible management of safety events (see “Investigation and Management of Safety in Clinical Trials” and “No Samples, No Science”). As described earlier (see “Challenge and Opportunity”), serious, but very low incidence side effects may not be manifest until after drug approval and launch and extensive exposure in the disease population. Opportunities exist to use pharmacogenetics to manage risk as a component of post-marketing surveillance strategies, but such approaches are currently in their infancy (EFPIA, 2005) and will likely require changes in regulatory and healthcare practices to enable their fuller application. Clinical study designs that incorporate pharmacogenetic objectives and methods will lead to faster and improved implementation of pharmacogenetic markers into clinical development and practice. Generally, the stages of pharmacogenetic marker evaluation and validation parallel the clinical development process. Pharmacogenetic marker hypotheses may be generated from preclinical data (e.g., disease susceptibility genetic data and metabolic profiling using in vitro recombinant ADME gene cell systems) or early clinical data (e.g., Phase I pharmacokinetic data or Phase IIa efficacy data). Marker hypothesis generation may also be achieved by retrospective patient stratification by candidate genotypes or alleles in “all comer” (not-genetically randomized) Phase I and II studies (see CYP2C19 genotype on incidence of Tykerb induced rash and diarrhea, “Investigation and Management of Safety in Clinical Trials”). Subsequent marker hypothesis testing and validation can be conducted in follow-on studies. Hypothesis testing can be achieved through different study designs, including prospective genotype or allele enrichment of study patient population in Phase III studies (see rosiglitazone and APOE 4 allele status for efficacy response in AD, “Efficacy Pharmacogenetics – Here and Now”), or by prospective comparison of genetic marker impact on clinical outcome in studies following successful launch which attempt to further support effective clinical management of serious side effects (see abacavir and comparison of HLA B*5701 marker
348
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
status determination, or not, on hypersensitivity reaction (HSR) incidence, “Investigation and Management of Safety in Clinical Trials”). As shown later in this chapter, marker evaluation and even validation can be achieved within the established clinical development timeframe.
EFFICACY PHARMACOGENETICS – HERE AND NOW Pharmacogenetics/pharmacogenomics seeks to associate genetic/ genomic variations with differences in the response to drugs in terms of efficacy and safety and by classifying trial subjects and patients using genetic/genomic variation in order to guide drug development decision-making and improve therapeutic outcome for patients. There is a need to generate pharmacogenetic classifiers that are accurate, sensitive, specific and, above all, clinically meaningful for utility in clinical trials and subsequently in healthcare. The focus of efficacy pharmacogenetics is on reducing attrition in drug development. Where the means to classify patients using genetic variation is available, pharmacogenetics can be used to enrich the study population with responders and not enroll those subsets genetically predicted to be poor or inadequate responders. Increased success by selectively enrolling genetically predicted responders in efficacy studies will lead to more successful drug launches, less attrition and, with less drug development failures to pay for, a more successful pharmaceutical pipeline. As illustrated below, this paradigm is happening now. Alzheimer’s Disease is a debilitating disease of the elderly. There are estimated to be more than 30 million sufferers worldwide and with demographic changes leading to an increased aging population, estimates indicate that the number of AD sufferers could triple by 2050. Current pharmacological treatment of AD is based on enhancing cholinergic neurotransmission through use of acetylcholinesterase inhibitors (ACHEi). Clinical evidence shows that ACHEis are not fully effective with estimates of between 20 and 60% of AD patients showing little or no response. The burden of disease places a strong requirement for novel, safe and effective therapies for AD. Based on clinical imaging data linking decline in cerebral glucose utilization with progression in AD (Reiman et al., 1996) and positive cognitive effects in a small pilot study in AD patients (Watson et al., 2005), the insulin sensitizing peroxisome proliferator-activated receptor gamma (PPAR) agonist rosiglitazone has been evaluated in a Phase II, 24 week, placebo controlled, dose-ranging (2, 4 and 8 mg/day) efficacy POC study in mild-to-moderate AD patients (Risner et al., 2006). Based on the previously described robust association of apolipoprotein E4 (APOE4) allele with increased risk and earlier onset of AD (Corder et al., 1993) and the association of APOE4 allele carriage with reduced cerebral glucose utilization in younger, unaffected subjects (Reiman et al., 1996), a prospective study analysis decision was made to genotype consenting patients in this study for common variants in the APOE gene. The pharmacogenetic study population was stratified by APOE4
allele carriage and efficacy outcomes were compared between the two groups (APOE4 positive, 44% of study subjects and APOE4 negative, 56% of study subjects). Following 24 weeks of treatment, rosiglitazone had no statistically significant treatment effect on cognition endpoints when analyzed in the “all-comer” (Intent to Treat) study population (Figure 29.3a). However, a significant interaction between APOE4 allele status and cognition endpoint (ADAS-Cog, p 0.014) was observed. Exploratory analysis showed a nominally significant improvement in cognition endpoints at 24 weeks observed at the highest dose (8 mg) in the subjects who were APOE4 negative (56% of study population) (Figure 29.3b). Furthermore, the corresponding treatment in the APOE4 positive subgroup (44% of the study population) failed to demonstrate cognitive improvement and also showed cognitive decline at the lowest dose of rosiglitazone (2 mg). No conclusions can be drawn about clinical management of patients with AD from this study, but it provided the means to identify a genetically defined population, as a function of APOE4 allele status, who demonstrated a clinically relevant beneficial efficacy response in this novel potential therapy for mild-to-moderate AD. Further Phase III trials that are powered to confirm and distinguish APOE allele specific response to rosiglitazone in AD are now underway. The Phase III program consists of three large studies (approximately 1400 patients recruited per study) evaluating rosiglitazone as monotherapy (one study) and as adjunctive combination with ACHEi’s (two studies). Patients will be recruited globally from North America, South America, Europe, India and Asia. These studies will compare efficacy response in two equally sized, prospectively stratified groups based on APOE 4 allele status ( 4 positive: one or two alleles, 4 negative: no alleles). During eligibility screening, individual patient’s APOE genotype and 4 allele status will be determined from a buccal swab sample which is sent to a central genotyping laboratory, and 4 allele status is used, along with other clinical criteria, to randomize the patient into the study. Inclusion of prospective APOE genotyping has been accommodated in these large Phase III studies with minimal projected impact on enrollment and timelines and modest cost. To our knowledge, this is the first example in a major drug program where a molecule has been progressed based on the ability to pharmacogenetically predict and define (classify) the subset of patients that demonstrate adequate efficacy. This real case study demonstrates the potential for pharmacogenomics to impact attrition in drug development. The means to genetically select the responder subset has converted a drug that would have failed at efficacy POC to a progression opportunity and so provide a potential, novel therapy in a disease of high unmet need and the means to significantly impact attrition in the drug development pipeline. By reducing attrition and increasing drug development success, the economics of drug development are impacted beneficially: there is less failure to pay for and so development costs are reduced and funds can be made available to invest in other drug opportunities. In addition, determination of responder/non-responder subsets in disease populations
Drug Exposure Pharmacogenetics to Tune Efficacy and Safety Profiles
■
349
2.5 2 1.5
DRUG EXPOSURE PHARMACOGENETICS TO TUNE EFFICACY AND SAFETY PROFILES
Clinical improvement
1 0.5 0 0.5
Clinical decline
1 1.5
Baseline
Week 8
RSG 8 mg
2.5 2 1.5 1 0.5 0 0.5 1 1.5 2 2.5 Baseline
RSG 4 mg
Week 8
Week 16 RSG 2 mg
Week 16
8 mg E4
4 mg E4
2 mg E4
8 mg E4
4 mg E4
2 mg E4
Week 24 Placebo
Week 24 Pbo E4 Pbo E4
Figure 29.3 Efficacy of Rosiglitazone in a genetically defined population with mild-to-moderate alzheimer’s Disease. Reproduced from Risner et al., 2006. (a) Effects of rosiglitazone on ADAS-Cog scores in the intent to treat (ITT) population. A positive number indicates worsening and a negative number indicates improvement. Change from baseline in ADAS-Cog total scores in the ITT population showed that treatment differences between rosiglitazone and placebo for ADAS-Cog total scores did not reach statistical significance at Week 24 LOCF. (b) ADAS-Cog scores in APOE E4-negative versus APOE E4-positive cohorts. Analysis of interaction between APOE carriage status and ADAS-Cog change from baseline to Week 24 was significant (P 0.014). Subsequent exploratory testing revealed that APOE E4-negative patients, after 24 weeks, showed an improvement that was statistically significant at the highest rosiglitazone dose (8 mg) compared to placebo (P 0.024, not adjusted for multiplicity). APOE E4-positive patients do not show an improvement in cognition but rather showed a decline at rosiglitazone 2 mg (P 0.012; not adjusted for multiplicity).
enables more focus on novel mechanisms for the provision of effective therapy for the “non-responder” populations. Using this paradigm, more drug programs are likely to be successful and the patient and healthcare providers are supported by the provision of affordable, more effective drugs to treat serious common diseases.
Extensive pharmacogenetic data are available to characterize the effects of genes responsible for the ADME of drugs (Evans and McLeod, 2003; Weinshilboum, 2003). Where exposure varies based on genotype, this can be corrected for routinely by genotype based dose selection, leading to reductions in the overall variability in efficacy and safety and a better understanding of drug performance during development. Based on modeling of CYP2D6 drug metabolism from a range of pharmacokinetic examples, it has been estimated that dose adjustments based on CYP2D6 genotype to achieve equivalent systemic drug exposure can be achieved within a fourfold range in dose option within the treated population (Kirchheiner et al., 2005). Assuming the standard indicated dose is 100 units, dose ranging would vary as follows: poor Metabolizers (PM, 7% Caucasian population) 60 units, Intermediate Metabolizers (IM, 40% Caucasian population) 80 units, Extensive Metabolizers (EM, 50% Caucasian population) 160 units and Ultra-Rapid Metabolizers (UM, 24% Caucasian population) 240 units. This simulation shows that the standard “global” dose for a CYP2D6 influenced drug overcompensates for the few individuals that are PM (7%). As a consequence, a majority of the Caucasian population (EM 50% and UM 3%) actually receive a lower dose than indicated by their empirical exposure, that is they are actually receiving a suboptimal dose for efficacy, based on the population risk of overexposure in PM individuals. Further analysis by Kirchheiner and colleagues shows a number of drug examples for various metabolizing enzymes (CYP2D6, CYP2C19, CYP2C9, TPMT and NAT2) where, based on DME influenced exposure, genotype based dose adjustment (sometimes 10× across the population) could be made to ensure the same level of exposure across the study population (Kirchheiner et al., 2005). These simulations have been applied and CYP 2D6 and CYP 2C19 genotype based recommendations have been proposed for a range of marketed antidepressants (Kirchheiner et al., 2001). Managing variability is a key factor in the understanding of drug performance in early studies. Covariate analysis of standard factors (age, gender, weight, disease severity, etc.) is routinely applied to explain variability in pharmacokinetic-pharmacodynamic modeling. Modeling and adjustment for dosage based on genotype will reduce overall study variability and so help identify other factors influencing efficacy and safety variability and support risk management and clearer evaluation of drug performance. An example of this is the identification of common polymorphisms, not only in CYP2C9, but also in VKORC1 (vitamin K epoxide reductase C1, efficacy pathway) that are associated with warfarin response variability (Wadelius et al., 2005). This work shows that the VKORC1 SNPs have a greater impact (30%) on dose than CYP2C9 SNPs (12%) and when combined with patient clinical characteristics (including bodyweight) that explain 60% of variability and provide the basis for
350
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
a planned prospective clinical evaluation of a dosing algorithm to support the rapid and safe attainment of anticoagulant efficacy in patients (www.warfarindosing.com). Preclinical in vitro data in hepatic microsomes characterized CYP2D6 as a major route of metabolism for the GSK investigational compound “A.”Variability in CYP2D6 enzyme activity has the potential to affect drug exposure levels and response (safety and efficacy) when administered alone and in drug combination therapies where the combination drugs are also CYP2D6 ligands. In a group of three small Phase I studies, comprising 35 subjects (volunteers and patients) treated with “A,” four subjects were characterized a priori as pharmacokinetic outliers based on review of pharmacokinetic phenotypic data. Subsequent genotypic analysis of CYP2D6 variation, analyzing nine polymorphisms associated with null/reduced enzyme function (Daly et al., 1996), in DNA collected from study subjects during their participation in the clinical trials and prior to pharmacokinetic analysis. This revealed that all four predefined outlier subjects had CYP2D6 Poor Metabolizer genotypes: two subjects were homozygous for the previously reported *4 null function genotype (G1934A, Caucasian minor allele frequency (MAF) 19%), two subjects had complex genotypes with multiple loss of function alleles, one was heterozygous for *4, *6 (delT1795, null function, Caucasian MAF 1%) and *10 (C188T, reduced function, Caucasian MAF 22%) alleles and one was heterozygous for *3 (delA2637, null function, Caucasian MAF 2%) and *9 (delAGA, reduced function, Caucasian MAF 2%) alleles. Subsequent Phase I studies for other analogs of “A” will prospectively include a cohort of CYP2D6 Poor Metabolizer subjects recruited based on genotype, to enable evaluation of the potential impact of increased exposure through impaired metabolism by CYP2D6 and inclusion of genotype as a covariate in pharmacokinetic modeling. The GSK drug Tykerb/Tyyerb (lapatinib) is approved in more than 20 countries for the treatment of patients with advanced or metastatic breast cancer whose tumours overexpress HER2/ErbB2 and who have received prior therapy including an anthracycline, a taxane and trastuzumab and is being developed for the treatment of breast cancer and other solid tumors (Nelson and Dolder, 2006). Preclinical in vitro studies showed that Tykerb was metabolized predominantly by CYP3A4 and CYP3A5 genes and to a lesser extent, CYP2C19 and that it interacted with the transporters MDR1 (ABCB1) and BCRP (ABCG2). During early phase studies in volunteers and patients, interindividual variation of Tykerb plasma concentrations was observed. In addition, 15% of study subjects experienced rash and diarrhea side effects. A retrospective pharmacogenetic study was undertaken using the above five genes evaluated in DNA samples (107 Caucasians, comprising 34 patients and 73 healthy volunteers) collected prospectively during eight clinical trials (Zaks et al., 2006). Single point (genotypic and allelic) association analyses were performed on 246 single nucleotide polymorphisms located within 10 Kb of the five genes. Whilst there was no evidence for association of side effects for ABCG2, ABCB1, CYP3A4 or CYP3A5, 22 SNPs in CYP2C19 showed significant association with incidence of rash and six of these
SNPs also showed significant association with incidence of diarrhea. Notably, three out of three subjects (two healthy volunteers, one patient) homozygous for the CYP2C19 *2 allele experienced both rash and diarrhea. The ability to investigate associations in the integrated setting of human subjects in clinical trials has complemented and extended the preclinical, in vitro data and provide a hypothesis that CYP2C19 metabolism is likely more important in the development of Tykerb induced side effects than other drug metabolizing genes. No conclusions can be drawn about the clinical management of Tykerb treated patients from this exploratory study, however, subject to further clinical evolution, identification of patients without this allele may allow clinicians to treat the cancer more aggressively with higher doses and less risk of side effects. Further work is ongoing in additional, follow up clinical trials to confirm these exploratory findings and determine the magnitude and clinical significance of these associations and their potential application in the management of side effects during chronic Tykerb treatment. These examples demonstrate the feasibility and provision of decision making utility of pharmacogenetics in early studies with small subject numbers, in contrast to previous perspectives (Cardon et al., 2000). Genotype-based dose selections will support the regulatory imperative to make more informed decisions around dose selection for therapy. Of course, application of genotype-based dose selection into therapy would require a sea change in clinical practice and pharmaceutical provision, not the least of which requiring more than one tablet strength and provision of therapy on a mg/kg basis, as is done in animals, children and cancer patients receiving chemotherapy. The benefits of such change would likely be more molecules making it into healthcare and better management of efficacy and safety profiles for patients.
INVESTIGATION AND MANAGEMENT OF SAFETY IN CLINICAL TRIALS In addition, Phase II studies often provide adequate population exposure to demonstrate common safety and tolerability issues. Genetic classifiers have been used successfully to explain and reduce risk associated with potential safety findings in Phase II studies and so enable the NCE to progress to further clinical studies. The previously reported pharmacogenetic studies on the coronary artery restenosis inhibitor, Tranilast (Danoff et al., 2004; Xu et al., 2004) demonstrated the utility of markers at UGT1A1 to characterize and explain the hyperbilirubinemia observed in patients treated with Tranilast as the benign and reversible condition Gilbert’s syndrome. These markers have subsequently been used successfully to discharge potential elevated bilirubin risk in two further GSK investigational drugs during Phase II (Figure 29.4, data for one of these compounds). Whilst these results are exploratory, generated from studies with small subject numbers and not balanced for genotypes, they provide confidence and hypothesis generation for integration and evaluation in follow up studies on these compounds. Similar data generated by
Investigation and Management of Safety in Clinical Trials
2.2
2.2
2.0
2.0
1.8
1.8
1.6 1.4 1.2 1.0 0.8 0.6
A,A A,G G,G
0.4 0.2
351
Treatment Group (N 244)
Treatment Phase Bilirubin
Treatment Phase Bilirubin
Placebo Treatment Group (N 168)
■
1.6 1.4 1.2 1.0 0.8 0.6
A,A A,G G,G
0.4 0.2
0.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
0.0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2
Pre-Treatment Phase Bilirubin
Pre-Treatment Phase Bilirubin
Adjusted Means of Standardised Bilirubin (on-treatment phase) 1 0.95 0.9 Adjusted Mean
0.85 0.8 0.75 0.7 0.65 0.6 0.55 0.5 G,G
A,G
A,A
Genotype Marker
Placebo
Drug
Figure 29.4 Elevated bilirubin in a Phase II drug trial explained by UGT1A1 polymorphism associated with Gilbert’s Syndrome. (a) Pre- and post-treatment bilirubin values for placebo and GSK investigational drug treatment for individual subjects coded by UGT1A1 RS887829 genotype. Data taken from a Phase II clinical study with a GSK investigational drug “B.” Plasma bilirubin levels were measured at baseline and during the study using standard clinical chemistry methods. Plots show individual patient bilirubin normalized values measured at pre-treatment (x axis) and during treatment (y axis) for placebo study arm (left panel, n 168) and GSK investigational drug treatment arm (right panel, n 244). Values that occur in the upper left quadrant relate to individuals with increased bilirubin values above the standard normal level following treatment. Patients were genotyped using a UGT1A1 SNP (RS887829), which has been shown to be in absolute linkage disequilibrium with the reported 7/7 TA repeat polymorphism. A significant association between RS887829 genotype and bilirubin levels was observed (p 0.001). Furthermore, comparison of placebo and treatment panels in above figure (individual data points coded according to RS887829 genotype) illustrates a higher proportion of patients receiving GSK investigational drug had the RS887829 A, A genotype than placebo, suggestive of a pre-existing condition of Gilbert’s Syndrome, more than an effect of the investigational drug. (b) No difference in bilirubin levels between placebo or GSK investigational drug “B” treatment within UGT1A1 RS887829 genotypes. Using standardized bilirubin values as a dependent variable, analysis of covariance showed no interaction with treatment (p 0.423), but significant interactions were observed for pre-treatment bilirubin (p 0.001) and UGT1A1 RS887829 genotype (p 0.001). Furthermore, as shown in the figure above, treatment by genotype interactions were not significant (p 0.137).
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
Bristol Myers Squibb has resulted in inclusion of UGT1A1 *28 genotype information in a product information sheet to provide context for hyperbilirubinemia observed in some patients treated with HIV antiretroviral Atazanavir (Bristol Myers Squibb). Abacavir is a cornerstone of highly effective combination therapy for the management of HIV infection. Whilst efficacious, approximately 5% of abacavir-treated patients experience a hypersensitivity reaction (HSR) that may be life-threatening. Genetic classifiers of some of the hypersensitivity risk have been identified, including SNPs in the HLA B*57 region (Hetherington et al., 2002; Mallal et al., 2002) and confirmed in some, but not all subsequent studies comprising different ethnic populations (Hughes et al., 2004). Whilst clinical awareness and early signs monitoring of hypersensitivity risk provide good clinical management and adequate maintenance of the benefit/ risk profile for abacavir, these data enabled the opportunity to evaluate the potential for genetic classifiers to improve the management of hypersensitivity risk and reduce incidence to below 5%. A study was conducted by GSK to prospectively investigate the relevant genetic markers by comparing clinical outcome (incidence of hypersensitivity) in groups randomized to standard care (all patients included) and standard care plus prescribing based on genotype (HLA B*5701 marker positive patients excluded) (Mallal et al., 2008). Prospective HLA B*5701 screening reduced the incidence of clinically diagnosed HSR and eliminated HSR confirmed by skin patch testing. Furthermore, when ascertainment of HSR cases is supplemented with skin patch testing, the association between HLA B*5701 and HSR appears to be generalizable to other ethnic populations (Saag et al., 2008).
OTHER GENOMIC METHODS: RNA INTERFERENCE TO DIRECT DRUG USAGE Response classifiers are not limited to genetic variation. Tykerb (Lapatinib) is being developed for the treatment of tumors that are regulated through ERbB2 and EGFR tyrosine kinase receptor pathways, with breast cancer as the major indication for registration (Nelson and Dolder, 2006). The presence of estrogen receptor (ER), progesterone receptor (PR), and bcl–2 in tumor samples is associated with a lack of biological and clinical response after only 10–14 days of treatment. Furthermore, the presence of ER/PR has been shown to be a negative predictor of response to Tykerb in two large phase II breast cancer studies. Immunohistochemistry of breast tumors from patients who became resistant to Tykerb showed a switch from ER /PR to ER+/PR. Therefore, the role of ER alpha in the development of resistance to Tykerb was determined in a target-selective manner using RNA interference (RNAi) (see Chapter 16). The results of this study were considered to have high clinical interest since combinations of anti-ER along with ErbB inhibitors (like Tykerb) might delay or prevent the onset of resistance to Tykerb. RNAi was used to determine the role of ER signaling
in the development of Tykerb resistance in ErbB2-overexpressing breast cancer cell lines (Xia et al., 2006). Specifically, this technology was used to investigate whether molecular knockdown of ER gene expression could convert Tykerb-resistant BT474 cells to become Tykerb-sensitive. Tykerb-resistant ErbB2overexpressing BT474 cell lines showed a conversion from ER to ER, and therefore reflected the situation seen in the clinic. RNAi effector reagents (siRNAs) were generated in Tykerbsensitive breast carcinoma cells, and then tested in Tykerb-resistant cells. It was found that knockdown of ER lead to increased cell death of Tykerb-resistant BT474 breast carcinoma cells. More profoundly, ER knockdown was found to lead to a reversing of the resistance of these cells to Tykerb treatment (Figure 29.5). These results have made a significant impact on the selection of patients in ongoing and planned phase II/III registration trials for Tykerb. The clinical protocol for the inflammatory breast cancer phase II trial has been amended, where ER/PR expressors will now be excluded. If tumor biopsies convert to ER+/PR+ then these patients will be offered a combination therapy involving Tykerb and an anti-ER therapy.
“NO SAMPLES, NO SCIENCE” A key pre-requisite to successful pharmacogenomic study is the prospective collection of appropriate samples and data, even in situations where the study phenotype(s) are not yet known. During the conduct of clinical trials it is necessary to collect all relevant data to enable full phenotype description and determination of covariates (gender, body weight, disease severity,
Gated Cells (%)
352
100 80 60 40 20 0 Cells
500nM lapatinib
siPool siPool500nM lapatinib
ER
ER500n Mlapatinib
Treatment Sub2N
G0/G1
S
G2/M
4N
Figure 29.5 siRNA knockdown of ER in Tykerb-resistant BT474 cells restores Tykerb efficacy. Anti-tumor effects were measured as increased cell apoptosis. Columns show percent of cells residing in five stages of the cell cycle (color coded as per legend). Cell cycle analysis was conducted using FACS analysis and cell apoptosis was measured as the percent of cells in the sub2N phase. In resistant BT474 cells, Tykerb (500 nM lapatinib) had little effect on cell apoptosis rate (sub2N column 2), whilst selective knockdown of ER by siRNA increased apoptosis to 60% (sub2N column 5). Furthermore, combination of ER siRNA plus Tykerb (50 M lapatinib) increased apotosis further (sub2N column 6). These data are representative of three independent experiments.
“No Samples, No Science”
concomitant medications, etc.). Indeed, it is simpler and less costly to collect all relevant data at this time than attempt its collection once the clinical trial has completed. Retrospective studies of clinical data suffer from challenges of collecting accurate data from patients and subjects no longer participating in trial. These include ascertainment bias limited access to the patient, limited recall by the patients and limited verification of data. Collection of samples with appropriate informed consent during clinical studies provides context for the sample collection and is logistically more efficient than retrospective follow up. In addition, successful (maximal) prospective sample collection reduces potential misrepresentation of the study population and sample bias that may occur retrospectively. The strategy to collect and store DNA from clinical trials has enabled pharmacogenetic study of clinical endpoints when new biological information emerges, even after the study has completed, as described below. The PPAR agonist farglitazar has been evaluated for efficacy and safety in a large Phase III program (10 studies) in Type 2 Diabetes patients, but was discontinued for Type 2 Diabetes due to dose-related fluid retention safety concerns in this population. Collection and storage of DNA samples during the studies enabled subsequent pharmacogenetic investigation of PPAR induced fluid retention and edema side effects some time after the studies had been completed. The impetus to conduct this investigation was the emergence of new data implicating epithelial sodium channel pathways in the mechanism of PPAR induced fluid retention and edema (Hong et al., 2003). This work led to selection of a candidate panel of 25 genes based on involvement in epithelial sodium transport. DNA samples were collected from 647 of 1106 patients participating in the Phase III studies (Spraggs et al., 2007). Four hundred and sixty six evaluable, Caucasian subjects were available for analysis and were divided into three groups based on study
TABLE 29.2
■
353
treatment: farglitazar alone (n 329), farglitazar plus glyburide (n 72) and farglitazar plus insulin, (n 140), with incidences of edema adverse events in each study group of 38%, 46% and 31%. Case-control analysis of edema reported as an adverse event in farglitazar treated subjects was conducted. Consistent and significant associations were observed for three polymorphisms in SCNN1B (p 0.05 to 0.0005). These associations were observed for farglitazar combination treatment groups (plus glyburide or insulin), but not in the group receiving farglitazar monotherapy. Additional sequencing of SCNN1B in 207 Caucasian subjects receiving combination therapy identified additional polymorphisms that were also significantly associated with edema (p 0.0005, Table 29.2). These studies have provided clinical pharmacogenetic evidence in distinct study groups to support a pivotal role for epithelial sodium channel regulation in PPAR induced edema and provide insight into mechanisms and possible management of this side effect using the selective sodium channel diuretic, amiloride in preference to loop diuretics, such as furosemide and hydrochlothiazide. This data has provided hypotheses for the further evaluation and possible management of edema side effects in future PPAR agonist studies in a range of disease indications. This work was only possible due to the collect and storage of DNA samples and data for subsequent investigation, after the clinical trials had completed. The pharmacogenomic model employed at GSK has established default, consented DNA collection and extraction in all Phase I, II, III and IV trials, with selective collection of other tissues (plasma, serum, urine, etc.) and phenotypes for biomarker studies, including imaging. Identification of pharmacogenomic opportunities (efficacy and safety) are identified through interactions of clinical project teams with genetics research physicians and scientists to support clinical development decision making.
SCNN1B variants associate with farglitazar induced fluid retention and edema in Type 2 diabetic patients Farglitazar /glyburide (33 cases, 39 controls)
Farglitazar /insulin (44 cases, 96 controls)
Genotypic Chi Sq p
Genotypic Chi Sq p
Allelic Chi Sq p
Global ID
Position relative to ATG
Gene region
MAF
Allelic Chi Sq p
rs2887481
–5724
5 flank
0.342
0.0136
0.0175
ss46565647
–3817
5 flank
0.032
0.00086
0.00096
ss46565649
–3771
5 flank
0.035
0.0008
0.0009
ss46565622
–786
5 flank
0.034
0.0006
0.0007
ss46565623
–454
5 flank
0.034
0.0006
0.0007
rs250563
65640
Exon 5
0.065
0.02821
0.3695
rs889299
68275
Intron 5
0.222
0.0029
0.0005
rs2303157
69608
Intron 7
0.226
0.00003
0.00051
SNPs are presented in genomic order, 5 to coding region; Chi square p values shown where p 0.05 and Hardy Weinberg Equilibrium p value; 0.01, where cell is blank, p 0.05; MAF: minor allele frequency
354
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
Pre-clinical
Rosiglitazone and APOE4 for efficacy in AD
Tykerb and CYP2C19 for rash and diarrhoea AEs
Phase I
APOE as AD progression gene
2C19 and other genes implicated in vitro
Phase II
APOE allele stratification for efficacy
Retrospective stratification for AEs by candidate genes
Abacavir and HLA B*5701 for HSR AE
Phase III
Post-launch and Phase IV
APOE allele enrichment for efficacy
Prospective stratification for AEs by 2C19
Retrospective stratification for HSR by candidate genes
Prospective evaluation of HLA B*57 on HSR incidence
Figure 29.6 Stages of pharmacogenetic marker validation during clinical development programs is illustrated using the examples of (1) Rosiglitazone and efficacy response classification by APOE ε4 allele status (see “Efficacy Pharmacogenetics – Here and Now”), (2) Tykerb and identification of patients at increased risk of rash and diarrhea adverse events by CYP 2C19 genotype (see “Drug Exposure Pharmacogenetics to Tune Efficacy and Safety Profiles”) and (3) Abacavir and identification of patients at increased risk of hypersensitivity reaction (HSR) adverse events by HLA B*5701 marker status (see “Investigation and Management of Safety in Clinical Trials”). Hypothesis marker generating studies are shown in orange and hypothesis marker testing studies are shown in green.
As shown in Figure 29.6, the availability of prospectively collected DNA (via appropriate informed consent) in all clinical studies for pharmacogenetic purposes provides the means to investigate genetic associations with drug response in early studies, generate marker hypotheses for drug safety and efficacy and test and/or validate marker hypotheses in subsequent studies.
CONCLUSIONS Clearly, after a decade of small retrospective studies, pharmacogenetics has come of age and is poised to translate genomics research into gains both in reduction of pipeline attrition and better clinical outcomes. In this review, through use of illustrative examples, we have demonstrated the potential of genomic technologies to influence clinical decision making and reduce attrition in the drug development pipeline. The capability to classify drug response (safety and efficacy) by genomic means enables enrichment
of clinical trials with appropriate subject and patient subsets. Current examples demonstrate the opportunity to manage variability in systemic drug exposure, reduce risk around possible safety concerns and convert drugs that are efficacy failures in the whole population into potential therapies in patient subsets where there is high unmet need. Pharmacogenetics will not just be an additional exploratory part in the drug development process but will become integrated as a clear objective in clinical trials. This becomes imperative as clinical trials access geographically more extensive populations and proof of efficacy studies are spread across different ethnic groups, where information pertaining to the causal variant will provide greater precision than ethnic categorizations (Tate and Goldstein, 2004). Multidisciplinary genomic approaches including genetics, transcriptomics and proteomics and incorporation of non-traditional endpoints such as imaging into clinical trials will improve understanding of the heterogeneity of response to drugs and refine clinical trials. In conjunction with better validated preclinical targets using tools such
References
as transcriptional profiling that predict toxicities used in early preclinical studies, this increases the ability to discover and develop new drugs for common important diseases with a higher probability of efficacy and a lower risk of safety concerns. Improving pharmaceutical productivity in this manner will benefit patients, healthcare providers and payers.
ACKNOWLEDGEMENTS
■
355
GlaxoSmithKline who have generated the data that is illustrated in the examples used in this paper. The authors particularly appreciate the contributions of Anthony Akkari, Cathy Burrows, Kirstie Davies, Olivier Delrieu, Suzanne Edwards, Priti Hegde, David Hosford, Louise Hosking, Arlene Hughes, Intisar Husain, Linda McCarthy, Mike Mosteller, Marc Risner, Ann Saunders, Ganesh Sathe, Neil Spector, Liling Warren and Wenle Xia. We thank the patients and physicians who contributed to the provision of DNA samples and clinical data for these studies.
The authors acknowledge the contributions of many colleagues in Genetics Research and Worldwide Development at
REFERENCES Bowman, C. and Delrieu, O. (2005). Visualisation of gene and pathway determinants of disease. In Quantitative Biology, Shape Analysis and Wavelets (S. Barber, P.D. Baxter, K.V. Mardia and R.E. Wall, eds), University of Leeds, pp. 21–24. Cardon, L.R., Idury, R.M., Harris, T.J., Witte, J.S. and Elston, R.C. (2000). Testing drug response in the presence of genetic i nformation: Sampling issues for clinical trials. Pharmacogenetics 10, 503–510. Carlson, C.S., Eberle, M.A., Rieder, M.J., Yi, Q., Kruglyak, L. and Nickerson, D.A. (2004). Selecting a maximally informative set of single-nucleotide polymorphisms for association analyses using linkage disequilibrium. Am J Hum Genet 74, 106–120. Corder, E.H., Saunders, A.M., Strittmatter,W.J., Schmechel, D.E., Gaskell, P.C., Small, G.W., Roses, A.D., Haines, J.L. and Pericak-Vance, M.A. (1993). Gene dose of apolipoprotein E type 4 allele and the risk of Alzheimer’s disease in late onset families. Science 261, 921–923. Cross, J., Lee, H., Westelinck, A., Nelson, J., Grudzinskas, C. and Peck, C. (2002). Postmarketing drug dosage changes of 499 FDA-approved new molecular entities, 1980–1999. Pharmacoepidemiol Drug Saf 11, 439–446. Daly, A.K., Steen,V.M., Fairbrother, K.S. and Idle, J.R. (1996). CYP2D6 multiallelism. Meth Enzymol 272, 199–210. Danoff,T.M.,Campbell, D.A.,McCarthy, L.C.,Lewis, K.F.,Repasch, M.H., Saunders, A.M., Spurr, N.K., Purvis, I.J., Roses, A.D. and Xu, C.F. (2004). A Gilbert’s syndrome UGT1A1 variant confers susceptibility to tranilast-induced hyperbilirubinemia. Pharmacogenomics J 4, 49–53. Delrieu, O. and Bowman, C. (2006). Visualizing gene determinants of disease in drug discovery. Pharmacogenomics 7, 311–329. Einarson, T.R. (1993). Drug-related hospital admissions. Ann Pharmacother 27, 832–840. Evans, W.E., Hon, Y.Y., Bomgaars, L., Coutre, S., Holdsworth, M., Janco, R., Kalwinsky, D., Keller, F., Khatib, Z. and Margolin, J. et al. (2001). Preponderance of thiopurine S-methyltransferase deficiency and heterozygosity among patients intolerant to mercaptopurine or azathioprine. J Clin Oncol 19, 2293–2301. Evans, W.E. and McLeod, H.L. (2003). Pharmacogenomics – drug disposition, drug targets, and side effects. N Engl J Med 348, 538–549. Evans, W.E. and Relling, M.V. (2004). Moving towards individualized medicine with pharmacogenomics. Nature 429, 464–468. EFPIA (2005). The Innovative Medicines Initiative (IMI) Strategic Research Agenda. Creating biomedical RandD leadership for Europe
to benefit patients and society. Accessed at http://www.ec.europa. eu/research/fp6/pdf/innovative_medicines_sra_final_draft_en.pdf. EMEA (2004). EMEA Road Map to 2010. Accessed at http://www. emea.eu.int/htms/general/direct/roadmap/roadmapintro.htm. FDA (2004). Innovation or stagnation: Challenge and opportunity on the critical path to new medical products. Accessed at http:// www.fda.gov/oc/initiatives/criticalpath/whitepaper.html. FDA (2005a). Guidance for industry: Pharmacogenomic data submissions. Accessed at http://www.fda.gov/cber/gdlns/pharmdtasub.pdf. FDA (2005b). Third Wave Technologies UGT1A1 assay FDA Approval. Accessed at http://www.ons.org/fda/documents/ FDA93005.pdf Gelb, A.F., Hogg, J.C., Muller, N.L., Schein, M.J., Kuei, J., Tashkin, D.P., Epstein, J.D., Kollin, J., Green, R.H. and Zamel, N. et al. (1996). Contribution of emphysema and small airways in COPD. Chest 109, 353–359. Gupta, E., Lestingi, T.M., Mick, R., Ramirez, J., Vokes, E.E. and Ratain, M.J. (1994). Metabolic fate of irinotecan in humans: Correlation of glucuronidation with diarrhea. Cancer Res 54, 3723–3725. Hetherington, S., Hughes, A.R., Mosteller, M., Shortino, D., Baker, K.L., Spreen, W., Lai, E., Davies, K., Handley, A. and Dow, D.J. et al. (2002). Genetic variations in HLA-B region and hypersensitivity reactions to abacavir. Lancet 359, 1121–1122. Hong, G., Lockhart, A., Davis, B., Rahmoune, H., Baker, S., Ye, L., Thompson, P., Shou, Y., O’Shaughnessy, K. and Ronco, P. et al. (2003). PPARgamma activation enhances cell surface ENaCalpha via up-regulation of SGK1 in human collecting duct cells. Faseb J 17, 1966–1968. Hughes, A.R., Mosteller, M., Bansal, A.T., Davies, K., Haneline, S.A., Lai, E.H., Nangle, K., Scott, T., Spreen, W.R. and Warren, L.L. et al. (2004). Association of genetic variations in HLA-B region with hypersensitivity to abacavir in some, but not all, populations. Pharmacogenomics 5, 203–211. Innocenti, F., Undevia, S.D., Iyer, L., Chen, P.X., Das, S., Kocherginsky, M., Karrison, T., Janisch, L., Ramirez, J. and Rudin, C.M. et al. (2004). Genetic variants in the UDP-glucuronosyltransferase 1A1 gene predict the risk of severe neutropenia of irinotecan. J Clin Oncol 22, 1382–1388. Kirchheiner, J., Brosen, K., Dahl, M.L., Gram, L.F., Kasper, S., Roots, I., Sjoqvist, F., Spina, E. and Brockmoller, J. (2001). CYP2D6 and CYP2C19 genotype-based dose recommendations
356
CHAPTER 29
■
Role of Pharmacogenomics in Drug Development
for antidepressants: A first step towards subpopulation-specific dosages. Acta Psychiatr Scand 104, 173–192. Kirchheiner, J., Fuhr, U. and Brockmoller, J. (2005). Pharmacogeneticsbased therapeutic recommendations-ready for clinical practice?. Nat Rev Drug Discov 4, 639–647. Krejsa, C., Rogge, M. and Sadee, W. (2006). Protein therapeutics: New applications for pharmacogenetics. Nat Rev Drug Discov 5, 507–521. Lazarou, J., Pomeranz, B.H. and Corey, P.N. (1998). Incidence of adverse drug reactions in hospitalized patients: A meta-analysis of prospective studies. JAMA 279, 1200–1205. Littman, B.H. and Williams, S.A. (2005). The ultimate model organism: Progress in experimental medicine. Nat Rev Drug Discov 4, 631–638. Mallal, S., Nolan, D., Witt, C., Masel, G., Martin, A.M., Moore, C., Sayer, D., Castley, A., Mamotte, C. and Maxwell, D. et al. (2002). Association between presence of HLA-B*5701, HLA-DR7, and HLA-DQ3 and hypersensitivity to HIV-1 reverse-transcriptase inhibitor abacavir. Lancet 359, 727–732. Mallal, S., Phillips, E., and Carosi, G. et al. (2008).The PREDICT-1 study: a randomized, double-blind trial to determine the clinical utility of HLA-B*5701 pharmacogenetic screening for abacavir hypersensitivity in HIV-infected patients (Study CNA106030). N Engl J Med 358, 568–579. Nelson, M.H. and Dolder, C.R. (2006). Lapatinib: A novel dual tyrosine kinase inhibitor with activity in solid tumors. Ann Pharmacother 40, 261–269. Phillips, K.A. and Van Bebber, S.L. (2005). Measuring the value of pharmacogenomics. Nat Rev Drug Discov 4, 500–509. Rawlins, M.D. (2004). Cutting the cost of drug development?. Nat Rev Drug Discov 3, 360–364. Reiman, E.M., Caselli, R.J.,Yun, L.S., Chen, K., Bandy, D., Minoshima, S., Thibodeau, S.N. and Osborne, D. (1996). Preclinical evidence of Alzheimer’s disease in persons homozygous for the epsilon 4 allele for apolipoprotein E. N Engl J Med 334, 752–758. Risner, M.E., Saunders, A.M., Altman, J.F., Ormandy, G.C., Craft, S., Foley, I.M., Zvartau-Hind, M.E., Hosford, D.A. and Roses, A.D. (2006). Efficacy of rosiglitazone in a genetically defined population with mild-to-moderate Alzheimer’s disease. Pharmacogenomics J 6, 246–254. Ritchie, M.D. and Motsinger, A.A. (2005). Multifactor dimensionality reduction for detecting gene-gene and gene-environment interactions in pharmacogenomics studies. Pharmacogenomics 6, 823–834. Roses, A.D. (2004). Pharmacogenetics and drug development: The path to safer and more effective drugs. Nat Rev Genet 5, 645–656. Roses, A.D., Burns, D.K., Chissoe, S., Middleton, L. and St Jean, P. (2005). Disease-specific target selection: A critical first step down the right road. Drug Discov Today 10, 177–189. Saag, M., Balu, R., and Phillips, E. et al. (2008). High sensitivity of HLA-B*5701 in immunologically-confirmed cases of abacavir hypersensitivity in White and Black patients. Clinical Infectious Diseases, e-publication doi: 10.1086/529382.
Schaeffeler, E., Fischer, C., Brockmeier, D., Wernet, D., Moerike, K., Eichelbaum, M., Zanger, U.M. and Schwab, M. (2004). Comprehensive analysis of thiopurine S-methyltransferase phenotype-genotype correlation in a large population of GermanCaucasians and identification of novel TPMT variants. Pharmacogenetics 14, 407–417. Spraggs, C.F., Pillai, S.G., Dow, D., Douglas, C., McCarthy, L., Manasco, P.K.,Stubbins, M. and Roses,A.D. (2005). Pharmacogenetics and obesity: Common gene variants influence weight loss response of the norepinephrine/dopamine transporter inhibitor GW320659 in obese subjects. Pharmacogenet Genomics 15, 883–889. Spraggs, C., McCarthy, A., McCarthy, L., Hong, G., Hughes, A., Lin, X., Sathe, G., Smart, D., Traini, C., Van Horn, S., Warren, L., and Mosteller, M. (2007). Genetic variants in the epithelial sodium channel associate with oedema in Type 2 Diabetic patients receiving the PPARγ agonist Farglitazar. Pharmacogenet Genomics 17, 1065–1076. Tate, S.K. and Goldstein, D.B. (2004). Will tomorrow’s medicines work for everyone?. Nat Genet 36, S34–42. Towse, A. (2003). CMR workshop on regulating personalized medicine. CMR International Institute for Regulatory Science. Accessed at http://www.cmr.org/institute/pdf/RD39.pdf. Wadelius, M., Chen, L.Y., Downes, K., Ghori, J., Hunt, S., Eriksson, N., Wallerman, O., Melhus, H., Wadelius, C. and Bentley, D. et al. (2005). Common VKORC1 and GGCX polymorphisms associated with warfarin dose. Pharmacogenomics J 5, 262–270. Watson, G.S., Cholerton, B.A., Reger, M.A., Baker, L.D., Plymate, S.R., Asthana, S., Fishel, M.A., Kulstad, J.J., Green, P.S. and Cook, D.G. et al. (2005). Preserved cognition in patients with early Alzheimer disease and amnestic mild cognitive impairment during treatment with rosiglitazone: A preliminary study. Am J Geriatr Psychiatry 13, 950–958. Weinshilboum, R. (2003). Inheritance and drug response. N Engl J Med 348, 529–537. Weinshilboum, R. and Wang, L. (2004). Pharmacogenomics: Bench to bedside. Nat Rev Drug Discov 3, 739–748. Xia, W., Bacus, S., Hegde, P., Husain, I., Strum, J., Liu, L., Paulazzo, G., Lyass, L., Trusk, P. and Hill, J. et al. (2006). A model of acquired autoresistance to a potent ErbB2 tyrosine kinase inhibitor and a therapeutic strategy to prevent its onset in breast cancer. Proc Natl Acad Sci U S A 103, 7795–7800. Xu, C.F., Lewis, K.F.,Yeo, A.J., McCarthy, L.C., Maguire, M.F.,Anwar, Z., Danoff, T.M., Roses, A.D. and Purvis, I.J. (2004). Identification of a pharmacogenetic effect by linkage disequilibrium mapping. Pharmacogenomics J 4, 374–378. Yates, C.R., Krynetski, E.Y., Loennechen, T., Fessing, M.Y., Tai, H.L., Pui, C.H., Relling, M.V. and Evans, W.E. (1997). Molecular diagnosis of thiopurine S-methyltransferase deficiency: Genetic basis for azathioprine and mercaptopurine intolerance. Ann Intern Med 126, 608–614. Zaks T, Akkari A, Briley L, et al., (2006). Role of pharmacogenetic studies in early clinical development: Phase I studies with Lapatinib (Tykerb). ASCO Annual Meeting 2–6 June 2006.
CHAPTER
30 Clinical Implementation of Translational Genomics Wendy K. Chung
INTRODUCTION In the 55 years since the structure of DNA was described by Watson and Crick, there have been remarkable advances in our understanding of the genetic basis for human variation and disease. We have now defined the genetic basis for over 2130 monogenic human disorders, and clinical genetic tests are available for over 1410 disorders (http://www.genetests.org/). Despite the remarkable advances in our scientific understanding of genetics and genomics, the impact of such information on medical care for common conditions has to this point been modest. With the complete sequence of the human genome and many other model organisms as well as detailed characterization of human haplotype structure, we have now begun elucidating the complex genetic basis for common diseases such as macular degeneration, diabetes, obesity, inflammatory bowel disease, and breast cancer using genome-wide association studies. In some cases the same allele will increase risk for one disease and protect against another. The ApoE4 allele is associated with a protective effect for age-related macular degeneration yet susceptibility to Alzheimer disease and hyperlipidemia. As additional scientific discoveries unfold, we will be challenged to clinically integrate this new information into routine patient care to improve health and quality of life in a cost-effective, socially acceptable manner. Early adopters of this new genetic information will provide invaluable experience to guide future implementation strategies.
Genomic and Personalized Medicine, 2-vol set by Willard & Glinsburg
In this chapter, we will review the current and projected future use of genetics and genomics in clinical medicine and define the steps to successful integration of genomic medicine to improve the quality of health care. A schematic of the elements necessary to develop clinical genomic medicine is provided in Figure 30.1.
GENETIC STRATIFICATION WILL ALLOW MEDICAL CARE TO BE INDIVIDUALIZED AFTER A DIAGNOSIS IS MADE How will germline genomic variation be clinically utilized? There are emerging data that the genetic stratification of phenotypically similar diseases has important therapeutic implications. Genetic characterization following an initial diagnosis may clarify prognosis and response to therapy. Three examples to illustrate this point. 1.
Genetic testing may provide information about future risk. Women with breast cancer due to mutations in Breast Cancer 1/Breast Cancer 2 (BRCA1/BRCA2) have an increased risk for a second primary cancer, usually breast or ovarian cancer, for which increased surveillance, chemoprevention, or prophylactic surgery are recommended (Verhoog et al., 1998). However, women with breast cancer without BRCA1/BRCA2 mutations are not at increased risk for ovarian cancer (Kauff et al.,
Copyright © 2009, Elsevier Inc. All rights reserved. 357
358
CHAPTER 30
■
Clinical Implementation of Translational Genomics
Development of Genomic Medical Infrastructure • Improved genetic literacy of health care professionals • Increased numbers of genomic providers • Increased third-party coverage of genomic tests • Revisions of CPT billing codes for adequate reimbursement • Electronic medical records to efficiently integrate pharmacogenomic and other test results • Creation of laboratory standards for test performance and reporting results
Types of Genomic Tests
Pre-clinical Scientific Requirements • Demonstration of genomic basis for disease • Validation of clinical utility, including outcome data with intervention • Determination of spectrum of genomic variants and frequency by ethnicity • Development of sensitive, specific, reliable less expensive testing
• Population-based screening • Pre-symptomatic risk stratification • Defining molecular basis of diagnosed condition • Prognostic information • Molecularly based management • Pharmacogenetics • Expression profiling
Monitoring of Established Testing • Assessment of impact of innovations on • Health care cost • Morbidity, mortality • Quality of life • Assessment of unintended consequences • Assessment of public acceptance • Ongoing quality control and quality assurance of laboratories
Development of Ethical, Legal, Societal Infrastructure • Passage of Genetic Non-Discrimination Legislation • Limitation of exclusive licenses for genes • Improved genetic literacy of public • Privacy of sensitive genomic information
Figure 30.1 Development of clinical genomic medicine. The requirements to develop a health care system efficiently and effectively integrating genomic information are shown and include pre-clinical scientific data, improvements in medical infrastructure, changes in laws, policy, and public opinion, and monitoring of testing and its impact once established.
2005). Determining which women with breast cancer are at increased risk of ovarian cancer and should have prophylactic oophorectomies to reduce the risk of future ovarian cancer can be clarified with BRCA1 and BRCA2 testing. 2. Genetic testing may provide information to optimize therapy. Long QT syndrome (L\QTS) is an inherited predisposition to cardiac arrhythmias characterized by a prolonged QT interval on an electrocardiogram that can result in syncope and sudden cardiac death. There are currently nine different genetically identified causes of LQTS, all affecting cardiac ion channel conductance that are clinically difficult to distinguish. The three most common forms of LQTSs (LQT1, LQT2, and LQT3) have specific triggers for arrhythmias they can be avoided such as exerciseinduced tachycardia in LQT1 and auditory stimuli during sleep in LQT2 (Schwartz et al., 2001). Specific molecularly based pharmacological intervention is also now available. For example, patients with LQT3, due to mutations in the cardiac sodium channel gene SCN5A that inappropriately activate and open the sodium channel, respond to sodium channel blockers such as flecainide (Moss et al., 2005). In contrast, beta-blockers are routinely prescribed for LQTS1 and LQTS2. Therefore, identifying the molecular subtype
of LQTS allows for selection of the most appropriate medication and avoidance of specific triggers. 3. Genetic testing may provide prognostic information to guide management. An autosomal dominantly inherited form of diabetes called maturity onset diabetes of the young (MODY) is genetically heterogeneous and caused by mutations in six different genes. Prognosis, including diabetic complications, for patients with mutations in Glucokinase (MODY2) is much better than other forms of MODY. Molecular diagnosis can now provide reassurance that tight glycemic control is unnecessary to prevent diabetic complications in MODY2 patients who will remain only mildly hyperglycemic (Codner et al., 2006).
POPULATION-BASED GERMLINE GENOMIC SCREENING Rather than waiting until a diagnosis is made, genetic and genomic variation can be used to prevent disease. Populationbased genomic screening could provide the opportunity to improve health outcomes by disease prevention and increased surveillance to facilitate early diagnosis. It would allow population
Newborn Screening
stratification to identify individuals at increased risk who are most likely to benefit from preventive medications or interventions for which population-based therapy would not be appropriate based on cost or side effects of treatment. Data to support an individual’s susceptibility to disease could also provide the necessary motivation to increase compliance with recommended health behaviors such as exercise and weight control for individuals at increased risk for diabetes and smoking cessation for lung cancer. Screening individuals at risk for adverse outcomes to specific environmental exposures would allow susceptible individuals to modify their work or home environment to minimize exposure. Screening individuals for adverse pharmacological reactions would provide a rational basis for drug selection and avoid harmful side effects. Such a strategy would ultimately increase the total number of drugs available for clinical use and decrease the cost of drug development by identifying and eliminating drug exposure to the small number of patients who would have adversely responded to the medication. Although there is great promise for population-based screening, it is important to ensure that the natural history and clinical utility (see section on clinical utility) have been adequately defined before introduction into clinical care. As an example, consider the case of hereditary hemachromatosis. Two common mutations with a single gene (HFE) account for the majority of autosomal recessively inherited hereditary hemachromatosis in Europeans (Jazwinska et al., 1996). Furthermore, because the complications of hereditary hemachromatosis are completely preventable by reducing iron stores through phlebotomy (which is inexpensive), presymptomatic screening for hereditary hemachromatosis could be clinically useful. With a frequency of 1 in 400 Caucasians carrying one or both of two common mutations in HFE, large-scale population-based screening is technically feasible. As many experts considered introducing hereditary hemachromatosis screening on a population-wide basis, it became increasingly apparent that the natural history of mutation carriers was incompletely understood and that the disease penetrance is less than 1% (Beutler et al., 2002). Although population-based screening for hereditary hemachromatosis still has clinical utility, the implications of a positive genetic test are less significant than had once been anticipated, dramatically altering the cost-benefit ratio. Criteria to use in considering further tests for population-based screening include the disease prevalence, disease penetrance, availability of sensitive, specific, and cost-effective screening tests which may depend upon the number of genes involved and the mutation spectrum, availability of acceptable means of prevention or early detection, and the infrastructure available to perform the screening and follow up the results with confirmatory testing.
NEWBORN SCREENING Newborn screening is the largest population-based public health screening program currently in practice in the United States and throughout the world. Recent technological advances including tandem mass spectrometry and molecular genetic diagnostics
■
359
have allowed many states in the United States to cost-effectively increase the number of disorders for which newborns are screened well beyond the classically screened diseases such as phenylketonuria, hypothyroidism, and galactosemia. Over 30 disorders are now screened in many US states. Most of recent expansion in newborn screening is for inborn errors of metabolism detected by metabolomic characterization of a dried spot of blood from a newborn heel stick. Additional molecular genetic diagnostic platforms have recently been added by many states as a second-tier screen to decrease the number of false-positives from metabolomic characterization. Two-tiered screening provides a complementary method to confirm inherited diseases with a small number of well-characterized mutations such as in medium-chain acyl dehydrogenase deficiency (MCAD), cystic fibrosis, and congenital adrenal hyperplasia. The use of tandem mass spectrometry and molecular genetic testing have now opened the possibility to screen for a vast number of inherited conditions. At the request of the General Accounting Office and Health Resources and Services Administration, the American College of Medical Genetics recently re-examined the criteria for determining which disorders to include in newborn screening (http://mchb.hrsa.gov/screening/). Unlike universally accepted screening for disorders such as phenylketonuria for which early detection provides the opportunity for complete prevention of permanent disabilities, some of the newly added disorders allow for early diagnosis and treatment initiation but do not provide a cure or effectively prevent all the long-term disabilities. There is now an ongoing, evolving dialogue about what disorders should be included in newborn screening and whether the criteria proposed by Wilson and Jungner (1968) must be satisfied. Cystic fibrosis is currently screened in newborns in 28 states. Krabbe disease, treatable only by bone marrow transplantation, is now performed in New York. Many other common genetic conditions such a spinal muscular atrophy, Fragile X, and muscular dystrophy have been considered even though only supportive care is currently available for these conditions. What treatment is efficacious enough to warrant population-based screening is debated. Clearly an important factor limiting screening for many conditions is availability of technology to allow for inexpensive, sensitive, and specific screening. Multiplexed DNA microarrays to screen for inherited disease susceptibility and infectious agents in newborns have been suggested (Green and Pass, 2005) and are being piloted. In general, pediatricians have endorsed the expansion of screening in high-risk infants (Acharya et al., 2005) since it leads to reduced time to diagnosis and initiation of treatment, decreased expense and invasiveness of diagnostic testing, and allows for informed reproductive planning for the family. However, as the number of disorders screened increases, it is equally necessary to improve test parameters to increase specificity and reduce the number of false-positive tests to spare the parents of unaffected children the anxiety of a potential diagnosis that will never be made (Waisbren et al., 2003). Not all genetic testing is appropriate for newborns. Genetic testing should be defined and limited to conditions for which there will be medical implications for minors,
360
CHAPTER 30
■
Clinical Implementation of Translational Genomics
and no genetic testing should be performed for adult onset conditions for which there is no intervention during childhood.
PHARMACOGENETICS Screening for variation in drug metabolism and adverse side effects (pharmacogenetics) could allow for safer and more efficacious use of drugs. There is heritable variation in drug metabolism that alters the levels of active drugs and metabolites, thereby producing differential therapeutic effect and toxicity in different patients on the same dose of medication due to differences in drug levels. Pharmacogenetic variation is common within the population and affects the majority of patients taking medications. With genetic characterization of the genes involved in drug metabolism, it should be possible to more accurately predict the dosing of medication required to remain in the therapeutic window and circumvent the current practice of trial and error resulting in harmful side effects and delaying time to effective treatment (see Chapters 27 and 29). There is now a clinically available diagnostic panel to begin to approach pharmacogenetics. The amplichip CYP450 offers the ability to genotype 29 polymorphisms, mutations, deletions, and duplications for CYP2D6 and 2 polymorphisms in CYP2C19, two genes responsible for metabolism of approximately 25% of drugs, including tricyclic antidepressants, selective serotonergic reuptake inhibitors, beta-blockers, antipsychotics, benzodiazepines, and proton pump inhibitors. The information from the 31 assays is integrated to produce a simple interpretation quantifying a patient’s predicted metabolism to be poor, intermediate, extensive, or ultra-rapid. For some medications, package inserts now contain dosing information specific to the predicted metabolizing phenotype. Although clinical testing on the amplichip CYP450 has been available for over 2 years, clinicians have been slow to utilize the test in part due to insufficient outcome data to clearly guide use in clinical practice. Within the field of oncology, several commonly used medications are known to have common polymorphisms in genes that metabolize the drugs that can produce toxic side effects. 5Fluorouracil, irinotecan, thiopurines, and methotrexate have common mutations/polymorphisms in the drug-metabolizing genes dihydropyrimidine dehydrogenase (Gonzalez and FernandezSalguero, 1995), UDP-glycosyltransferase I (Iyer et al., 1998), thiopurine methyltransferase (Tai et al., 1996), and 5, 10-methylenetetrahydrofolate reductase (Ulrich et al., 2001), respectively. Mutations in these genes are associated with toxicity that can be easily screened at any time prior to administration of the medication. It is simple to imagine that each patient will have a pharmacogenetic profile for drug-metabolizing genes contained within their electronic medical record that will automatically warn prescribers against use of certain medications or suggest altered dosing. However, implementing the data systems necessary to integrate this information efficiently into molecular genetic diagnostic laboratories, pharmacies, hospitals, and physicians’ offices will not be trivial until electronic medical records are firmly established.
Using pharmacogenetics to rationally select the appropriate medication at the correct dose to prevent and treat disease should be economically efficient with less money spent on ineffective drugs or drug doses, fewer adverse outcomes and complications, and increased number of medications that can safely be made available to properly screened patients. As beneficial as pharmacogenetics appears to be, polymorphisms in the same genes such as CYP2D6 responsible for drug metabolism may also indicate increased susceptibility to gastrointestinal, lung, and liver cancer about which patients may not wish to be themselves aware or have in their medical record (Clapper, 2000). Other polymorphisms in ALDH2 and CYP2A6, two genes responsible for ethanol and nicotine metabolism, could be misinterpreted as socially stigmatizing markers for addiction (Agarwal, 1997). Therefore, careful consideration should be given to the genes included within pharmacogenetic profiles and clarify the intended and possible unintended use of the results.
SOMATIC GENOMIC VARIATION In addition to testing for heritable germline genetic characteristic, we will need to characterize somatic variation over the lifetime of an individual, especially in characterizing oncological specimens. Cancer treatment is likely to be rapidly redefined with genomic characterization of tumors using a combination of comparative genomic hybridization, quantification of gene amplifications/deletions, identification of acquired genetic mutations, and gene-expression profiling. These methods are already clinically utilized and used on a small scale. Molecularly targeted medications with fewer side effects will be more frequently developed as we rationally develop drugs based upon specific molecular targets. Examples already include the tyrosine kinase inhibitor imatinib mesylate (Gleevec) for chronic myelogenous leukemia and gastrointestinal stromal tumors that over express the tyrosine kinases, trastuzumab (Herceptin) for HER2 over expressing breast cancer, and gefitinib for activating mutations in Epidermal Growth Factor Receptor for non-small cell lung cancer (Lynch et al., 2004). These medications offer the possibility of increased efficacy and rely critically on accurate molecular tumor characterization to identify the subset of patients who will likely respond to these expensive therapies. Genomic Health Inc. offers another form of genomic tumor characterization, a limited expression profiling array (Oncotype Dx) of 21 genes for estrogen receptor positive, node negative, stage one or two breast cancer. The expression profile quantifies the likelihood of breast cancer recurrence in women with newly diagnosed, early stage breast cancer and assesses the benefit of chemotherapy with a numerical scoring system ranging from 1 to 100. The ability of this assay to analyze gene expression on paraffin embedded tissue offers an advantage over the requirement of freshly frozen tissue since new procedures for tissue handling and storage need not be developed to utilize this technology. The ability to simply, numerically quantify and integrate complex genomic information facilitates physician and patient communication and understanding to allow patients
Laboratory Standards to Ensure Analytic Validity
and physicians to make more rational decisions about therapy. Although not always covered by third-party payers, some patients are willing to personally bear the cost of testing to provide additional reassurance that chemotherapy will not benefit them and will allow them to avoid chemotherapy associated morbidity without increasing risk of recurrence. Third-party payer coverage will increase as clinical utility is demonstrated, patients and physicians requests for testing increases, and as diagnostic laboratories perform sufficient volumes of tests to negotiate carve out contracts with major health insurers and CMS to cover testing. There are currently multiple oncology clinical trials in progress to determine the precise algorithms by which expression profiles will be weighted and analyzed and to determine the clinical utility for prognosis and treatment efficacy (Buckhaults, 2006). Similar expression profiles from blood samples are currently being used to predict rejection for cardiac transplant recipients and obviate the need for an invasive cardiac biopsy if the expression profile indicates a low probability of rejection (Deng et al., 2006). It is likely that clinically relevant diagnostic tools will soon be available for multiple tumor types to assist with oncological management and multiple solid organ transplants to predict rejection.
NOVEL SOURCES OF GENOMIC VARIATION In addition to the nuclear germline and somatic mutations resulting from alteration in one or a small number of nucleotides, there are additional sources of genomic variation, the clinical significance of which is not yet fully appreciated. Human genomic architecture and variation are now being characterized. It is likely that previously undetected variation in the copy number of genes will underlie a significant fraction of birth defects, mental retardation, and autism (de Vries et al., 2005; Schoumans et al., 2005; Sebat el al., 2007). It is also possible that copy number variation increases susceptibility to common psychiatric, neurological, and medical conditions (Sebat et al., 2004). Somatic copy number variation has long been associated with cancer and amplification or deletion of specific genes that may provide additional prognostic and therapeutic information (Pinkel and Albertson, 2005). Bacterial artificial chromosome and oligonucleotide arrays are currently being developed and are available on a research basis to provide a molecular karyotype with resolution of 10–100 kb and are recently clinically available with a resolution of 100–500 kb. It is likely that such oligonucleotide arrays will replace standard karyotypes as the first means of analyzing chromosomes for an overall assessment of genomic balance. However, to maximize its clinical utility, it will be necessary to characterize normal copy number variation and demonstrate clinically relevant associations of copy number variations/alterations with disease. As it has become less expensive to genomically characterize gene copy number on many of the same platforms used to genotype single nucleotide polymorphisms, tests of association with common diseases will provide the data necessary to judge clinical validity and utility.
■
361
In addition to nuclear genetic variation, genetic variation also exists within the mitochondria which can vary from cell to cell and over the course of a lifetime due to mitochondrial heteroplasmy. Mitochips are currently available on a research and clinical basis to sequence the complete mitochondrial genome. As correlations between mitochondrial mutations are associated with diseases associated with degeneration and advancing age, it may become important to include methods to detect and quantify mitochondrial variation in blood as well as other tissues over the course of a patient’s lifetime.
LABORATORY STANDARDS TO ENSURE ANALYTIC VALIDITY As new methodologies and diagnostic tests such as those described above are developed, it will be necessary for diagnostic laboratories to maintain rigorous standards of quality control and quality assurance. New diagnostic assays will be developed in Clinical Laboratory Improvement Amendment (CLIA) certified laboratories. Some assays will use reagents developed and utilized only by that laboratory (“laboratory developed tests”) as has traditionally been done in molecular genetic diagnostic laboratories for rare and ultra-rare disorders. As molecular genetic testing is applied to more common disorders, it is more likely that analytespecific reagents (ASRs) will be commercially prepared and packaged for wider distribution to a large number of laboratories. It will be necessary for novel diagnostic platforms to be FDA-approved before the platforms and ASRs can be widely distributed to commercial diagnostic laboratories. The rigorous review provided by the FDA for products such as the AmpliChip CYP450 provide assurance of the analytical performance (sensitivity, specificity, and reproducibility) of the diagnostic assay and interpretability of results by clinicians. Increasingly, review and oversight of novel diagnostic tests will be necessary to assure analytical quality; however, regulation by the FDA if not expeditious could impede the efficient transfer of diagnostic methods into clinical practice. Each laboratory will need to rigorously validate the analytical performance of new assays including splitting of samples within and between laboratories to compare results between analytical methods. Standards and guidelines for genetic testing have been developed and are regularly revised by the American College of Medical Genetics (http://www.acmg. net/resources/s-g/s-g-yes-no.asp). Regular proficiency testing including the use of common samples between laboratories will ensure ongoing test quality. It is anticipated that as new ASRs are developed, the level of technical expertise necessary to perform the assays and the cost of running the assays should decrease simultaneously as throughput increases with automation and multiplex assays. It is likely that in the future the majority of molecular genetic testing will move from the many small academic boutique laboratories currently performing most testing to large commercial laboratories and/or hospitals with larger capacity and potentially one day could be performed as point of care testing.
362
CHAPTER 30
■
Clinical Implementation of Translational Genomics
Communicating test results clearly and concisely is as important as the laboratory’s analytical performance. Test reports should present sufficient detail about test methodologies to allow for comparison of methods between laboratories and allow the physician to determine the likelihood of a false-negative based upon the mutations included in the assay. As the scientific data and methods are likely to change over time, these reports will require periodic revision, especially with regard to clinical utility. Results should be succinctly reported in a final summary statement including sensitivity and specificity of the test, the possible implications of a positive and negative test result, and reference to or inclusion of necessary supporting data documenting clinical utility. Ideally, Gene Clinics (http://www.geneclinics.org/) or a similarly organized web-based format would provide periodically revised summaries of the genetic tests and their clinical utility. As data from these tests accumulate, de-identified centralized databases defined by gene or test should be maintained with phenotypic information and outcomes to allow for ongoing data collection and analysis to constantly improve test utility and interpretation, especially that of variants of unknown clinical significance.
CLINICAL VALIDATION AND CLINICAL UTILITY Genomic testing is expected to simplify diagnostic work ups, provide prognostic information, improve and refine clinical management, identify individuals at increased risk, and decrease adverse outcomes. However, before this can be implemented, genomic testing will require clinical validation and demonstration of clinical utility before acceptance into health care. The data for such validation may initially come from studies of patients with the most severe disease manifestations. However, it will be important to analyze population-based samples as well to accurately determine association between genetic susceptibility and disease in an unbiased manner. Studies in multiple ethnic groups will allow characterization of allele frequencies in various populations. It is likely that large clinical trials will provide much of these data necessary for validation since many trials now routinely integrate genetic and genomic information to stratify response to therapy and adverse outcomes by genotype. Integration of genetic data into existing and developing clinical data collection systems such as birth registries, vital statistics, cancer registries, and biorepositories linked to electronic medical records will facilitate testing and validation of preliminary genetic associations and facilitate rapid, independent confirmation of results before clinical introduction. Clinical utility is the ability in specific clinical circumstances of a genomic test to assist in clinical decision-making and improve health outcomes. Whenever possible, a genetic/genomic test should predict a defined clinical outcome with high positive and negative predictive value. Ideally, prospective, randomized, blinded controlled clinical trials genetically stratified prior to treatment would provide prospective data to validate clinical utility. To effectively evaluate the cost-effectiveness of a genetic
test, the population frequency of the at-risk genotype, the agerelated penetrance, morbidity and mortality of the disease, and the effects of interacting modifiable risk factors on genotypic expression must all be accurately known. Recommendations for clinical introduction of genetic testing should only be made when the diagnostic methodologies are reliable, patients have access to the clinical services necessary to make informed decisions and interpret the results of genetic testing, and when it becomes clinically apparent how to utilize the results of genetic testing. Once clinical testing is offered, it is also important to maintain ongoing data collection after testing is introduced to define any additional, unintended consequences. As examples, we can compare the clinical utility of genetic testing for hereditary breast/ovarian cancer conferred by BRCA1 and BRCA2 and the risk for venous thromboembolism due to Factor V Leiden. The lifetime risks of breast cancer for carriers with BRCA1 or BRCA2 mutations are approximately 65% and 45%, respectively (Antoniou et al., 2003). Mutations carriers can pursue increased surveillance for breast cancer, risk reducing prophylactic mastectomy, or chemoprevention with tamoxifen. Arguments can even be made that with a mutation frequency of 2.5% in the Ashkenazi population (Hartge et al., 1999), population-based screening for population-specific founder mutations in adult women should be considered. The clinical utility of BRCA1/BRCA2 testing is derived in large part from the high penetrance of the mutations. On the other hand, thrombophilia susceptibility conferred by Factor V Leiden is the most commonly requested molecular genetic test, although the risk conferred is much more modest and ranges from two- to eightfold (Rosendaal, 1999). It has been proposed that women should be tested for the Factor V Leiden mutation prior to initiation of oral contraceptives due to the 30-fold increased risk of venous thrombosis for mutation carriers on oral contraceptives (Vandenbroucke et al., 1994). However, although the relative risk of thrombosis is significantly increased, the absolute risk is only 28/10,000 person years and the mortality is low in young women. Failure to use oral contraceptives by the numerous Factor V Leiden mutations carriers could have other unforeseen implications including unwanted pregnancies. Thus, no professional consensus has yet emerged about the utility of testing.
COST A major impediment to integration of genetic/genomic information into health care has been the high cost of molecular genetic testing and the unwillingness of many third-party payers to cover this expense, many citing that it remains experimental. The cost of molecular genetic testing is determined largely by the costs of instrumentation, reagents, personnel, licensing fees, and professional liability insurance. As testing becomes increasingly automated and multiplexed using high-throughput assays on miniaturized scales, the cost of reagents and personnel will be reduced significantly. A goal of the National Human Genome Research Institute is to provide individual genome sequence for
Who will Provide Genomic Medical Care?
$1000 (Collins et al., 2003). Clearly if the technology can drive the cost of testing down to this extent, a major financial barrier to testing will be removed. Licensing fees have significantly increased the cost of testing for a limited number of conditions such as hereditary breast and ovarian cancer (BRCA1 and BRCA2) to $3120 (http://www.myriadtests.com/) and LQTS to $5400 (http://www.pgxhealth.com/genetictests/familion/ index.cfm). The genetic/genomic community is opposed to exclusive licenses on genes and is mixed in opinion on gene patenting, but it remains to be seen how much the gene patents and intellectual property regarding clinical utility of specific diagnostic testing will increase the average cost of testing. In addition to the absolute cost of testing, the extent to which these costs are borne by third-party payers differs widely by laboratory, test, and payer. In the 12 years since BRCA1/ BRCA2 genetic testing became clinically available, the number of individuals tested has increased dramatically, in large part as insurance companies, Medicare, and recently Medicaid began to cover the cost of testing for patients meeting medical eligibility. While many patients might elect to pursue genetic testing for health maintenance and disease prevention, third-party payers are reluctant to expose themselves to this financial burden until the utility is proven. The majority of molecular genetic testing currently performed and paid for by third-party payers surrounds reproductive issues such as carrier screening for cystic fibrosis, recommended by the American College of Obstetrics and Gynecology for consideration in all women contemplating conception. Standard of clinical care guidelines endorsed by major professional medical organization are influential in getting coverage for new tests. Without third-party payer coverage for genomic testing, genomic testing will be utilized only by a small, motivated segment of society who can afford to self-pay. A significant proportion of the financial benefit to genomic health care is attributable to preventive medicine and the opportunity to prevent chronic disease and cancer. Regional markets in which there are single, dominant insurers or organizations that are self-insured are more likely to initially support genomic health maintenance since they will see the financial benefits within a single organization. Markets with large numbers of competing insurers are reluctant to absorb large financial costs for which they will not personally see a return as consumers change between insurance carriers with regularity. However, as the cost incurred for genomic testing decreases and the utility becomes more apparent and as consumers demand this benefit, it is likely that additional major health insurance providers will add these services to both remain competitive and as a long-term cost-saving measure.
REIMBURSEMENT Many diagnostic tests, especially those that are highly multiplexed, have been slow to be developed and made clinically available because the testing laboratories cannot charge third-party payers enough to cover their operating expenses. As new technologies for molecular genetic testing are developed, it will be necessary
■
363
to revise and develop new Current Procedural Terminology (CPT) codes that accurately reflect the complexity of the testing. For many multiplex assays composed of hundreds to thousands of simultaneous assays, it will be necessary to raise the current ceiling for the maximum number of probes, hybridizations, and amplicons to reflect the complexity and cost of the new test methodologies. Once this ceiling is raised, laboratories may find it financially feasible to expand their testing platforms and test menus.
WHO WILL PROVIDE GENOMIC MEDICAL CARE? Genetic medicine is currently provided in large part by boardcertified medical geneticists and genetic counselors (see Box 1.1 in Chapter 1). Given the projected rate of growth and increasing specialization of genetic and genomic medicine and small number of 80 MD and/or PhD graduates per year in medical genetics (Korf et al., 2005), it will be necessary for all health care providers to integrate genomic information to some extent into their provision of health care. Genomics may follow other diagnostic tests such as pathology, microbiology, and diagnostic imaging with specialized physicians performing and interpreting the tests, while practitioners become largely responsible for determining which patients require testing and how to utilize the test results in heath care. Medical geneticists will continue to play important roles in research, clinical development of the field, education, laboratory medicine, and possibly treatment, in addition to providing direct clinical care for patients with disorders with which they have clinical expertise. Genetic counselors will continue to play vital roles in patient education and counseling. It will be important to recognize the role of genetic counselors by granting formal licensure and persuading insurers to reimburse for codes for genetic counseling to allow for reasonable reimbursement for genetic counselors’ services. Genetic counselors may be increasingly employed by diagnostic laboratories to offer counseling to patients referred by their physicians for testing. The demand for genetic counselors will surpass supply and will necessitate expansion of training programs that currently graduate only 160 new genetic counselors annually. In the future, there may be a specialized track for genomic counselors who will provide patient education and laboratory support not for rare monogenic disorders, but for common, polygenic conditions. Other health care professionals and/or multimedia educational programs will also need to assist with patient education. Additional responsibility will be placed upon the patient to actively gather and share medical information with their family through the use of interactive, computer-based family history tools. However, the majority of medical professionals outside genetics are currently ill equipped to provide genomic medical care. When surveyed, 65% of obstetricians and gynecologists who routinely utilize genetics in their practice did not consider themselves sufficiently educated in genetics. They rarely referred patients for genetic counseling and 86% did
364
CHAPTER 30
■
Clinical Implementation of Translational Genomics
not obtain informed consent for genetic testing (Wilkins-Haug et al., 2000). Genomic education on a massive, unprecedented scale will be necessary to implement genomic medicine (see Chapters 22 and 34).
GENOMIC LITERACY Integration of genomic information into health care will require at least rudimentary understanding of genetics and genomics by health care providers and health care consumers. Increasingly, there will be medicolegal implications for physicians if they do not recognize specific heritable risks based upon the patient’s personal or family history (Deftos, 1998). Additional time must be dedicated to genetics and genomics within medical training, beyond the average 29 hours currently allocated in 4 years of medical school (Friedman et al., 1998). Medical educators themselves will need to keep abreast of advances within genomics and rapidly revise curricula. Because genomic medicine is so rapidly changing, it will be necessary to also teach the standards by which genetic tests should be evaluated for scientific validity and clinical utility so practitioners will be equipped to judge new data as they emerge. As genomics is integrated into medical, dental, nursing, and allied professional education, a new generation of health care providers will hopefully be capable of rapidly assimilating genomic information into the life cycle of their patients. The younger generation of health care providers is already anxious to adopt this new technology in their practices. Perhaps, they even place more confidence in this new scientific field than is yet warranted. A greater challenge remains for health care professionals educated prior to the expansion of genomics – perhaps graduating from medical school only 5 years ago. For previously trained health care professionals, it will be necessary to learn general underlying principles of genetics and genomics as well as specific tools relevant to their own area of clinical expertise. The National Coalition for Health Professional Education in Genetics (NCHPEG) has produced a set of core competencies in genetics essential for all health care professionals (http://www. nchpeg.org/eduresources/core/Corecomps2005.pdf). NCHPEG is currently developing a core curriculum for genetic education (http://www.nchpeg.org/eduresources/core/coreprinciples. pdf) and has produced a CD ROM “Genetics and Common Disorders: Implications for Primary Care and Public Health Providers” that should assist in educating health care professionals currently in practice. Geneticists will play critical roles as genomic medical educators and discern when information is clinically relevant for various subspecialty areas with clinical care recommendations supported by major medical professional organizations. Genomic educators will need to continuously update medical professionals with new discoveries and applications in simple terms that emphasize clinical utility. It will remain a scientific challenge to define clinical situations in which complex genomic information can be distilled into an appropriately simple interpretation that has unambiguous clinical implications that impact clinical outcomes.
Genomic literacy in the public must simultaneously improve and remove the mysticism and misconceptions some patients may have regarding hereditary information. We must enlighten those who believe genetics is deterministic who would otherwise fail to modify health behaviors, believing instead in an inevitable, unalterable fate. There is a wide and growing divide between patients utilizing genetic and genomic information in their health care, driven in part by education and socioeconomic status. Younger patients tend to utilize genetic information more aggressively and seek genetic information both out of curiosity and to make more informed life and reproductive decisions. Younger generations have also become accustomed to advances in assisted reproductive technology and increasingly push the ethical and social boundaries to use such genetic and reproductive technology for such trivial genetic factors as elective sex selection. Older patients tend to be less educated about emerging genomic technologies, and often the reason they cite for seeking genetic testing is to provide information for their children or grandchildren rather than for themselves. As the public becomes more genetically literate, there will likely be a push from diagnostic laboratories to market genetic testing directly to consumers. This has already been done with BRCA1/BRCA2 testing and is done with nutriceuticals. While such advertising campaigns may be effective in increasing awareness and forcing health care professionals to learn new information about genetic testing, there will also be the danger that inappropriate tests will be ordered and potentially misused by patients and physicians who are not fully informed.
ETHICAL, LEGAL, AND SOCIAL ISSUES Disparities in genetic literacy are correlated with education which is itself correlated with race. Race independently further complicates the scientific interpretation of genomic data as minority populations are initially generally less well studied than Caucasians. Insufficient information about normal genetic diversity in various ethnic groups complicates and delays development and accurate interpretation of clinical diagnostic tests across in all populations. For instance, missense variants of unknown clinical significance in BRCA2 are clinically reported more commonly in minority populations due in part to deficiencies in our knowledge of the spectrum of normal variation in non-Caucasian populations. These ambiguous results can lead to increased anxiety and possibly unnecessary prophylactic surgeries when misinterpreted by patients (and physicians) who may be educationally ill equipped to comprehend complex and evolving test results. Within the United States, minority populations have historically been concerned about discrimination. It is important to present educational information about genomics in a culturally sensitive way, ideally by members within each community. Access to genomic information and treatment should be equitable. Currently there are great disparities based upon geography, inequities in coverage of genetic testing by insurance companies, Medicare, and Medicaid and cultural and
References
language barriers with 93% of genetic counselors being Caucasian (Farmer and Chittams, 2000). Patients have great concerns about the misuse of their genetic information. Although the Americans with Disabilities Act provides some reassurance against discrimination in the workplace and the Health Insurance Portability and Accountability Act (HIPAA) protects against eligibility discrimination for group health care for individuals with genetic susceptibility to disease, patients continue to feel vulnerable. Failure by Congress previously to pass the Genetic Information Nondiscrimination Health Act left many consumers with concern that genetic information could be used to discriminate against them for individual health insurance plans, life insurance, long-term disability, longterm care insurance, admission to educational institutions, or the military, jobs, loans, or social status. The Genetic Information Nondiscrimination Health Act has passed the House and the Senate. It should provide Americans with additional reassurance about genetic discrimination. Concerns about genetic discrimination are now even discussed on the sports page as Eddy Curry, a professional basketball player, refused to have genetic testing for hypertrophic cardiomyopathy because of concerns that he would be unable to play professional basketball if he tested positive. Although there have been no documented successful cases of health insurance discrimination on the basis of genetic predisposition, because more subtle forms of discrimination can be difficult to document, some patients still prefer to have genetic testing performed anonymously. Legislation ensuring genetic privacy will be necessary before many consumers will confidently embrace and utilize genomic medicine.
■
365
CONCLUSIONS There will come a time in the future when we will be able to determine, in real time at different points in our lives, our susceptibility to disease and specific steps we can take to intervene with individualized recommendations about lifestyle and dietary changes, pharmacological intervention, and increased surveillance to improve health outcomes and decrease adverse outcomes. It is important for all stakeholders to have realistic expectations about the time required to implement these changes. Clinical validation and utility should be rigorously demonstrated prior to clinical implementation. Genomic literacy among health care providers, insurers, and patients must increase, requiring a massive educational effort. Costs of testing must decrease dramatically and coverage for the cost of testing must be borne by third-party payers.This will not happen as quickly as some might naively expect, and fostering unrealistic public expectations will surely lead to disappointment. However, with advances in scientific knowledge and verification of clinical applications of human genetics, genomics, metabolomics, proteomics, and molecular imaging, it should be possible to improve health outcomes at decreased cost.We must be mindful to ensure equal access to genomic medicine and utilize these medical advances rationally and cost-effectively without patient’s fear of discrimination.
ACKNOWLEDGEMENTS Roberto Almazan and Josue Martinez provided assistance with manuscript preparation.
REFERENCES Acharya, K., Ackerman, P.D. and Ross, L.F. (2005). Pediatricians’ attitudes toward expanding newborn screening. Pediatrics 116, E476–E484. Agarwal, D.P. (1997). Molecular genetic aspects of alcohol metabolism and alcoholism. Pharmacopsychiatry 30, 79–84. Antoniou, A., Pharoah, P.D., Narod, S., Risch, H.A., Eyfjord, J.E., Hopper, J.L., Loman, N., Olsson, H., Johannsson, O., Borg, A. et al. (2003). Average risks of breast and ovarian cancer associated with BRCA1 or BRCA2 mutations detected in case Series unselected for family history: A combined analysis of 22 studies. Am J Hum Genet 72, 1117–1130. Beutler, E., Felitti, V.J., Koziol, J.A., Ho, N.J. and Gelbart, T. (2002). Penetrance of 845G→A (C282Y) HFE hereditary haemochromatosis mutation in the USA. Lancet 359, 211–218. Buckhaults, P. (2006). Gene expression determinants of clinical outcome. Curr Opin Oncol 18, 57–61. Clapper, M.L. (2000). Genetic polymorphism and cancer risk. Curr Oncol Rep 2, 251–256. Codner, E., Deng, L., Perez-Bravo, F., Roman, R., Lanzano, P., Cassorla, F., Chung, W. (2006). Glucokinase mutations in young children with hyperglycemia. In Diabetes/Metabolism Research and Reviews 22(5), 348–355. Collins, F.S., Green, E.D., Guttmacher, A.E. and Guyer, M.S. (2003). A vision for the future of genomics research. Nature 422, 835–847.
Deftos, L.J. (1998). The evolving duty to disclose the presence of genetic disease to relatives. Acad Med 73, 962–968. Deng, M.C., Eisen, H.J., Mehra, M.R., Billingham, M., Marboe, C.C., Berry, G., Kobashigawa, J., Johnson, F.L., Starling, R.C., Murali, S. et al. (2006). Noninvasive discrimination of rejection in cardiac allograft recipients using gene expression profiling. Am J Transplant 6(1), 150–160. de Vries, B.B., Pfundt, R., Leisink, M., Koolen, D.A., Vissers, L.E., Janssen, I.M., Reijmersdal, S., Nillesen,W.M., Huys, E.H., Leeuw, N. et al. (2005). Diagnostic genome profiling in mental retardation. Am J Hum Genet 77, 606–616. Farmer, J. and Chittams, J. (2000). Professional status survey 2000. Perspect Genet Couns 22, S1–S12. Friedman, J.M., Blitzer, M., Elsas, L.J., 2nd., Francke, U. and Willard, H.F. (1998). Clinical objectives in medical genetics for undergraduate medical students. Association of Professors of Human Genetics, Clinical Objectives Task Force. Genet Med 1, 54–55. Gonzalez, F.J. and Fernandez-Salguero, P. (1995). Diagnostic analysis, clinical importance and molecular basis of dihydropyrimidine dehydrogenase deficiency. Trends Pharmacol Sci 16, 325–327. Green, N.S. and Pass, K.A. (2005). Neonatal screening by DNA microarray: Spots and chips. Nat Rev Genet 6, 147–151. Hartge, P., Struewing, J.P., Wacholder, S., Brody, L.C. and Tucker, M.A. (1999). The prevalence of common BRCA1 and BRCA2 mutations among Ashkenazi Jews. Am J Hum Genet 64, 963–970.
366
CHAPTER 30
■
Clinical Implementation of Translational Genomics
Iyer, L., King, C.D.,Whitington, P.F., Green, M.D., Roy, S.K.,Tephly,T.R., Coffman, B.L. and Ratain, M.J. (1998). Genetic predisposition to the metabolism of irinotecan (CPT-11). Role of uridine diphosphate glucuronosyltransferase isoform 1A1 in the glucuronidation of its active metabolite (SN-38) in human liver microsomes. J Clin Invest 101, 847–854. Jazwinska, E.C., Cullen, L.M., Busfield, F., Pyper, W.R., Webb, S.I., Powell, L.W., Morris, C.P. and Walsh, T.P. (1996). Haemochromatosis and HLA-H. Nat Genet 14, 249–251. Kauff, N.D., Mitra, N., Robson, M.E., Hurley, K.E., Chuai, S., Goldfrank, D.,Wadsworth, E., Lee, J., Cigler,T., Borgen, P.I. et al. (2005). Risk of ovarian cancer in BRCA1 and BRCA2 mutation-negative hereditary breast cancer families. J Natl Cancer Inst 97, 1382–1384. Korf , B.R., Feldman, G. and Wiesner, G.L. (2005). Report of Banbury Summit meeting on training of physicians in medical genetics, October 20–22, 2004. Genet Med 7, 433–438. Lynch,T.J., Bell, D.W., Sordella, R., Gurubhagavatula, S., Okimoto, R.A., Brannigan, B.W., Harris, P.L., Haserlat, S.M., Supko, J.G., Haluska, F.G. et al. (2004). Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefitinib. N Engl J Med 350, 2129–2139. Moss, A.J.,Windle, J.R., Hall,W.J., Zareba,W., Robinson, J.L., McNitt, S., Severski, P., Rosero, S., Daubert, J.P., Qi, M. et al. (2005). Safety and efficacy of flecainide in subjects with Long QT-3 syndrome (DeltaKPQ mutation): A randomized, double-blind, placebo-controlled clinical trial. Ann Noninvasive Electrocardiol 10, 59–66. Pinkel, D. and Albertson, D.G. (2005). Comparative genomic hybridization. Annu Rev Genomics Hum Genet 6, 331–354. Rosendaal, F.R. (1999). Venous thrombosis: A multicausal disease. Lancet 353, 1167–1173. Schoumans, J., Ruivenkamp, C., Holmberg, E., Kyllerman, M., Anderlid, B.M. and Nordenskjold, M. (2005). Detection of chromosomal imbalances in children with idiopathic mental retardation by array based comparative genomic hybridisation (array-CGH). J Med Genet 42, 699–705. Schwartz, P.J., Priori, S.G., Spazzolini, C., Moss, A.J., Vincent, G. M., Napolitano, C., Denjoy, I., Guicheney, P., Breithardt, G., Keating, M.T. et al. (2001). Genotype-phenotype correlation in
the long-QT syndrome: gene-specific triggers for life-threatening arrhythmias. Circulation 103, 89–95. Sebat, J., Lakshmi, B., Malhortra, D.,Troge, J., Lese-Martin, C., Walsh, T., Yamrom, B., Yoon, S., Krasnitz, A., Kendall, J. et al. (2007). Strong association of de novo copy number mutations with autism. Science 316, 445–449. Sebat, J., Lakshmi, B.,Troge, J.,Alexander, J.,Young, J., Lundin, P., Maner, S., Massa, H.,Walker, M., Chi, M. et al. (2004). Large-scale copy number polymorphism in the human genome. Science 305, 525–528. Tai, H.L., Krynetski, E.Y., Yates, C.R., Loennechen, T., Fessing, M.Y., Krynetskaia, N.F. and Evans, W.E. (1996). Thiopurine S-methyltransferase deficiency: two nucleotide transitions define the most prevalent mutant allele associated with loss of catalytic activity in Caucasians. Am J Hum Genet 58, 694–702. Ulrich, C.M., Yasui, Y., Storb, R., Schubert, M.M., Wagner, J.L., Bigler, J., Ariail, K.S., Keener, C.L., Li, S., Liu, H. et al. (2001). Pharmacogenetics of methotrexate: Toxicity among marrow transplantation patients varies with the methylenetetrahydrofolate reductase C677T polymorphism. Blood 98, 231–234. Vandenbroucke, J.P., Koster, T., Briet, E., Reitsma, P.H., Bertina, R.M. and Rosendaal, F.R. (1994). Increased risk of venous thrombosis in oral-contraceptive users who are carriers of factor V Leiden mutation. Lancet 344, 1453–1457. Verhoog, L.C., Brekelmans, C.T., Seynaeve, C., van den Bosch, L.M., Dahmen, G., van Geel, A.N.,Tilanus-Linthorst, M.M., Bartels, C.C., Wagner, A., van den Ouweland, A. et al. (1998). Survival and tumour characteristics of breast-cancer patients with germline mutations of BRCA1. Lancet 351, 316–321. Waisbren, S.E.,Albers, S.,Amato, S.,Ampola, M.,Brewster,T.G.,Demmer, L., Eaton, R.B., Greenstein, R., Korson, M., Larson, C. et al. (2003). Effect of expanded newborn screening for biochemical genetic disorders on child outcomes and parental stress. JAMA 290, 2564–2572. Wilkins-Haug, L., Erickson, K., Hill, L., Power, M., Holzman, G.B. and Schulkin, J. (2000). Obstetrician-gynecologists’ opinions and attitudes on the role of genetics in women’s health. J Womens Health Gend Based Med 9, 873–879. Wilson, J.M. and Jungner,Y.G. (1968). Principles and practice of screening for disease.World Health Organization, Geneva.
RECOMMENDED RESOURCES http://www.genetests.org http://www.geneclinics.org/ http://mchb.hrsa.gov/screening
http://www.acmg.net/resources/s-g/s-g-yes-no.asp http://www.nchpeg.org/eduresources/core/Corecomps2005.pdf http://www.nchpeg.org/eduresources/core/coreprinciples.pdf
CHAPTER
31 Translating Innovation in Diagnostics: Challenges and Opportunities Matthew P. Brown, Myla Lai-Goldman and Paul R. Billings
INTRODUCTION The last decade has seen rapid advances in diagnostic tools and novel therapies as a result of basic research in fields as diverse as genomics, drug discovery, information science, imaging, and nanotechnologies (Mankoff et al., 2004; Reece, 2006). These areas of advancing knowledge are changing the practice of modern medicine. However, their uptake in the health care setting is often slow and haphazard (Graham et al., 2006; Kerner, 2006). It is clear that new knowledge and clever technologies are not enough to improve patient care. Basic discoveries must be linked to insights surrounding the development, delivery, deployment, and application of new information or techniques in health care for translation to successfully occur. The study of translating knowledge in health care or “translational medicine” has been characterized in many ways, but one definition is that efforts to find cures for affected individuals are complemented by the pursuit to understand human diseases and their complexities (Mankoff et al., 2004). Within this idealized framework, advances in basic research are properly tested in the clinical setting, then knowledge gained is both fed back to basic researchers and efficiently translated into new therapeutic strategies to treat, cure or prevent disease (Marincola, 2003; Sonntag, 2005). However, gaps in translation and the provision of new knowledge in the health care setting (so-called “translational gaps”) do occur. They can adversely impact the quality of medical care, decrease patient safety, and escalate costs. An example Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
of an important topic in translational medicine is the estimated 30–45% of all patients who do not receive medical care according to the best scientific evidence. In fact, it has been reported that 20–25% of medical care currently provided is either unnecessary or potentially harmful (Graham et al., 2006). Efforts to close these translational gaps have focused in large part on the perceived obstacles and challenges in translating basic research into viable diagnostic and therapeutic strategies. These include: communication and education between researchers from diverse fields, institutional support for translational research activities, regulatory hurdles, the high cost of developing diagnostics and novel therapeutics, the need for bioinformatics and complex data handling capabilities, adapting new technologies to high-throughput automation, intellectual property and licensing protection, and the willingness to move away from standard therapies and embrace new paradigms for diagnostic and therapeutic strategies (Horig and Pullman, 2004; Horig et al., 2005; Humes, 2005; Marincola, 2003; Reece, 2006;Tsongalis, 2005). There is a need to recognize the key role of technological innovation in closing translational gaps in medicine as compared to incremental improvements with current practice. However, significant health gains can only occur when the fidelity with which medical advances are delivered is optimized. These two processes can sometimes be seen as competing (Woolf and Johnson, 2005), but in actuality represent two factors necessary for medical progress. There may be no area where this is more clearly demonstrated than in translational diagnostics, a field Copyright © 2009, Elsevier Inc. All rights reserved. 367
368
CHAPTER 31
■
Translating Innovation in Diagnostics: Challenges and Opportunities
driven by innovation (Ratner, 2006). The implementation of genomics and other innovative diagnostics for personalized medicine requires the translation of a vast array of technologies and therapeutic strategies from basic science to clinical application. Delivery of these diagnostics to patients requires physician and patient education, as well as evidence-based demonstration of efficacy for healthcare uptake, payment, and regulatory approval. How have some of these challenges been addressed? Just a few years ago, academic medical centers were the primary mechanism to foster translational research and facilitate collaboration between researchers and clinicians (Puderbaugh, 2006). Multidisciplinary educational programs were funded primarily by private institutions, such as the Doris Duke Charitable Foundation (Mankoff et al., 2004). In the United States, the National Institutes of Health (NIH) and the Food and Drug Administration (FDA) then began initiatives to identify, finance, and develop translational research programs. The NIH Roadmap for Medical Research aimed to identify and support research beyond the scope of any single NIH component (NIH, 2006). Roadmap initiatives sought to improve the development and availability of modern scientific tools and information resources, foster novel methods of research collaboration, and enhance the nation’s clinical research enterprise. The NIH funded the first 12 Clinical and Translational Science Awards (CTSAs) in 2006 for academic centers conducting projects on translational research (NCRR, 2006); additional centers are being named in subsequent years. The FDA plays a key regulatory role in medical practice and has promoted translational medicine. The FDA’s Critical Path Initiative, subtitled “Stagnation to Innovation,” identified key areas for development: better evaluation tools – Biomarkers and Disease Models; streamlining clinical trials; harnessing bioinformatics; moving manufacturing into the 21st century; and products to address urgent public health needs and at-risk populations (FDA, 2006a). In early 2006, the FDA released The Critical Path Opportunities List, which outlined 76 science projects to bridge the translational research gap (FDA, 2006b, c). Of note, the first 33 items on the FDA’s Opportunities List were in diagnostic-related areas, reflecting the importance of diagnostics in driving innovation in medical practice (Ratner, 2006). Translational diagnostics is the subfield of translational medicine concerned with diagnostic methods and information. Given the importance of diagnostics, it is crucial to look at translational issues affecting this area of innovation with the intent of closing the gap between basic research and the medical system. As the role of diagnostics continues to evolve, examining translational diagnostics from the view of the largest providers of labbased tests, the national reference laboratories, will be important. Reference labs translate everything from “home brew” technologies (also called laboratory-developed tests) licensed directly from academic institutions, to FDA-approved kits purchased from manufacturers, and sophisticated in vitro diagnostic multivariate index assays (IVDMIAs) used to generate patient specific risk data (FDA, 2006d). They provide complex data analysis of laboratory tests, educate both physicians and patients, and in this and other ways spearhead the uptake of new diagnostic technologies.
With this in mind, this chapter will focus on translational diagnostics in the context of genomic and personalized medicine and on the practical lessons learned from bringing innovative diagnostics to market at larger reference laboratories. The large national labs in the United States – including Quest Diagnostics, Laboratory Corporation of America Holdings, Sonic Healthcare, Genzyme, Mayo Clinic, Associated Regional University Pathology (ARUP), and others – provide about 50% of all lab testing delivered in this country and complete several million assays a day in support of patient care. Understanding translational diagnostics as practiced by these national reference laboratories should shed light on both challenges and successes in bridging the translational diagnostics gap from bench to bedside. The potential for translational diagnostics in cancer, infectious disease, and other health care sectors will also be discussed.
NOVEL DIAGNOSTICS The goal of translational diagnostics is to provide improved diagnostic procedures leading to more efficacious therapies and improved medical outcomes. Early laboratory medicine consisted primarily of the direct examination of tissue or the analysis of simple analytes in blood. Molecular biology techniques such as positional DNA cloning, DNA sequencing and nucleic acid hybridization assays led to early molecular diagnostics. Southern blot hybridization assays were used to detect gene deletions in Duchenne muscular dystrophy, repeat expansions in fragile X syndrome, and gene rearrangements diagnostic for B- and T-cell lymphoma (Tsongalis and Silverman, 2006). The first molecular diagnostic products to reach the market included tests for the detection of viral RNA or DNA, genetic tests for single gene diseases, and tests to determine risk for developing certain cancers, such as breast or colon cancer. Many of the new diagnostic technologies entering the clinical laboratory are part of an evolution from nucleic acid amplification technologies for use in amplifying, identifying, and sequencing target single gene sequences towards a systems biology approach which looks at the expression of multiple genes in response to a variety of stimuli and conditions such as infectious disease, drug therapy, and cancer (Billings and Brown, 2004). Systems biology is often used as a synonym for functional genomics, a description of the genomic and epigenomic influences on a trait and interactions with environmental variants (see Chapter 6). New Diagnostic Technologies To understand translational diagnostics and the gaps that currently exist in the movement of knowledge, it is necessary to understand what technological innovations and potential applications are coming forth (Table 31.1). Microarrays Microarray technology typically involves tethering numerous probes to a solid substrate. For DNA microarrays, the probes usually consist of small oligonucleotides or complementary
Novel Diagnostics
TABLE 31.1
Diagnostic applications and technologies
Application
Example
Technology
Pathogen quantity
HIV viral load
QPCR
Pathogen detection
Bordetella pertussis
Real-time PCR
Emerging infectious disease detection
SARS
Protein microarray
Drug selection
HER2
IHC/FISH
Disease screening
HPV Testing
Hybrid capture
Predisposition
BRCA
Gene sequencing
Prognosis
Oncotype Dx
Expression array
Therapeutic response monitoring
HIV-1 genotyping
PCR
Prediction
Huntington’s disease genotyping
PCR
Pharmacogenomics companion diagnostic
Irinotecan selection UGT1A1 genotyping
PCR
Pharmacogenomics drug selection/ dosage
Warfarin Test CYP2D9/VKORC1 genotyping
PCR
HIV, Human immunodeficiency virus; QPCR, quantitative polymerase chain reaction; SARS, Severe acute Respiratory syndrome; IHC, immunohistochemistry; FISH, fluorescence in situ hybridization; HPV, human papillomavirus.
DNAs (cDNAs). DNA targets, commonly in the form of fluorescently labeled cDNA or genomic DNA fragments, are then hybridized to the probe. Detection of a fluorescent signal from directly labeled nucleic acid or protein samples is the most common microarray detection method (Tefferi et al., 2002). DNA microarray technology provides data on DNA sequence variation (mutations and polymorphisms) and gene expression levels (see Chapters 8, 9, and 13). Gene expression profiles can identify upregulated and downregulated genes, which are then targets for novel therapeutics. cDNA arrays have also been used to classify pathological subgroups of specific disorders. When polymorphisms are identified, genomic DNA targets are probed with oligonucleotides which define allelic differences. Recent advances in microarray technology involve the use of liquid-phase or microparticle-based arrays. One example would be bead-based multiplexing, which allows multiple analytes to be assayed in the same well. Molecular diagnostic applications of microarray technology include cancer diagnosis, typing, and prognosis; infectious disease identification, biodefense applications, and pharmacogenomics (Petrik, 2006).
■
369
Another application of array technology is array comparative genome hybridization. This is a powerful technique that detects high-level amplification and homozygous deletions in small genomic regions. Array technology has been used to detect highresolution copy number changes in breast, renal, and bladder cancer (Gabriele et al., 2006; Petrik, 2006). Using the genomic clone sequence data, this technique can lead quickly to the identification and cloning of genes associated with disease for use in diagnostics and targeted therapeutics. Array technology promises to revolutionize the field of medical cytogenetics by providing molecular karyotyping without the need to culture cells or stain chromosomes for visualization (see Chapter 9). Proteomics Proteomics seeks to understand the structure, function, and expression of all proteins encoded by a given genome (see Chapter 14).The human genome project identified approximately 25,000 human genes. However, it is estimated that the number of protein products is 10–30 times higher due to alternative splicing and post-translational modification of gene products (Petrik, 2006). Proteomics also encompasses protein expression profiles as a function of age, state, and environment. The basic technique of proteomics is the separation of proteins from a biological sample by various separation technologies (2-D PAGE, liquid chromatography, affinity capture) followed by detection of proteins peptide fragments by technologies such as peptide fingerprinting by mass spectrometry, protein arrays, or antibody arrays (Rice et al., 2006). Proteomic analysis based on mass spectrometry may be better suited for the identification of useful protein biomarkers, while technologies such as automated chip technology using protein arrays may have higher clinical diagnostic utility. Protein arrays use protein probes that in solution retain the ability to interact specifically with other proteins or molecules. Protein arrays can be used to identify protein–protein interactions, enzyme–substrate interactions, and antibody–antigen interactions. Glycomics Glycosylation is a critically important post-translational modification of cellular proteins. Glycan moieties are involved in a wide variety of intracellular, cell–cell, and cell–matrix recognition events. Glycomics is the study of the glycans produced by humans and their role in protein function in health and disease (Morrelle and Michalski, 2005). Early technologies used separation methodologies and mass spectrometry analyses that were similar to proteomic methods. New approaches are being developed for clinical applications and include such technologies as microcapillary chromatography, lectin affinity chromatography, and carbohydrate microarray and mass spectrometry (Miyamoto, 2006). Glycan analysis has contributed to drug discovery, clinical assays, and basic research into the underlying biology of cancer. However, its greatest promise will be to enhance genomic and proteomic analysis of clinical samples to determine the effects of the genetic, environmental, lifestyle, and nutritional state of patients on their health status.
370
CHAPTER 31
■
Translating Innovation in Diagnostics: Challenges and Opportunities
Epigenetics Epigenetics involves the study of heritable changes in gene function that occur without a change in DNA sequence (see Chapter 1). Epigenetic mechanisms include DNA methylation, histone acetylation, and RNA interference (see Chapters 5 and 11). Some neuropsychiatric and rheumatologic diseases and cancers may affect both the genotype and epigenotype. Small interfering RNAs (siRNAs) can induce specific post-transcriptional gene silencing in mammalian systems and hold great promise as therapeutic agents. Diagnostic assays will have to be developed to analyze the effectiveness of these new therapies. Cancer cells are known to undergo changes in 5-methylcytosine distribution including hypomethylation and hypermethylation of promoter CpG islands, which are associated with tumor-suppressor genes (Esteller, 2002). An integrated analysis of the epigenetic state of cancer cells can involve DNA methylation, transcription factor binding analysis, and histone acetylation analysis. Treatments being evaluated include histone deacetylase inhibitors and DNA methyltransferase inhibitors. DNA methylation patterns are being developed as diagnostic biomarkers for detection and risk assessment (Laird, 2005) (see Chapter 11). Biological Imaging Biological imaging has begun to play an increasingly important role in the early detection and staging of disease (see Chapter 43). Recent advances in imaging include multislice computer tomography (CT) and whole-body magnetic resonance imaging (MRI) to provide earlier detection and more accurate determination of the extent of disease (Schwaiger and Peschel, 2006). Tracer techniques in combination with positron emission tomography (PET) have aided in discriminating between normal and tumor tissues. Such molecular imaging technologies seek to visualize biological processes and molecular binding sites. New advances such as PET-CT provide qualitative and quantitative assessment of tumor tissue. Taken together with other molecular diagnostic methods, biological imaging should aid in disease detection, be predictive of therapeutic efficacy, and provide markers for disease treatments. Nanotechnology The near future of nanobiosensing envisions a merger of molecular diagnostic technology with nanotechnology to produce nanobiosensors capable of detecting biological processes at the molecular level with continuous operation in real time (see Chapter 51 and Demidov, 2004). Molecular beacons comprised of both nucleic acid probes and self-reporting optical transducers should be capable of providing feedback for monitoring therapeutic intervention or for periodic checkup. Nanoparticlebased molecular detection is being used currently to develop multiplexed molecular recognition arrays and label free ways to quantify specific binding events (Ming-Cheng Cheng et al., 2006). Quantum dot nanocrystals provides a nearly unlimited range of sharply defined color indicators that can be tagged to biomolecules of interest and provide long-lived, sensitive probes. Nanodevices using microfluidic technologies may produce useful cancer, infectious diseases and other tests. These technologies
will aid the continuing miniaturization of molecular diagnostic platforms; a necessary evolution if molecular diagnostics are to be expanded to point of care use and for system biology approaches that require the simultaneous detection of specific biological processes associated with disease. Applications Infectious Disease Factors affecting disease progression and therapy management in infectious disease associated with viral, bacterial, fungal, and parasitic pathogens include rapid detection and identification, quantifying pathogen burden, and pathogen genotyping (NIH, 2002). As with the Human Immunodeficiency Virus (HIV), quantifying the viral load of a wide variety of viral pathogens is needed to monitor both the progression of the disease and the response to therapy. Pathogen genotyping is also essential in order to identify variants resistant to therapy, and to differentiate closely-related species or viral types. Currently, the gold standard for pathogen detection is growth of the pathogen in culture. This process is time-consuming and in many cases non-informative. Diagnostic methods for viral detection consist primarily of immunological identification of antibodies produced in response to viral infection. Molecular diagnostics utilizing nucleic acid amplification testing (NAAT) have been used for viral load monitoring and genotyping for HIV and screening for HIV1, hepatitis B and C (HBV, HCV), and West Nile virus (WNV) (Paxton, 2004). In HIV and HCV there is a window between infection and detection of antibody production in which immunological methods fail to detect viral infection. This is a very important issue for blood safety and for the identification of acute infections which may account for a high percentage of new clinical cases. NAAT testing has been implemented for blood screening and has been estimated to prevent about five HIV-1 and 56 HCV infections per year. Another advantage of NAAT testing lies with the flexibility of the platform in the face of emerging infectious disease threats. When WNV became a public health concern in 2002, these platforms were converted for the detection and monitoring of WNV within 9 months. With the incorporation of DNA microarray technologies, rapid pathogen detection and genotyping may be accomplished simultaneously. Recently, a protein microarray for Coronaviruses, including the severe acute respiratory syndrome (SARS) virus, has been tested and found to distinguish between SARS and other Corona viruses (Zhu et al., 2006). Low-density microarray tests are being researched for use in subtyping influenza A and avian type H5N1. These types of arrays may soon have clinical utility for monitoring influenza worldwide, developing vaccines, and eventually for point of care applications (Mehlmann et al., 2007). This last application is very important, as determining the presence of influenza on solely clinical grounds is very difficult due to poor specificity and sensitivity of clinical findings, the other pathogens that cause similar symptoms, and the influenza subtypes that cause different symptoms. For example, up to 70% of patients with influenza symptoms are not infected with influenza virus, and up to 30% of those are due to Coronaviruses
Novel Diagnostics
(Montalto, 2003; Brown, 2006). CombiMatrix has developed a DNA microarray to detect and type influenza strains and Tm Bioscience Corp. (now Luminex) has developed a major human respiratory virus array. Molecular diagnostics can be utilized for sexually transmitted diseases such as Chlamydia trachomitis and Nisseria gonorrhea, as well as other bacterial pathogens such as Legionella pneumophilia (Legionnaire’s disease), Borrelia burgdoreri (Lyme disease), Mycobacterium tuberculosis, and Bordetella pertussis and B. parapertussis (whooping cough). Despite effective vaccination programs, Bordetella pertussis remains an important and highly contagious disease due to the fact that newborns do not appear to gain passive maternal protection and vaccine-induced immunity is far less complete than expected. A real-time PCR assay that accurately detects and differentiates the DNA from both agents of pertussis syndromes has been developed (Vincart et al., 2007). This real-time PCR assay provides results within 24 h of sample receipt and provides clinicians with more timely and accurate information than that provided by culture, direct fluorescence antibody, serological testing, or even conventional PCR methods. Additional applications include tissue product testing, bioterrorism monitoring, and characterization of host–pathogen interactions in the search for additional diagnostic biomarkers. Oncology Standard diagnostic procedures for human tumors use a combination of histopathology, special stains, immunohistology, radiology, and clinical data. Diagnostics derived in this manner provide data on tissue origin, tumor type, stage, and grade, along with information on completeness of surgical tumor removal (Dietel and Sers, 2006). However, these tests are neither sensitive nor specific enough to differentiate patients who will respond to treatment from those who will not. The application of high-throughput DNA array technologies, which provide gene expression profiles, has revolutionized tumor diagnostics and led to the identification of novel tumor subgroups for otherwise indistinguishable tumors (Dietel and Sers, 2006). Researchers have also identified genes the expression of which can be used to predict the metastatic potential of breast carcinomas. In the future, technologies that take a systems approach will be needed to identify patients who will benefit from novel therapies. These methods must be capable of detecting multiple oncogenic pathways in tumors both before and during therapy and they must be adaptable to routine clinical diagnostic testing. Possible methods include: tissue microarrays; forward-phase protein microarrays, capable of simultaneously detecting multiple protein interactions with a single sample; and reverse-phase protein microarrays, capable of probing multiple analytes with specific antibodies (Gulmann et al., 2006). These testing modalities may be applied to primary or metastatic tumor biopsies, or to ‘blood biopsies’ of rare circulating tumor cells. Pharmacogenomics Understanding the role of human genetic variation in disease susceptibility and in the selection and efficacy of therapeutics is
■
371
an essential component of optimizing care. Pharmacogenomics looks at the impact of patient genetic variation on response to therapeutics (see Chapter 27). Pharmacogenomics also influences the delivery and cost of healthcare (Phillips et al., 2004). Sensitivity, toxicity, and resistance to medications at standard doses are prime examples of the effect of genotypic heterogeneity within the patient population. Well known, clinically relevant examples include differential responses to warfarin, codeine, thiopurine antileukemic drugs, and succinylcholine. Genetic differences in receptor structure and affinity, drug metabolism, and drug transport systems are now known to play an important role in both differential responses to drug therapy and adverse drug reactions (ADRs). a) CYP450
Adverse drug reactions result in 6.7% of hospitalizations and 0.32% of mortalities (Montgomery and Louie, 2001). The genes most associated with ADR encode receptors, metabolic enzymes, and metabolite transport proteins, the same genes have been implicated in environmental toxin susceptibility and cancer predisposition. Drug efficacy is directly related to the binding of the drug molecule to cell surface receptors. For example, it has been demonstrated that patients expressing high levels of beta-adrenergic receptors are more responsive to beta-agonists and antagonists. Conversely, those expressing low levels of this receptor require higher-drug levels to achieve a comparable pharmacological effect. Once inside the cell, the drug is metabolized by a number of enzymes catalyzing alterations in the molecular structure of the therapeutic drug. Genetic variation in the genes encoding metabolic enzymes leads to differences in enzyme activity that can be characterized as extensive (standard dosage), intermediate (slower than extensive, altered dosage), poor (enzyme deficiency, do not treat with some drugs), and ultrarapid (break down drugs faster than extensive, no effect or reduced effectiveness from drug therapy). One class of metabolic enzymes is the cytochrome P450 superfamily (CYP) that comprises more than 40 isozymes. These enzymes metabolize a large number of drugs, small molecules, mutagens, and carcinogens by modifying parent molecule functional groups. Polymorphisms in the gene CYP2D6 can result in a homozygous recessive inactive genotype that cannot convert codeine into the active metabolite morphine (Rabinowitz and Poljak, 2003). In 2005, the FDAapproved Roche Diagnostic’s AmpliChip CYP450 test, a P450 molecular array which detects mutations in both the CYP2D6 and CYP2C19 genes, which contribute to the metabolism of about a quarter of all prescribed drugs. The data provided by the AmpliChip aids physicians in determining drug selection and dosage. An example of the potential impact of CYP450 typing is illustrated by the drug warfarin which is metabolized by enzymes encoded by the gene CYP2C9. Polymorphisms in this gene influence warfarin dosing requirements and warfarin-associated bleeding (Wittkowsky, 2002). It has been demonstrated that the maintenance warfarin dose can be estimated from demographic, clinical, and pharmacogenetic factors (Gage et al., 2004). A second gene, VKORC1, has been associated with
372
CHAPTER 31
■
Translating Innovation in Diagnostics: Challenges and Opportunities
the average weekly warfarin dose required to maintain patients at their desired anticoagulation target. With more information it is expected that predictive models for warfarin dosing can be developed and help the more than one million patients which each year take warfarin at appropriate dosage to avoid thromboembolic events (Li et al., 2006). Twelve percent of those users currently experience major bleeding episodes and death results in as many as 2%. b) HLA Genotyping
Abacavir is a nucleoside analog that is a potent inhibitor of the HIV reverse transcriptase enzyme. However, between 4–8% of patients develop hypersensitive reactions which can result in life threatening hypotension. The possession of human leukocyte antigen (HLA) B*5701 is a risk factor for abacavir hypersensitivity. Use of HLA-B*5701 genotype screening has been shown to effectively reduce the incidence of abacavir hypersensitivity through the exclusion of patients positive for this allele from abacavir combination treatment (Mallal et al., 2008). Adoption of this pharmacogenomic test was impacted by drug safety issues and the demonstration of its cost-effectiveness (Hughes et al., 2004). Additional associations between specific HLA genotypes and life threatening ADRs include Allopurinol and HLA-B*5801 (Hung et al., 2005), and carbamazeprine and HLA-B*1502 (Dainichi et al., 2007). In December, 2007, the FDA released guidelines calling for genetic testing of the HLAB*1502 genotype prior to starting carbamazeprine therapy for patients with Asian ancestry (FDA, 2007e). c) Cancer Pharmacogenomics
Cancer pharmacogenomics impacts patient safety and therapeutic efficacy and can be used to select treatment choice based on tumor genomics and patient genotype. Methodologies such as molecular cytogenetics, somatic mutation detection, and gene expression profiling have been used to examine genotypic differences in cancer tissue and in patient response to therapy. Chronic myelogenous leukemia (CML) is one of the most common forms of leukemia. Most cases of CML result from a chromosome abnormality whereby DNA from chromosome 9, which contains most of the proto-oncogene c-abl, is translocated onto chromosome 22 within the breakpoint cluster region (BCR) gene (Billings and Brown, 2004). This results in a gene fusion constitutive for expression of a protein with tyrosine kinase activity. This activity affects intracellular signaling pathways that result in unregulated cell proliferation. Molecular diagnosis of CML utilizes quantitative PCR. Fluorescence in situ hybridization (FISH) can also be used to visualize the translocated chromosomes. The oral drug imatinib (Gleevec) was specifically designed as a selective inhibitor of the BCR-ABL tyrosine kinase and has demonstrated therapeutic superiority over conventional drug therapy. Molecular diagnostic testing can be used to monitor responsiveness to Gleevec or the development of therapeutic resistance. Hereditary nonpolyposis colorectal cancer (HNPCC) is the most common hereditary cause of colon cancer, accounting for about 2–5% of all colon cancer cases (see Chapter 73).
It is caused by mutations in any of at least five DNA mismatch repair genes, and DNA tests are available for the most common genes. The HNPCC syndrome predisposes a person to developing colon cancer at a young age. Presymptomatic and predispositional testing in families have been conducted. Since approximately 90% of tumors from HNPCC patients show microsatellite instability, testing for this phenomenon alone can be a good guide to the necessity for further molecular characterization. Gene sequencing can be used to identify the precise mutation. Mutational data, along with DNA microsatellite instability testing results, can be used to identify first-degree relatives with HNPCC. Once HNPCC has been implicated from clinical data, immunohistochemistry testing can be conducted on tumor tissue for confirmation of diagnosis. In this way, members of families with HNPCC have a means to know if they need aggressive monitoring and treatment or not. Thiopurines, thioguanine, and mercaptopurine are commonly used anticancer therapeutics. The thiopurine methyltransferase (TPMT) enzyme catalyses the methylation of thiopurines. The TPMT gene is polymorphic, and one in 300 patients is deficient in enzyme activity. At standard doses, this can lead to toxic accumulation of thiopurines, which can be fatal. Three mutations account for the majority of mutant alleles, and genetic testing is available. Children with leukemia who receive these medications are routinely screened for these deficiency genes (Billings and Brown, 2004). In late 2005, the FDA-approved Third Wave Technologies’ Invader UGT1A1 molecular assay, which is used to identify patients who could experience adverse reactions to Camptosar (irinotecan), a colorectal chemotherapy drug (Grebow, 2005). The test detects mutations in the UGT1A1 gene, which produces a metabolizing enzyme active on the therapeutic. This was one of the first pharmacogonomic tests to be FDA approved for use as a companion diagnostic to a specific drug. Finally, genetic testing to predict future disease risk based on an inherited germline mutation is also available. The BRCA1/2 mutations are associated with a higher risk of breast and ovarian cancer. Approximately 60–80% of women with BRCA1/2 will develop breast cancer. BRCA testing, the BRAC Analysis test, is expensive and provides information with limited therapeutic response. Tests such as these can experience initial resistance among insurers to provide reimbursement (Phillips et al., 2004). Adoption of Pharmacogenomic Testing What are the factors that influence the uptake of pharmacogenomic diagnostic testing? Clinical validity and utility need to be demonstrated and communicated to physicians and patients. Rapid, reliable, inexpensive, and easily interpretable molecular tests will increase the clinical use of these diagnostic tests (Shah, 2004). Clinical relevance must also be demonstrated. This may limit the use of this technology to applications where there exists choice of treatments. In addition, due to the screening nature of these tests, cost-effectiveness must be demonstrated (Flowers and Veenstra, 2004). However, determining cost-effectiveness of genetic technology will not be simple, a fact that has led to some opposition to their large scale uptake in healthcare
Novel Diagnostics
(Rogowski, 2007). Additional obstacles to the deployment of pharmacogenomic testing include obtaining reasonable reimbursement for testing, and the education of physicians and patients as to the value of tests and interpretation of the results. Examples of Translational Diagnostics Investigation of specific examples of translational diagnostics can reveal crucial issues and challenges faced as new methods enter health care. HIV Viral Load Testing Prior to the 1990s, therapeutics that truly impacted the course of viral diseases were not available. The symptoms could be affected, but there was no antibiotic equivalent for treating viral diseases. However, by the mid-1990s, pharmaceutical companies had a number of antiviral compounds; some specifically tailored to HIV, well into clinical trials and needed an accurate biomarker to analyze effectiveness of the new therapies. This convergence of therapeutics and diagnostics would lead to the birth of the theranostic, a diagnostic test linked to the application of specific therapies (Warner, 2004). In 1996, Roche Diagnostics began the Roche Amplicor® Access Program to provide two free baseline HIV viral load tests to anyone in the United States with HIV (James, 1996). Reference laboratories helped execute this program, which began the successful uptake of HIV viral load testing in the marketplace. As with traditional diagnostics, molecular diagnostic tests needed to demonstrate utility and performance before their adoption into practice. First, there must be an unmet clinical need, which was clearly the case with HIV. Secondly, with the development of antiviral therapies, HIV viral load testing provided information that resulted in a therapeutic action, and also provided the means to continually monitor the efficacy of the treatment. Without a therapeutic, the test would have provided information that would not have informed the treatment of HIV. Additional factors sped the uptake of HIV viral load testing. Initial testing involved a relatively small group of highly focused clinicians. This made it much simpler to educate physicians in the use of the viral load test for the management of HIV. Highly motivated patients fueled demand for the test. The viral load test had a significant role in early clinical trials of antiviral therapeutics and became an objective measurement tool, even before any protease inhibitors were approved for physician use. Publications on the antiviral therapeutics included the HIV viral load results. Thought leaders who participated in the trials became early adopters of the test, even before the test was FDA approved. The viral load test was put into a kit format and became FDA approved at approximately the same time as the first protease inhibitors became available. Recommendations for viral load testing became part of standard guidelines for treating HIV infection (Chesebro and Everett, 1998). The theranostic became tied to the use of the drug. Roche’s Amplicor Access Program led to rapid test utilization. As volumes increased, manufacturers developed automation. HIV management requires relatively frequent monitoring, as often as four times per year. Therefore, the test was not a “one off ” test as in constitutional
■
373
genetic testing. HIV viral load testing also obtained its own CPT code, facilitating reimbursement. Additionally, the kit manufacturer, the national reference labs, patients, and doctors all worked to ensure that the test was reimbursed. These factors helped make HIV viral load testing cost effective and a foundation of HIV prevention efforts. The combination of viral load testing and new therapeutics transformed HIV from a uniformly fatal disease to a chronic treatable disease. Therefore, HIV has become a disease with an “actionable” result. Recent Centers for Disease Control (CDC) recommendations now call for general population screening for HIV status. This example of translating a diagnostic into patient care was successful because so many of the hallmarks of successful diagnostic tests were in place. There was an unmet clinical need, an actionable result was obtained from testing, the test was ultimately cost effective, high-throughput analysis became available, and reimbursement issues were rapidly resolved. Breast Cancer, HER2 Receptor Assay, and Herceptin Similar to the convergence of HIV therapeutics and the need for HIV viral load assays, the development of the anti-cancer therapeutic Herceptin required a clinical laboratory test to determine Her2 receptor status (Tsongalis and Silverman, 2006). Amplification of the human epidermal growth factor receptor 2 gene (HER2) in primary breast cancer carcinomas had been shown to correlate to poor clinical prognosis for breast cancer patients (Ferretti et al., 2007). The Her2/neu oncogene is active in 25–37% of breast cancers, and overexpression of the HER2 protein on cell surfaces was found to stimulate uncontrolled tumor growth. Herceptin (trastuzumab) is a recombinant DNA-derived monoclonal antibody that binds with high affinity and specificity to the extracellular domain of the HER2 receptor and inhibits the proliferation of HER2 overexpressing tumor cells. Herceptin alone was associated with an objective response in 15% of extensively pretreated patients with metastatic breast cancer overexpressing HER2, and 26% of previously untreated patients (McKeage and Perry, 2002). Cardiac dysfunction occurred in 13% of patients receiving Herceptin and paclitaxel and in 4.7% of patients receiving only Herceptin. Given the high cost of Herceptin therapy, the significant side effects of Herceptin treatment for some patients, and the targeted nature of the therapeutic, it was necessary to restrict therapy to patients expected to respond to treatment. In 1998, the FDA approved both Herceptin and the HerCepTest for the treatment of metastatic breast cancer. HER2 protein is detected primarily by IHC or FISH methods. HercepTest and PathVysion, a FISH assay for HER2 amplification, are now used to select patients for Herceptin therapy. Uptake of diagnostic testing for HER2 receptor status was greatly accelerated due to safety issues associated with the therapy. Additionally, the assay did not require new tissue samples as in many cases archived samples in paraffin blocks were already being sent to reference labs for estrogen and progesterone analysis, and physicians could merely check another box on the test order sheet. Randomized clinical trials have recently demonstrated significant differences in survival when comparing chemotherapy to chemotherapy plus
374
CHAPTER 31
■
Translating Innovation in Diagnostics: Challenges and Opportunities
Herceptin in women with HER2 overexpressing breast cancer in either the metastatic or adjuvant setting (Ferretti et al., 2007). Thus, what was once a prognosticator of poor outcome for the patient has become a predictive marker of response to therapy. Unfortunately, a lack of concordance among detection techniques, different scoring systems used to determine HER2 status, and a lack of lab standardization and quality based on test experience have led to significant variation in HER2 testing (Nelson, 2000). Recent CAP/ASCO guidelines, that include a testing algorithm, QA requirements, lab evaluations, and a scoring system, should help address these issues (CAP/ASCO, 2007). Cervical Cancer, HPV, and Gardisil Human papillomavirus (HPV) is the major cause of cervical cancer, a disease that kills more than 200,000 women worldwide each year (Gottlieb, 2002). In the United States more than 6 million new cases are reported annually and at least 20 million people in this country are already infected (NIAID, 2006). Approximately 40 of the 100 types of HPV virus can be sexually transmitted, but most rarely cause symptoms or disease. Types 6 and 11 are low-cancer risk types that cause genital warts. Lowand high-risk types can cause the growth of abnormal cells which can be detected when a Pap test is done during a gynecologic exam. The Bethesda system divides the most common clearly abnormal Pap results into either: low-grade squamous intraepithelial lesions (LSILs) which are mild cell changes associated with HPV; high-grade squamous intraepithelial lesions (HSILs) which are precancerous cell changes which should be treated by a physician; and cancer. Clinicians managing a notcompletely-normal Pap result, known as ASCUS (atypical squamous cells of undetermined significance), may benefit from the additional information provided by HPV testing. In 2000, the FDA-approved Digene Corporation’s hc2 High-risk HPV DNA test for use in women with abnormal Pap test results. In 2003, the FDA expanded the use of HPV testing in conjunction with the Pap test for routine screening. The test uses hybrid capture technology to directly detect HPV virus DNA and the current system allows high-throughput testing. A woman fitting the appropriate clinical criteria with both a negative Pap and negative HPV DNA test has less than a one in one thousand chance of developing cervical cancer. In 2006, the FDA-approved Merck and Co’s Gardisil, a recombinant vaccine which is designed to prevent the majority of HPV-related clinical diseases, those caused by HPV 6, 11, 16, and 18. HPV types 16 and 18 account for approximately 70% of cases of cervical cancer, while HPV 6 and 11 cause approximately 90% of genital wart cases. In clinical studies, Gardasil prevented 100% of HPV 16- and 18-related cervical cancer in women not previously exposed to the relevant HPV types. These studies were conducted on 21,000 women ages 16–26 and the vaccine was nearly 100% effective in preventing precancerous lesions and genital warts. Gardisil was evaluated and approved in 6 months under the FDA’s priority review process, a process for products with potential to provide significant health benefits (FDA, 2006f ). On June 28, 2006 the CDC’s Advisory Committee on Immunization Practices (ACIP) recommended that all females
between the ages of 11–26 receive the HPV vaccine as part of routine primary care. HPV diagnostic tests expanded in utility from a test limited to abnormal Pap test results, to a diagnostic test that in conjunction with the Pap test is now standard of care for routine screening for women 30 years of age or older. The move to a liquid Pap test from previously common smear methods removed the need for additional sampling for HPV testing and the development of an effective vaccine bolstered the public awareness of HPV. Technological innovations reducing sampling barriers and improving therapeutic options along with other factors aided the rapid uptake of HPV testing.
CONCLUSIONS: TRANSLATIONAL CHALLENGES FOR INNOVATIVE DIAGNOSTICS Diagnostic tests are being developed at a rapid rate, and the technologies of diagnostic testing are expanding. However, not all diagnostic tests find their way to application in health care. For some, clinical utility may not be demonstrated due to the lack of an actionable result, cost-effectiveness, high-throughput methods, or adequate reimbursement. For example, early asymptomatic glaucoma testing was widely abandoned because early detection did not affect the outcome. Other tests may not achieve the analytical performance characteristics required of a good diagnostic. Assay sensitivity, specificity, and predictive value must be high (Gaeta, 2005). Reproducible precision and a clinically reportable range must be demonstrated (Irwig, et al., 2002). For example, before microarray technology can be translated widely to diagnostic applications, problems with sensitivity and reproducibility will have to solved (Petrik, 2006). A rigorous evaluation of diagnostic tests prior to their introduction into clinical practice is the goal of the standards for the reporting of diagnostic accuracy (STARD) initiative (Bossuyt et al., 2003) and evaluation of genomic applications in practice and prevention (EGAPP), a CDC project to establish an evidence-based process for assessing applications of genomic technology (CDC and National Office of Public Health Genomics, 2008). Ultimately, the role of diagnostics in clinical care must be balanced with other tools available to the physician, such as patient history and physical exam (Halkin et al., 1997). Additionally, technological advancements must produce an increase in efficacy of treatment in excess of that produced by improving the delivery of older treatments (Woolf and Johnson, 2005). The future of innovative diagnostics holds great promise for providing rapid identification of disease susceptibility and status, for monitoring disease progress and therapeutic efficacy, for reducing ADRs and speeding appropriate therapy selection, for the development of personalized medicine, and for optimizing disease prevention and cures. To see this promise unfold, a number of challenges must be overcome. Diagnostic testing must address an unmet medical need, lead to an actionable event, and demonstrate its clinical utility. These diagnostics must also demonstrate high specificity, sensitivity, and predictive value.
References
T A B L E 3 1 . 2 Features of successfully translated diagnostic tests and technologies ● ● ● ● ● ● ● ● ● ● ● ● ●
Addresses an unmet medical need and leads to an actionable result Promotes patient safety and/or stratifies patient risk Highly sensitive and specific, with significant positive or negative predictive value Monitors disease progress or treatment efficacy Speeds clinical trials of innovative therapeutic treatments Clinical validity and utility confirmed in multiple peer reviewed publications Demonstrable cost-effectiveness leading to CPT coding and reimbursement Amenable to automation and easy clinical sample acquisition Elicits patient advocacy and thought leader support Included in guidelines developed by professional societies or government agencies Easy technical adoption by high volume reference labs Favorable reviews by technology assessment groups Obtains FDA clearance or approval, if applicable
For a novel diagnostic to be ultimately translated effectively, additional obstacles must also be overcome. The diagnostic test must prove functional in the health care setting. For example, do new sample types need to be procured, or can the diagnostic test replace or add on to existing assays while retaining current sampling procedures? Will the clinical value of the test be clearly demonstrated to thought leaders, early adopters, patient advocacy groups, and other stakeholder; required for rapid uptake of the diagnostic? The role of professional society guidelines, CMS coverage, technology review by the insurance industry, and regulators must be considered. Will reimbursement issues delay the uptake and access of the test to a large sector of the patient population? Are there patient safety issues, within specific treatments, that warrant rapid clinical uptake? What are the training requirements? How does the clinical lab select the appropriate technology from a pool of technical approaches to the diagnostic problem? What is the best way to deliver the diagnostic to healthcare providers and users? How should a new diagnostic test best be used to optimize the healthcare benefit, as a stand alone diagnostic, or linked to a treatment protocol (a companion diagnostic or theranostics)? Finally, will new accurate and reliable diagnostic methods make the biases and inaccuracies that have characterized the practice of clinical medicine and patient reports a thing of the past? And, if they do, what will be lost and what will be gained?
TABLE 31.3 ● ● ● ● ● ● ● ●
● ●
■
375
Improving diagnostic translation
Document health value of diagnostic innovation Enhance support for translational research Identify gaps and key role of technological innovation in their reduction Speed development of innovation and delivery ease Increase multidisciplinary communication and training Reduce regulatory hurdles Enhance clinical trials for new test validity, utility and health economic outcomes Improve communication of diagnostic test functionality to thought leaders, patient advocacy groups, early adopters, regulatory agencies, and payers Support test standardization, quality assurance standards, and laboratory self-evaluations Improve awareness of the lessons learned from previously translated diagnostics
These are crucial concerns and challenges that must be addressed for virtually any new diagnostic to reach its full potential to improve health care. Closing the translational gaps highlighted by these issues and others requires research, its own knowledge translation, financial support, and an awareness of the lessons learned from previously translated diagnostics (Table 31.2). Even though diagnostics influence 60–70% of healthcare decision-making, they make up less than 5% of hospital costs and 1.6% of Medicare costs (Nordhoff, 2005). In the next decade, new diagnostics are predicted to have an even greater impact on healthcare decision making by providing improved disease prediction and prognostication, and therapy guidance. Studies have found a 30–50% reduction in direct hospital and outpatient charges when changes in patient health status are accurately monitored.Yet, diagnostic tests recommended as standard of care are underused 51% of the time (Olsen, 2006). Low compliance with diagnostics-based quality measures for diabetes, cardiovascular disease, colorectal cancer, and breast cancer can be linked to 34,000 avoidable deaths and $899 million in avoidable healthcare costs (Olsen, 2006). Efficient translation of new knowledge and technology is critical so that the promise of innovative diagnostics and personalized medicine can be realized. Improving the translation of innovative diagnostics will require not only an understanding of specific lessons learned from previous translations, but also an understanding of the conceptual framework within which diagnostic translations occur (Table 31.3).
REFERENCES Billings, P.R. and Brown, M.P. (2004). The future of clinical laboratory genomics. MLO, 8–15. Bossuyt, P., Reitsma, J.B., Bruns, D.E., Gatsonis, C.A., Glasziuo, P.P., Irwig, L.M., Lijmer, J.G., Moher, D., Rennie, D. and de Vet, H.C.W. (2003). Towards complete and accurate reporting of studies of diagnostic accuracy: The STARD initiative. BMJ 326, 41–44.
Brown, I.H. (2006). Advances in molecular diagnostics for avian influenza. Dev Biol (Basel) 124, 93–97. CAP/ASCO. (2007). HER2 and You: Guidelines provided by CAP and ASCO: http://www.cap.org/apps/docs/diseases/cancer/ AboutHer2Testing.pdf. CDC, National Office of Public Health Genomics. (2008). (http:// www.cdc.gov/ genomics/gtesting/EGAPP/about.htm).
376
CHAPTER 31
■
Translating Innovation in Diagnostics: Challenges and Opportunities
Chesebro, M.J. and Everett, D.E. (1998). Understanding the guidelines for treating HIV disease. Am Fam Physician 57(2), 315–322, 328. Dainichi, T., Uchi, H., Moroi, Y. and Furue, M. (2007). Stevens-Johnson syndrome, drug-induced hypersensitivity syndrome, and toxic epidermal necrolysis caused by allopurinol in patients with a common HLA allele: What causes the diversity?. Dermatology 215, 86–88. Demidov, V.V. (2004). Nanobiosensors and molecular diagnostics: A promising partnership. Expert Rev Mol Diagn 4(3), 267–268. Dietel, M. and Sers, C. (2006). Personalized medicine and development of targeted therapies: The upcoming challenges for diagnostic molecular pathology – A review. Vinchows Arch 448, 744–755. Esteller, M. (2002). CpG hypermethylation and tumor suppressor genes: A booming present, a bright future. Oncogene 21(35), 5427–5440. FDA. (2006a). FDA news: FDA unveils Critical Path Opportunities List outlining blueprint to modernize medical product development by 2010 (http://www.fda.gov/bbs/topics/news 2006/NEW01336. html). FDA. (2006b). Critical Path Opportunities initiated during 2006 (http://www.fda.gov/oc initiatives/criticalpath/opportunities06. html). FDA. (2006c). Critical Path Opportunities List (http://www.fda.gov/ oc/initiatives /criticalpath/reports/opp_report.pdf ). FDA. (2006d). Draft guidance for industry, clinical laboratories, and FDA staff: in vitro diagnostic multivariate index assays (http:// www.fda.gov/cdrh/oivd/guidence/1610.pdf ). FDA. (2007e). Information for healthcare professionals: Carbamazepine (marketed as Carbatrol, Equetro, Tegretol, and generics). FDA Alert: http://www.fda.gov/cder/drug/infopage/carbamazepine/default. htm. FDA. (2006f ). FDA Licenses New Vaccine for Prevention of Cervical Cancer and Other Diseases in Females Caused by Human Papillomavirus (http://www.fda.gov/bbs/topics/NEWS/2006/ NEW01385.html). Ferretti, G., Felici, A., Papaldo, P., Fabi, A. and Cognetti, F. (2007). HER2/neu role in breast cancer: From a prognostic foe to a predictive friend. Curr Opin Obstet Gynecol 19(1), 56–62. Flowers, C.R. and Veenstra, D. (2004). The role of cost-effectiveness analysis in the era of Pharmacogenomics. Pharmacoecon 22(8), 481–493. Gabriele, L., Moretti, F., Pierotti, M.A., Marincola, F.M., Foa, R. and Belardelli, F.M. (2006). The use of microarray technologies in clinical oncology. J Transl Med 4, 8–12. Gaeta, T. (2005) Screening and diagnostic tests (http://www.emedicine. com/emerg/topic779.htm). Gage, B.F., Eby, C., Milligan, P.E., Banet, G.A., Duncan, J.R. and McLeod, H.L. (2004). Use of pharmacogenetics and clinical factors to predict the maintenance dose of warfarin. Thromb Haemost 91, 87–94. Gottlieb, N. (2002) A primer on HPV (http://newscenter.cancer.gov/ newscenter/ benchmarks-vol2-issue4/page2). Graham, I.D., Logan, J., Harrison, M.B., Straus, S.E.,Tetroe, J., Caswell,W. and Robinson, N. (2006). Lost in knowledge translation: Time for a map? J Contin Ed Health Prof 26, 13–24. Grebow, M. (2005). Third wave pharmacogenetic test cleared by FDA. IVD Technol Nov, 14. Gulmann, C., Sheehan, K.M., Kay, E.W., Liotta, L.A. and Petricoin, E.F. (2006). Array-based proteomics: Mapping of protein circuitries for diagnostics, prognostics, and therapy guidance in cancer. J Pathol 208, 595–606.
Halkin, A., Reichman, J., Schwaber, M., Paltiel, O. and Brezis, M. (1997). Likelihood ratios: Getting diagnostic testing into perspective. Q J Med 91, 247–258. Horig, H. and Pullman, W. (2004). From bench to clinic and back: Perspective on the 1st IQPC translational research conference. J Transl Med 2, 44–51. Horig, H., Marincola, E. and Marincola, F.M. (2005). Obstacles and opportunities in translational research. Nat Med 11(7), 705–708. Hughes, D.A., Vilar, F.J., Ward, C.C., Alfirevic, A., Park, B.K. and Pirmohamed, M. (2004). Cost-effectiveness analysis of HLAB*5701 genotyping in preventing abacavir hypersensitivity. Pharmacogenet 14(6), 335–342. Humes, H.D. (2005). Translational medicine and the National Institutes of Health road map: Steep grades and tortuous curves. J Lab Clin Med 146, 51–54. Hung, S.-L., Chung, W.-H., Liou, L.-B., Chu, C.-C., Lin., M. et al. (2005). HLA-B*5801 allele as a genetic marker for severe cutaneous adverse reactions caused by allopurinol. PNAS 102(11), 4134–4139. Irwig, L., Bossuyt, P., Glasziou, P., Gatsonis, C. and Lijmer, J. (2002). Evidence base of clinical diagnosis: Designing studies to ensure that estimates of test accuracy are transferable. BMJ 324, 669–671. James, J.S. (1996). Viral load approved; Free test offered to all in U.S. with HIV. (http://www.aids.org/atn/a-248-01.html). Kerner, J.F. (2006). Knowledge translation versus knowledge integration: A “funders” perspective. J Contin Ed Health Prof 26, 72–80. Laird, P.W. (2005). Cancer epigenetics. Hum Mol Genet 14(Review Issue 1), R65–R76. Li, T., Lange, L., Li, X., Susswein, L., Bryant, B., Malone, B., Lange, E., Huang, T.-T., Stafford, D. and Evans, J.P. (2006). Polymorphisms in the VKORC1 gene are strongly associated with warfarin dosage requirements in patients receiving anticoagulation. J Med Genet 43(9), 740–744. Mallal, S., Phillips, E., Carosi, G. et al. (2008). HLA-B*5701 screening for hypersensitivity to Abacavir. NEJM 358, 568–579. Mankoff, S.P., Brander, C., Ferrone, S. and Marincola, F.M. (2004). Lost in translation: Obstacles to translational medicine. J Transl Med 2, 14–18. Marincola, F.M. (2003). Translational medicine: A two-way road. J Transl Med 1, 1–2. McKeage, K. and Perry, C.M. (2002). Trastuzumab: A review of its use in the treatment of metastatic breast cancer overexpressing HER2. Drugs 62(1), 209–243. Mehlmann, M., Bonner, A.B.,Williams, J.V., Dankbar, D.M., Moore, C.L., Kuchta, R.D., Podsiad, A.B., Tamerius, J.D., Dawson, E.D. and Rowlen, K.L. (2007). Comparison of the MChip to viral culture, reverse transcription-PCR and the QuickVue Influenza A B test for rapid diagnosis of influenza. J Clin Microbiol 45(4), 1234–1237. Ming-Cheng Cheng, M., Cuda, G., Bunimovich, Y.L., Gaspari, M., Heath, J.R., Hill, H.D., Mirkin, C.A., Nijdam, A.J., Terracciano, R., Thundat, T. et al. (2006). Nanotechnologies for biomolecular detection and medical diagnostics. Curr Opin Chem Biol 10, 11–19. Miyamoto, S. (2006). Clinical applications of glycomic approaches for the detection of cancer and other diseases. Curr Opin Mol Ther 8(6), 507–513. Montalto, N.J. (2003). An office-based approach to influenza: Clinical diagnosis and laboratory testing. Am Fam Physician 67(1), 111–118. Montgomery, M. and Louie, L. (2001). Pharmacogenomics: What pharmacists need to know. Calif J Health-System Pharm 13(5), 5–12. Morrelle, W. and Michalski, J.C. (2005). Glycomics and mass spectrometry. Curr Pharm Des 11(20), 2615–2645.
References
NCRR. (2006). Clinical and translational science awards to transform clinical research (http://www.ncrr.nih.gov/ncrrprog/roadmap/ CTSA_9-2006.asp). Nelson, N.J. (2000). Experts debate value of HER2 testing. J Natl Cancer Inst 92(4), 292–294. NIAID. (2006). Human Papillomavirus and genital warts (http:// www.newscenter.cancer.gov/http://newscenter.cancer.gov/ newscenter/benchmarks-vol2-issue4/page2). NIH publication number 02-4987. (2002). Deciphering pathogens: Blueprints for new medical tools (http://www.niaid.nih.gov/ publications/pdf/deciphering_pathogens.pdf ). NIH. (2006). NIH roadmap for medical research: re-engineering the clinical research enterprise (http://nihroadmap.nih.gov/ clinicalresearch/overview-translational.asp). Nordhoff, H.L. (2005 Nov.). Commentary: Realizing the potential of molecular diagnostics. IVD Technol, 18–20. Olsen, T. (2006 Sep.). Commentary: How molecular diagnostics has responded to industry trends. IVD Technol, 20–22. Paxton, A. (2004). HIV, HCV windows nearly shut with minipool NAT. CAP Today 6, 1–4. Petrik, J. (2006). Diagnostic applications of microarrays. Transfus Med 16, 233–247. Phillips, K.A., Veenstra, D.L., Ramsey, S.D., Van Bebber, S.L. and Sakowski, J. (2004). Genetic testing and pharmacogenomics: Issues for determining the impact to healthcare delivery and costs. Am J Manag Care 10, 425–432. Puderbaugh, A. (2006). NIH launches national consortium to transform clinical research. NIH Record LVIII(23), 3 (http://www.nih.gov/ nihrecord/11_17_2006/11172006 _Record.pdf ). Rabinowitz, P.M. and Poljak, A. (2003). Host-environment medicine: A primer on medical genomics part VI: Genomics and molecular genetics in clinical practice. Mayo Clin Proc 78(3), 307–317. Ratner, M. (2006). Looking for solid ground along the critical path. Nat Biotechnol 24(8), 885–887. Reece, E.A. (2006). A clarion call for translational and collaborative research. Am J Obstet Gynecol 194, 1507–1509.
■
377
Rice, G.E., Georgiou, H.M., Ahmed, N., Shi, G. and Kruppa, G. (2006). Translational proteomics: Developing a predictive capacity – A review. Placenta Trophoblast Research 20(27: Suppl. A), S76–S86. Rogowski, W. (2007). Current impact of gene technology on healthcare: A map of economic assessments. Health Policy 80, 340–357. Schwaiger, M. and Peschel, C. (2006). Biological imaging for selecting and monitoring cancer therapy; a pathway to individualized therapy. Eur J Nucl Med Mol Imaging 33, S1–S5. Shah, J. (2004). Recent developments: Criteria influencing the clinical uptake of pharmacogenomic strategies. BMJ 328, 1482–1486. Sonntag, K.-C. (2005). Implementations of translational medicine. J Transl Med 3, 33–35. Tefferi, A., Bolander, M.E., Ansell, S.M.,Wieben, E.D. and Spelsberg,T.C. (2002). Primer on medical genomics part III: Microarray experiments and data analysis. Mayo Clin Proc 77, 927–940. Tsongalis, G.J. (2005). Anniversary essays: Molecular diagnostics. IVD Tech, 62–64. Tsongalis, G.J. and Silverman, L.M. (2006). Molecular diagnostics: A historical perspective. Clinica Chemica Acta 369, 188–192. Vincart, B., De Mendonca, R., Rottiers, S.,Vermeulen, F., Struelens, M.J. and Denis, O. (2007). A specific real-time PCR assay for the detection of Bordetella pertussis. J Med Microbiol 56, 918–920. Warner, S. (2004). Diagnostics Therapy Theranostics. Scientist 18(16), 38. Wittkowsky, A.K. (2002). Pharmacogenomics of warfarin. Anticoag Forum 6(3), 1–2. Woolf , S.H. and Johnson, R.E. (2005). The break-even point: When medical advances are less important than improving the fidelity with which they are delivered. Ann Fam Med 3, 545–552. Zhu, H., Hu, S., Jona, G., Zhu, X., Krieswirth, N., Willey, B.M., Mazzulli, T., Liu, G., Song, Q., Chen, P. et al. (2006). Severe acute respiratory syndrome diagnostics using a Coronavirus protein microarray. PNAS 103(11), 4011–4016.
CHAPTER
32 The Role of Genomics in Enabling Prospective Health Care Ralph Snyderman
INTRODUCTION In the early 1900s, the emerging sciences of physiology, chemistry, immunology, microbiology, and radiology began to be introduced into the practice of medicine, which at that time, considered disease to be due to imbalances of bodily humors and transmitted through climacteric miasmas. For the first time in history, the application of science to medicine laid a solid foundation for understanding the pathophysiology of disease (Bynum, 2002; Snyderman, 2004; Snyderman and Williams, 2003). In particular, microbiology, through the identification of “causative” agents for numerous infectious diseases, had a profound impact on concepts concerning the cause and potential cure of illnesses. Given the increasing number of diseases determined to have specific “causes” along with the new found ability of chemistry to synthesize therapeutic “magic bullets,” early 20th Century medicine logically pursued a scientific, reductionist approach to health care which continues today: that is, identify a disease’s root cause and eliminate it. This approach has enabled medicine to reverse some diseases, prolong life, and at times, effect wondrous cures. The reductionist approach to medicine, however, is limited in that the evolution of virtually all diseases is more complicated than a single root cause. The development of disease is based upon one’s genetic susceptibilities, complex interactions between initiating factors, and health status. In the aggregate, these can lead to disease progression, or not (Figure 32.1). For example,
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 378
while the tubercle bacillus is the causative agent of tuberculosis, the clinical manifestation of exposure, as a consequence of host susceptibility, can range from transient subclinical infection to rapidly fatal miliary disease. Even a well-defined genetic disease such as sickle cell anemia, known to be due to a single mutation, can have manifestation ranging from death in adolescence to far milder forms compatible with long-term survival. By focusing primarily on the reductionist model of disease development, our health care system has, unfortunately, not focused on the prevention or treatment of complex chronic diseases. At present, individuals with any of five chronic conditions account for roughly two-thirds of all health care expenses (Figure 32.2). A more rational approach to health care is now possible. New emerging sciences, notably genomics, can transform medical practice now just as new applications of science and know-how transformed medicine 100 years ago (Figure 32.3). Genomic and related expanding fields of research can impact health care in numerous ways (Figure 32.4), but most profound may be their ability to predict risk, track disease progression and anticipate clinical events, thereby enabling truly personalized, predictive, preventative care. To do this, the strengths of today’s diseaseoriented reductionist approach must be integrated with an approach that focuses on prevention and minimization of disease with an emphasis on long-term strategies. This personalized, predictive, and preventative approach to health care is known as “prospective health care” (Snyderman and
Copyright © 2009, Elsevier Inc. All rights reserved.
Introduction
■
379
Reductionism: Single factor (20th Century) Causative factor
Disease
Emergence: Multiple factors (21st Century) Baseline risk Preclinical progression
Disease initiation
Disease progression
Irreversible damage
Environmental factors
Figure 32.1 Concept of disease: reductionism versus emergence. The reductionist approach considers disease to be a consequence of a pathogenic factor; that is microbe. More accurately, disease development depends on the host’s susceptibility to pathogenic factors and exposure to them (Snyderman and Yoediono, 2006).
Total US health care spending 2002: $1.5 trillion $1.6T
$1.5T
$1.2T
All other care
$0.8T
$0.4T
Treating chronic disease
Leading chronic diseases
Population affected
National bill
Diabetes
16 million
$44–$98B
Cancer
15 million
$37–$107B
Congestive heart failure
5 million
$20–$40B
Respiratory disease
30 million
$24–$36B
Hypertension
50 million
$23–$33B
US healthcare spending
Initiating events
Earliest molecular detection
Earliest clinical detection
Typical current intervention Cost
Disease burden
Baseline risk
1/reversibility
Figure 32.2 Consequences of current approaches to health care. While little is spent on prevention, nearly three-quarters of health care expenditures fund the treatment of late-stage chronic disease. (Source: American Heart Association, American Cancer Society, American Lung Association, National Institute of Diabetes and Digestive and Kidney Disease.)
Time
Figure 32.3 Disease progression. Diseases develop over time, with pathology increasing, reversibility decreasing and cost of care increasing (Snyderman and Yoediono, 2006).
380
CHAPTER 32
■
Technologies
The Role of Genomics in Enabling Prospective Health Care
Impact
PREDICTIVE MODELS • Genomics
• Disease risk prediction
• Proteomics
• Quantification disease burden
• Metabolomics
• Tracking of pathogenses
• Medical technologies
• Disease event prediction
• Informatics
• Therapeutic evaluation
• Systems Biology
Figure 32.4 Impact of new technologies on prospective health care. Last century’s science enabled reactive responses to disease while this century’s science enables prospective approaches as well.
Traditional medical evaluation and record • Chief complaint • History of illness • Past medical history • Family history • Social history • Physical exam • Diagnostic tests • Assessment and plan
Prospective evaluation and record
At the core of prospective health care are personalized health risk assessment and predictive modeling. A predictive model uses statistical algorithms to identify factors that predict events (Anderson et al., 1991). The predictive model can be used to forecast events based on factors or biomarkers, which are most likely to correlate with or be causative of the future clinical event. The development of a predictive model uses various methods, such as logistical regression or neural network, to differentiate predictive factors from other variables which are not as useful for anticipating the clinical outcome of interest. The predictive model can then be applied to a current situation to determine the risk and timing of an event occurring. The better the model, the better the accuracy in predicting the occurrence and timing of the event. The details involved in developing a predictive model, as well as other related components such as individualized patient databases and risk model libraries, are beyond the scope of this chapter, but a simplified version is shown in Figure 32.6.
• Health risk assessment • Current health status • Disease burden and pathogenesis tracking • Clinical event prediction • Therapeutic plans and evaluation
Figure 32.5 Current medical record versus prospective approach to health care. Physicians currently evaluate patients using a disease-focused approach. A more effective approach would include strategic health planning as indicated by the prospective approaches shown on the right (Snyderman and Yoediono, 2006).
Williams, 2003; Snyderman and Yoediono, 2006;West et al., 2006; Williams et al., 2003). Key features of prospective health care include personalized health risk assessment, tracking of pathogeneses, prediction of clinical events, and strategic planning to mitigate risks and evaluate therapeutic benefits. With a prospective health care model, individuals will be provided with personalized health plans which promote strategic health planning by incorporating individualized risk assessments for diseases, disease tracking, clinical event predictions and therapeutic planning (Figure 32.5). Genomic research will play an integral role in providing these capabilities given that the molecular processes underlying disease susceptibility, progression, and therapeutic response differ among individuals and can be measured, in part, through genomic analyses. Technological advances allowing the measurements at the DNA, mRNA, protein, and metabolic levels are already beginning to enable inferences regarding health risks or clinical outcomes (Snyderman and Langheier, 2006).
PREDICTIVE FACTORS The predictive factors most commonly used for disease-related risk assessments encompass clinical, demographic, family history, and laboratory data. These types of data can provide insight into the likelihood of an individual developing a condition or event and are relatively cost-effective and collected routinely. However, generally they are limited in terms of being able to accurately predict when disease will occur. Furthermore, many false-positives and false-negatives are associated with these types of data, which are often epidemiologic in nature (i.e., Framingham Study) (Brindle et al., 2003; The International HapMap Consortium, 2003). Predictive factors, which serve as more accurate markers of disease susceptibility, likely will come in the form of biomarkers which are directly related to the cause of the clinical event. A biomarker is a characteristic (i.e., expressed gene, protein or metabolite) that is objectively measured and evaluated as an indicator of biological or pathogenic processes or pharmacological responses to therapeutic interventions. Genomic research will play an integral role in identifying predictive biomarkers which, in combination with other types of predictive factors, will enable more accurate risk analyses for baseline risk assessment, disease tracking, clinical event predictions and response to therapeutics. Biomarkers can be thought of as belonging to two general categories: (1) stable biomarkers that are inherited and/or change rarely over a lifetime, and (2) dynamic biomarkers which express ongoing biological activities.The latter include gene expression, protein expression and the measurement of metabolic factors. Measurement of genomic factors that lie in the causal pathway of a disease or a therapeutic response, or factors such as single-nucleotide polymorphisms (SNPs) that are highly associated with causal genes, will serve as better predictors of adverse outcomes than much of the demographic data now being collected. Stable DNA gene predictors will enhance baseline clinical risk assessment and primary prevention, while dynamic mRNA, protein and metabolic factors will reflect ongoing biological or pathological processes.
Predictive Factors
■
381
Out of sample populations
Clinical data cohort Data Clinical
Genetic
Proteomic
Other Prediction models Statistical validation
Baseline risk Disease progression Event prediction Decision support
y f (x1, x2,…xn) Predictive model
Patient’s data
Predictive modeler
Additional data New biomarkers
Patient
Physician
Figure 32.6 Methodology for creating clinical risk predictive models. Relevant clinical data and/or biomarkers are statistically analyzed to create validated risk models for particular clinical conditions or events (Snyderman and Yoediono, 2006).
Assess risk
Refine assessment
Initiating events
Earliest molecular detection
Earliest clinical detection
Monitor progression Predict events Inform therapeutics Typical current intervention Cost
Disease burden
Baseline risk
Predict/diagnose
1/reversibility
Decision support tools:
Time Baseline risk Sources of new biomarkers:
Figure 32.7
Stable genomics: • Single-nucleotide • Polymorphisms (SNPs) • Haplotype mapping • Gene sequencing
Preclinical progression Dynamic genomics: • Gene expression • Proteomics • Metabolomics • Molecular imaging • Clinical risk models
Disease initiation and progression Therapeutic decision support
Contributions of new technologies to disease prediction and prediction of clinical events (Snyderman and Yoediono, 2006).
Analysis of these will enhance refined risk assessments to track disease progression, predict events, and guide therapeutic choices (Figure 32.7). The advantage of genotypic data for baseline prediction is that they can be collected at birth or any time during one’s life
and, theoretically, needs to be collected only once. Baseline risk assessments using demographic data or static genomic information will likely have lower specificity (a higher number of false positives) than molecular measures that are dynamic and change as a function of time. Nonetheless, as more information regarding
382
CHAPTER 32
■
The Role of Genomics in Enabling Prospective Health Care
inherited risks is gathered, baseline risk assessment will become increasingly useful. For secondary prevention (for example, predicting heart attack in an individual with diabetes), stable genomic data may be less valuable. In this case, relevant dynamic indicators will provide a more powerful predictor of the disease event. Over time, the cost of genotyping should plummet and the value of baseline risk analysis should increase markedly. Moreover, the identification of clinically valuable dynamic biomarkers is an area of fertile research and will enhance accurate ongoing risk assessment timing of disease events and prediction of therapeutic outcomes. Identifying accurate predictors for baseline risk assessment will be facilitated greatly by clinical research and the HapMap project and the Genetic Association Information Network (GAIN) initiative, a public/private partnership between the NIH, foundation of NIH and industry. The GAIN study will fund $20 million of research to identify SNPs associated with common diseases. The International HapMap Consortium is characterizing common patterns of DNA sequence variation and the extent of linkage disequilibrium in the human genome. This will facilitate the characterization of genotypes and identification of key SNPs related to chronic disease; traditional and advanced association algorithms will allow the analysis of the HapMap (International HapMap Project, http://www.hapmap.org)(Niu, 2004; Liu et al., 2004; Evans et al., 2004). Online Mendelian Inheritance in Man (OMIM), a database of disease risk genes, is already revealing an increasing number of stable genomic factors that should be useful in risk assessment (Online Mendelian Inheritance in Man, http:// www.ncbi.nlm.nih.gov/entrez/query.fcgi?dbOMIM). The role of an individual’s gene variants in altering their metabolism and response to drugs is becoming important in drug development and in certain areas of clinical practice, particularly oncology (Lee et al., 2005). For individuals whose genes, SNPs, family history, or clinical information identify them as high risk for a particular disease, surveillance will be needed to track possible disease progression and, when relevant, to evaluate therapeutic support. Such tracking will likely include the measurement of dynamic factors, including gene-expression, proteomics, and metabolomic assessments. Analyses to track baseline risk and disease development will hopefully be incorporated into personalized health plans in the future (Langheier and Snyderman, 2004). For example, children with a family history of type 1 diabetes could have a baseline risk assessment that evaluates various predictive SNPs. Those determined to be at high risk could undergo a surveillance protocol, tracking their levels of biomarkers associated with actual disease development and progression. (Barker et al., 2004; Eisenbarth, 2004). As effective preventative therapies are identified, these analyses could guide appropriate intervention before -cells are destroyed. Initial applications of genomic technologies are being applied to predict outcomes in defined clinical conditions. For example, gene-expression microarrays and proteomic techniques show promise for identifying the aggressiveness of cancer, allowing the creation of predictive models for likely survival time with and
without treatment (Anderson and LaBaer, 2005; Berchuck et al., 2005; Pittman et al., 2004; Rich et al., 2005). Moreover, gene expression in circulating mononuclear cells is being used to predict organ rejection in patients with heart transplants, obviating the need for myocardial biopsy in some conditions (Deng et al., 2006). Many gene-expression tests are being developed to gauge appropriate chemo-therapeutic regimens for a given patient’s cancer (Bild et al., 2006).
RISK ASSESSMENT FOR BREAST CANCER Breast cancer provides a useful example for how genomic research and predictive models can improve clinical care. For baseline risk measurement, a tool was developed in 1989 to estimate the likelihood that a woman at a given age with defined risk factors will develop breast cancer over a specified time (Gail et al., 1989). This Gail breast cancer model aids physicians in developing a personalized strategy for screening and treatment. The model was constructed from case-control data of the Breast Cancer Detection Demonstration Project and included age at menarche, age at first live birth, number of previous biopsies, and number of first-degree relatives with breast cancer as indicators. Newer and improved predictive models include more robust family history (i.e., the Claus model) (Santen et al., 2007) and causal disease genotypes such as BRCA1 or BRCA2 (for example, in the BRCAPRO model) (Euhus, 2001) as predictors. In contrast to the Gail model, which uses logistic regression, the Claus model uses genetic modeling to determine age-specific breast cancer development probabilities from family history. BRCAPRO employs a Bayesian model, focused on BRCA1 and BRCA2 to determine the risk of breast cancer. Accurate prediction of breast cancer will require continued research. While the incorporation of BRCA1 and BRCA2 disease alleles as predictors facilitates the risk assessment of cancer, these alleles account for only a small proportion (5%) of breast cancer in the United States. Other genes of interest will certainly be found as breast cancer is a feature of many other syndromes with known genetic mutations, such as, the Li-Fraumeni syndrome (caused by a germline p53 mutation), the Cowden syndrome (a PTEN mutation), and the Peutz-Jegher syndrome (an STK11 mutation) (Garber and Offit, 2005; Nagy et al., 2004). Other genotypes associated with increased risk of breast cancer are located in several genes, including BRCATA on 11q, BRCA3 on 13q21, RB1CC1 on 8q11, BWSCRIA on 11p15.5, and BRIP1 on 17q22 (OMIM #114480 Breast cancer, http://www.ncbi.nlm.nih.gov/ entrez/dispomim.cgi?id114480). Further research may enable baseline analysis of breast cancer risk to be used for primary care screening (Nelson et al., 2005). A validated “SNP chip” to test for the presence of disease genotypes for multiple alleles should help improve the quality of the baseline risk assessments in the broader population (Listgarten et al., 2004). When they become cost-effective, early screening of a broader range of relevant genotypes could be incorporated into
Acknowledgements
personal health plans. Although no high-throughput genotyping tool is currently available for breast-cancer onset prediction, Genomic Health, Inc. has commercialized its “Oncotype Dx” 21-gene predictor of breast cancer recurrence (Paik et al., 2004), and Veridex, LLC has published research on its gene-expression tests, reporting improvements in the accuracy of predicting cancer prognosis (Wang et al., 2005). These enhancements are based on molecular tumor analysis; the Oncotype Dx test has already been used to enhance “Adjuvant Online!,” a predictive model for cancer recurrence and survival (Surveillance epidemiology and end results, http://seer.cancer.gov; Adjuvant! http://www.adjuvantonline.com). Gene-expression strategies are also being made to customize cancer chemotherapy. By analyzing expressed genes in breast cancer specimens, the operative pathogenic pathways are being identified and specific chemotherapies chosen to block them (Bild et al., 2006). These applications indicate that clinicalgenomic predictive models may soon have broad utility in clinical practice.
■
383
major side effect of inducing peripheral neuropathy. Genetic studies have shown that isoniazid-related peripheral neuropathy occurs among individuals with specific alleles for a polymorphic gene for N-acetyltransferase 2, a liver enzyme. Persons with these alleles acetylate isoniazid slowly, which in turn increases their risk for peripheral neuropathy as well as liver damage. Individuals found to have these alleles based on a pharmacogenetic profile, would most likely not receive isoniazid. A “genotyping” microarray chip for the cytochrome P450 (CYP) metabolism genes CYP2D6 and CYP2C19 might effectively aid physicians in assessing genetic susceptibilities to toxicity from a wide range of drugs (http://circ. ahajournals.org/cgi/content/full/101/14/1749; http://www.aafp. org/afp/ 20070801/391.html). Polymorphisms detected in these two enzymes can affect the rate at which an individual metabolizes up to 25% of drugs on the market, including those used to treat cardiovascular disease, high blood pressure, depression, and Attention Deficit Hyperactivity Disorder (ADHD) (http://us. diagnostics.roche.com/press_room/2003/062503.htm Roche Diagnostics – press release; http://www.devicelink.com/ivdt/ archive/03/04/002.html).
PHARMACOGENOMICS Pharmacogenomics determines how genetic inheritance affects an individual’s response to drugs.This field holds the promise that the use of drugs might be adapted to reach individual’s genetic makeup (http://www.ornl.gov/sci/techresources/Human_Genome/ medicine/pharma.shtml). In addition to improving the accuracy of risk assessments related to disease, biomarkers will play a role in prospective health care by enabling individualized therapeutic approaches. Just as there are differences among individuals in regards to disease susceptibility and progression, there are differences among individuals in terms of how they respond to therapeutics. Drug responsiveness is impacted by the interaction among many factors underlying those determined through genomics. From a genetic perspective, individual variations can influence how a drug is absorbed, distributed, metabolized and excreted, which in turn impact how an individual responds to a particular drug. Pharmacogenomics determines how individual genetic differences impact one’s response to a medication (see Chapter 27). Biomarkers associated with drug response could be utilized to develop individualized pharmacogenetic profiles, which would determine an individual’s likelihood of responding to a drug as well as adverse reactions for which an individual could be at high risk. For example, isoniazid, a drug used to treat tuberculosis, has a
CONCLUSION Genomic research is critical to enabling personalized, predictive, and preventative medicine through the identification and validation of predictive biomarkers. Prospective approaches to health care require individualized baseline risk assessments, disease tracking, clinical event prediction and analysis of an individual’s therapeutic response. The application of genomic risk predictive technologies to health care provide a far more detailed understanding of health and its evolution toward disease. Importantly, it will support the ability to predict, and thereby prevent, clinical events through appropriate and personalized interventions. Highly accurate risk assessment is an important component of a shift of disease-based medicine to a more rationale approach to prospective health care.
ACKNOWLEDGEMENTS The author would like to acknowledge the outstanding editorial assistance of Dr Ziggy Yoediono and Ms Cindy Mitchell.
384
CHAPTER 32
■
The Role of Genomics in Enabling Prospective Health Care
REFERENCES Anderson, K.M., Wilson, P.W., Odell, P.M. and Kannel, W.B. (1991). An updated coronary risk profile. A statement for health professionals. Circulation 83, 356–362. Anderson, K.S. and LaBaer, J. (2005). The sentinel within: Exploiting the immune system for cancer biomarkers. J Proteome Res 4, 1123–1133. Barker, J.M., Barriga, K.J., Yu, L., Miao, D., Erlich, H.A., Norris, J.M., Eisenbarth, G.S. and Rewers, M. (2004). Diabetes Autoimmunity Study in the Young. Prediction of autoantibody positivity and progression to type 1 diabetes: Diabetes Autoimmunity Study in the Young (DAISY). J Clin Endocrinol Metab 8, 3896–3902. Berchuck, A., Iversen, E.S., Lancaster, J.M., Pittman, J., Luo, J., Lee, P., Murphy, S., Dressman, H.K., Febbo, P.G., West, M. et al. (2005). Patterns of gene expression that characterize long-term survival in advanced stage serous ovarian cancers. Clin Cancer Res 11, 3686–3696. Bild, A.H., Yao, G., Chang, J.T., Wang, Q., Potti, A., Chasse, D., Joshi, M.B., Harpole, D., Lancaster, J.M., Berchuck, A. et al. (2006). Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439(7074), 274–275. Brindle, P., Emberson, J., Lampe, F., Walker, M., Whincup, P., Fahey, T. et al. (2003). Predictive accuracy of the Framingham coronary risk score in British men: Prospective cohort study. BMJ 327, 1267–1270. Bynum, W. (2002). The evolution of germs and the evolution of disease: Some British debates, 1870–1900. Hist Philos Life Sci 24(1), 53–68. Deng, M.C., Eisen, H.J., Mehra, M.R., Billingham, M., Marboe, C.C., Berry, G., Kobashigawa, J., Johnson, F.L., Starling, R.C., Murali, S. et al. (2006). Non-invasive discrimination of rejection in cardiac allograft recipients using gene expression profiling. Am J Transplant 6, 150–160. Eisenbarth, G.S. (2004). Prediction of type 1 diabetes: The natural history of the prediabetic period. Adv Exp Med Biol 552, 268–290. Euhus, D.M. (2001). Understanding mathematical models for breast cancer risk assessment and counseling. Breast J 7(4), 224–232. Evans, D.M., Cardon, L.R. and Morris, A.P. (2004). Genotype prediction using a dense map of SNPs. Genet Epidemiol 4, 375–384. Gail, M., Brinton, L., Byar, D., Corle, D., Green, S., Schairer, C. and Mulvihill, J. (1989). Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. J Natl Cancer Inst 81, 1879–1886. Garber, J.E. and Offit, K. (2005). Hereditary cancer predisposition syndromes. J Clin Oncol 23, 276–292. Langheier, J.M. and Snyderman, R. (2004). Prospective Medicine: The role for genomics in personalized health planning. Pharmacogenomics 5(1), 1–8. Lee, W., Lockhart, A.C., Kim, R.B. and Rothenberg, M.L. (2005). Cancer pharmacogenomics: Powerful tools in cancer chemotherapy and drug development. Oncologist 2, 104–111. Liu, T., Johnson, J.A., Casella, G. and Wu, R. (2004). Sequencing complex diseases with HapMap. Genetics 168, 503–511. Listgarten, J., Damaraju, S., Poulin, B., Cook, L., Dufour, J., Driga, A., Mackey, J., Wishart, D., Greiner, R. and Zanke, B. (2004). Predictive
models for breast cancer susceptibility from multiple single nucleotide polymorphisms. Clin Cancer Res 10, 2725–2737. Nagy, R., Sweet, K. and Eng, C. (2004). Highly penetrant hereditary cancer syndromes. Oncogene 23, 6445–6470. Nelson, H.D., Huffman, L.H., Fu, R. and Harris, E.L. (2005). U.S. Preventive Services Task Force: Genetic risk assessment and BRCA mutation testing for breast and ovarian cancer susceptibility: Systematic evidence review for the U.S. Preventive Services Task Force. Ann Intern Med 143, 362–379. Niu, T. (2004). Algorithms for inferring haplotypes. Genet Epidemiol 27, 334–347. Paik, S., Shak, S., Tang, G., Kim, C., Baker, J., Cronin, M., Baehner, F.L., Walker, M.G., Watson, D., Park, T. et al. (2004). A multigene assay to predict recurrence of tamoxifen-treated, node-Negative breast cancer. N Engl J Med 351, 2817–2826. Pittman, J., Huang, E., Dressman, H., Horng, C.F., Cheng, S.H., Tsou, M.H., Chen, C.M., Bild, A., Iversen, E.S., Huang, A.T. et al. (2004). Integrated modeling of clinical and gene expression information for personalized prediction of disease outcomes. Proc Natl Acad Sci USA 101, 8431–8436. Rich, J.N., Hans, C., Jones, B., Iversen, E.S., McLendon, R.E., Rasheed, B.K., Dobra, A., Dressman, H.K., Bigner, D.D., Nevins, J.R. et al. (2005). Gene expression profiling and genetic markers in glioblastoma survival. Cancer Res 65, 4051–4058. Santen, R., Boyd, N., Chlebowski, R., Cummings, S., Cuzick, J., Dowsett, M., Easton, D., Forbes, J., Key, T., Hankinson, S. et al. (2007). Critical assessment of new risk factors for breast cancer: Considerations for development of an improved risk prediction model. Endocr Relat Cancer 14(2), 169–187. Snyderman, R. (2004). AAP Presidential address: The AAP and the transformation of medicine. J. Clin Invest 114, 1169–1173. Snyderman, R. and Langheier, J. (2006). Prospective health care: The second transformation of medicine. Genome Biol 7, 104–127. Snyderman, R. and Williams, R.S. (2003). Prospective medicine: The next health care transformation. Acad Med 78(11), 1079–1084. Snyderman, R. and Yoediono, Z. (2006). Prospective care: A personalized, preventative approach to medicine. Pharmocogenomics 7(1), 5–9. The International HapMap Consortium (2003). The International HapMap Project. Nature 426, 789–796. Wang, Y., Klijn, J.G., Zhang, Y., Sieuwerts, A.M., Look, M.P., Yang, F., Talantov, D., Timmermans, M., Meijer-van Gelder, M.E., Yu, J. et al. (2005). Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet 365, 671–679. West, M., Ginsburg, G., Huang, A. and Nevins, J. (2006). Embracing the complexity of genomic data for personalized medicine. Genome Res 16, 559–566. Williams, R.S., Willard, H.F. and Snyderman, R. (2003). Personalized health planning. Science 300, 549.
Recommended Resources
■
385
RECOMMENDED RESOURCES Adjuvant! [http://www.adjuvantonline.com] Human Genome Project Information [http://www.ornl.gov/sci/ techresources/Human_Genome/medicine/pharma.shtml] International HapMap Project [http://www.hapmap.org] Medical Device Link [http://www.devicelink.com/ivdt/archive/ 03/04/002.html] Online Mendelian Inheritance in Man [http://www.ncbi.nlm.nih.gov/ entrez/query.fcgi?dbOMIM]
Online Mendelian Inheritance in Man #114480 Breast cancer [http:// www.ncbi.nlm.nih.gov/entrez/dispomim.cgi?id114480] Roche Diagnostics – press release [http://www.us.diagnostics.roche. com/press_room/2003/062503.htm] Surveillance epidemiology and end results [http://www.seer.cancer.gov]
This page intentionally left blank
Section
Policy Challenges in Genomic and Personalized Medicine
33. 34. 35. 36. 37.
5
From Sequence to Genomic Medicine: Genome Policy Considerations Educational Strategies in Genomic Medicine Federal Regulation of Genomic Medicine Economic Issues and Genomic Medicine Public–Private Interactions in Genomic Medicine: Research and Development
CHAPTER
33 From Sequence to Genomic Medicine: Genome Policy Considerations Susanne B. Haga
INTRODUCTION The success of the Human Genome Project will be measured largely with respect to advancements in biology and the realization of a new era of medicine based on the use of genomic data to predict, prevent, diagnose, and treat disease. However, the scientific challenge of sequencing the human genome will likely pale in comparison to the efforts required to translate genome sciences into personalized medicine. As we begin to apply the data and technologies arising from the Human Genome Project and subsequent projects including the HapMap Project (International HapMap Consortium, 2005) and the ENCODE Project (The ENCODE Project Consortium, 2007), the initial discovery, validation, and the development of a clinical test or therapy will be influenced by a range of science, health, and public policies. Although many of the policy issues are not unique to those faced by other new medical innovations, the genome revolution raises issues that span both the traditional science and health policy arenas (see Figure 33.1), some warranting resolutions specific to genomic medicine. This chapter will provide an overview of the major policy issues pertaining to the research, development, and translation of genomic medicine applications, including: research allocation and prioritization; the use and analyses of race in genome Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 388
Research issues
Acceptance and implementation issues
Economic issues
Legal issues
Education issues
Figure 33.1 The spectrum and inter-connectedness of genome policy issues, based on (Haga and Willard, 2006).
studies; the ethical issues linked to large-scale genome efforts such as biobanks, oversight, coverage, and reimbursement of new genomics applications; privacy and discrimination; and enhancing public and professional awareness. An in-depth analysis of several of the issues introduced here can be found in other chapters in this volume (see Chapters 34–37). Copyright © 2009, Elsevier Inc. All rights reserved.
Genome Research After the Human Genome Project
GENOME RESEARCH AFTER THE HUMAN GENOME PROJECT The completion of the Human Genome Project marked the beginning of a new era, rather than the end of an era (Collins, 2003). Although a substantial investment in basic research, it is unlikely that anyone could have foreseen the widespread impact of not only the ultimate product – a reference sequence of the human genome – but also the parallel technology development that enabled the early completion of the sequence of the human genome and the genomes of countless other species. Its impact on the biotechnology sector and other fields as diverse as evolutionary biology, agriculture, biodefense, and medicine is still in the early stages but its scope has been staggering already. Research Allocation and Prioritization In 2003, after the completion of the Human Genome Project, a new vision for genomics research was created (Collins et al., 2003). The next 5 years of genomic research would concentrate on three major areas – society, health, and biology. Across these three areas, six crosscutting research tracts were identified, including technology development and ethical, legal, and social issues. This differs markedly from past strategic plans that focused predominantly on mapping and sequencing of the human genome and model organisms (see Table 33.1; Collins and Galas, 1993; Collins et al., 1998, 2003). Despite the huge investment in basic biomedical research and the excitement generated by the Human Genome Project, concerns have been raised that the investment has not resulted in the anticipated development of new products, as the number of submissions and approval of new drugs and diagnostics has been on the decline (FDA, 2004). Because of this and political pressures to demonstrate return on investment from the doubling of the US NIH budget, the emphasis over the past several years has shifted toward translational and clinical science (Zerhouni, 2005). Recognition of the importance of translational research has led to changes in science policy such as the development of the NIH Roadmap (Zerhouni, 2003) and emphasis on the development of new tools and applications (FDA, 2006a). Translational genomics research will be an important part in each of these new efforts (see Chapter 22). Over the past decade, Europe has also increased its supports for both basic and translational genomics research. In 2001, the European Commission allocated €100 million for an initiative called Genome Research for Human Health. Starting in 2002, the European Community Framework Program for Research, Technological Development and Demonstration (otherwise known as FP6) identified genomics as one of its seven major thematic areas (European Commission, 2002). The research budget for life sciences, genomics, and biotechnology was more than €2 billion. The program aimed to promote European genomics research through three major strategies: establishing a robust infrastructure; fostering interdisciplinary research with other fields such as physics, chemistry, mathematics, and
■
389
T A B L E 3 3 . 1 Strategic goals for genomics research (1993–2008) in the United States (Collins and Galas, 1993; Collins et al., 1998, 2003) 5-Year period
Major goals
1993–1998
• Genetic map • Physical map • DNA sequencing technologies • Gene identification • Technology development • Model organisms • Informatics • Ethical, legal, social implications • Training • Technology transfer • Outreach
1998–2003
• Human DNA sequence • Sequencing technology • Human genome sequence variation • Technology for functional genomics • Comparative genomics • Ethical, legal, social implications • Bioinformatics and computational biology • Training
2003–2008
• Three thematic areas: Genomics to Biology, Genomics to Health, Genomics to Society • Six cross-cutting elements: Resources, technology development, computational biology, training, ethical, legal and social implications, and education
computer science; and integrating genomics into medicine and biotechnology. In 2007, FP7 was revised with the primary goal of supporting Europe’s global competitiveness. Genomics and biotechnology continue to have a prominent role in the development of new applications in medicine and food and agriculture. For the first time, investigator-driven projects will be funded in addition to the continued support of network initiatives (European Commission, 2002). The field of genome sciences has already taken the next step forward toward applying this new knowledge and capability to medically relevant initiatives. For example, the NIH Chemical Genomics project involves the development of a public repository of small organic molecules to study cellular pathways in health and disease and hasten the identification of new drug targets and drug development (NIH Chemical Genomics Center, 2005). The Medical Sequencing Project involves the sequencing of well-phenotyped patients with autosomal and X-linked Mendelian diseases as well as common diseases to identify the underlying genetic cause of disease (National Human Genome Research Institute, 2007).
390
CHAPTER 33
■
From Sequence to Genomic Medicine: Genome Policy Considerations
With respect to diagnostic screening and testing, the emphasis on translational research has provided opportunities to gather data about the clinical utility of new tools. Demonstrating clinical validity and utility is critical to the development of clinical guidelines and reimbursement policies that will support the appropriate use of these tests. For example, the National Cancer Institute is supporting a long-term prospective study called The Trial Assigning IndividuaLized Options for Treatment (Rx), or TAILORx (National Cancer Institute, 2006). This study will explore whether an expression profile of 16 genes associated with risk of recurrence for women with early stage breast cancer is useful in identifying those women who will most likely benefit from chemotherapy in addition to radiation and hormonal therapy. The study will follow 10,000 women for a period of 10 years, with follow-up of up to 20 years after initial therapies. While clinical trials represent the gold standard of evidence, these types of studies will not be feasible for every new genome application given the cost and time required, and other study approaches will need to be utilized. The continued support across translational genome research will be critical to the development of safe and effective applications in genomic medicine. Race and Genomics Population genetics research has spurred debate regarding the validity and use of racial categories as a variable in biological research. In particular, genomics research has provided insight into the genetic diversity (or lack thereof) of the human population, giving rise to questions about the use of race as a variable in biomedical research and whether genomics could provide improved measurements of variation in place of the concept of race (Burchard et al., 2003; Cooper et al., 2003). While a major conclusion of the Human Genome Project was the high degree of sequence identity between individuals, subsequent research has focused on the small differences between individuals and populations, particularly with respect to disease (Foster, 2005). For instance, the 3-year US-led HapMap project generated a map of one million common genetic variants through analyses of more than 200 genomes of individuals from four different populations (International HapMap Consortium, 2005). Similar to the HapMap project, the Mexican Genome Project, started in 2005, sought to determine the genetic variation of the heterogeneous mestizo population in Mexico (Mothelet and Herrera, 2005). Mestizo refers to individuals with mixed ancestry of European (primarily Spaniards) and Indian. The goal of the project was to determine whether any genetic differences between populations may be correlated to known health differences. Preliminary analysis of data from the project concluded that mestizo ancestry is a mixture of 35 ethnic groups (Wall, 2007). In general, 65% of Mexican ancestry can be traced to indigenous populations and 35% is due to non-indigenous groups (African, Asian, and European). Not surprisingly, regional differences were also found as numerous indigenous populations existed for some time prior to the arrival of the Europeans. The medical implications of these data remain to be determined. However, it is anticipated that the data will benefit other Latin
American countries as well and will aid in the development of safe and effective drugs for these populations which are not typically studied in clinical trials conducted by United States and European drug makers (Mothelet and Herrera, 2005). For genomic medicine, the main question is whether race can be used as a surrogate for biological variance to aid prevention, diagnosis and treatment of disease. Studies have detected not only differences in the prevalence of genetic variants (Hall, 1999; Monaghan et al., 1996), but also in the risks associated with the same genetic variant (Helgadottir et al., 2006), implying a role for additional genetic or environmental factors. While studies have shown that individuals from populations around the world can be clustered genetically into six groups (Rosenberg et al., 2002), other studies have demonstrated a range of mixed ancestry in African-Americans, Hispanics, and Mexicans (Parra et al., 1998; Shriver et al., 2003; Sinha et al., 2006). Our new understanding of human variation and ancestry has yet to impact federal policies regarding the use of race in biomedical research and clinical trials. Current policies such as the National Institutes of Health Revitalization Act (1993), the US Food and Drug Administration (FDA) (1998, 2005a), and the European Medicines Agency (1998, 2005) require that study participants be identified by race and/or ethnicity to ensure a diverse study population and to allow subset analysis. In light of recent findings, however, the application of race and ethnicity categories to biomedical research and the requirement to subset and analyze clinical trials data seem to be outdated and inadequate measures of differences in treatment outcome or response (Haga and Venter, 2003). Although self-identified race has been shown to correspond well with genetic clustering (Tang et al., 2005), it appears to be more problematic clinically since it has not always been shown to be a consistent and reliable measure (Hahn et al., 1996; Johnson, 1974; Rankin and Bhopal, 1999). Moreover, the context in which the self-identification is made may influence an individual’s decision, resulting in variable responses depending on who is asking the question and how the information might be used (Senior and Bhopal, 1994). In using race as an analytical variable, it is important to consider the two major constructs of race: the use of race as a social measure to detect differences of potentially important factors such as access to healthcare, environmental exposures, and lifestyle; or the use of race as a biological variable to identify genetic (and therefore, biological) differences between populations underlying differences in disease prevalence, severity, and outcome. While there is little disagreement that genetic epidemiology studies would benefit from more diverse sample populations (Ioannidis et al., 2004; Tang, 2006), which may be achieved through using race as a surrogate, it is still unclear what the significance (social and/or biological) of a positive association is between a given phenotype and race. Thus, difficulty in defining the impact of different factors makes race a poor surrogate in general, and direct measurement of relevant factors (e.g., diet, environment) may reduce some of the challenges arising from the use of such a broad undefined variable such as race. Genetics and genomics present an alternative measure of human variation for which race
Policy Issues in Large-Scale Genetics and Genomics Research
has been used as a surrogate that will lead to more definitive and quantifiable data points (Shields et al., 2005). Another major issue is the relationship between genetic and genomic data and health disparities. Differences in disease prevalence and outcome between groups, typically characterized by race, have been well-documented (Institute of Medicine, 2003). If individual differences in disease predisposition, prognosis or response to drugs may be accounted for by genetic variation, could group health differences also be accounted for by genetic variation between populations? Despite the small role that genetics is believed to play regarding disease prevalence and outcomes of common diseases, there is an inherent danger of extrapolating differences detected in individuals from one group to an entire race, potentially resulting in unvalidated support of race as a biological category (Wiegmann 2006). Indeed, the temptation to attribute health disparities to genetic causes may distract from other types of social and biomedical research that may lead to greater reduction in these observed differences (Sankar et al., 2004). Furthermore, race-based medicines such as BiDil, a drug approved by the FDA for the treatment of heart failure in self-identified black patients, have the added potential of reifying the biological concept of race and hinder efforts to reduce disparities (Haga and Ginsburg, 2006; Kahn, 2005).
POLICY ISSUES IN LARGE-SCALE GENETICS AND GENOMICS RESEARCH The shift to study the genetic etiology of common, complex diseases and the development of high-throughput technologies such as microarray analysis and new sequencing techniques at increasingly cheaper costs have enabled the expansion of study populations in genomic studies. Whereas genetic studies traditionally focused on small groups, particularly families, genomics studies are characterized by the collection of genetic data on hundreds or thousands of individuals, healthy and affected. The demand for DNA samples from individuals of various phenotypes has led to the creation of local and national biobanks or biorepositories worldwide. Given the enormity of national biobanks, several policy issues arise, particularly with respect to research policy. Among the many issues that have been debated about biobanks are the scientific need and merit of a national biobank, the cost and feasibility of successfully establishing and operating a biobank, the required infrastructure, accessibility to samples and data, informed consent, intellectual property (IP), privacy and confidentiality of data, and disclosure of research results to participants. While current policies may be applicable to some of these issues as they are not unique to biobanks, new policies may be required to address some of these issues such as data-sharing and data disclosure and to help reassure the public. Biobanking Several years prior to the completion of the Human Genome Project, a number of countries contemplated setting up a
■
391
national research resource or “biobank” that would contain both clinical specimens and data for the purposes of large-scale genetic epidemiology studies. Samples and data would be collected from 60,000 to one million individuals and stored for five to 30 years (Godard et al., 2004). Given that a national biobank depends on the public’s support and participation, engaging the public is critical to its success (Haga and Willard, 2006; Haga and Beskow, 2008).Two approaches have been used to communicate with the public about national biobanks – a communication or partnership approach (Godard et al., 2004). In Iceland, a brief public consultation took place in the form of radio and television programs, town hall meetings, and public surveys. In 1998, the Act on a Health Sector Database was passed, authorizing the Ministry of Health and Social Security in Iceland to license a private firm (deCODE Genetics, Inc.) to establish and maintain the Health Sector Database, which would store the health records of Iceland’s 270,000 citizens. Strong criticism from researchers and ethicists worldwide quickly followed, particularly with respect to the issue of “presumed consent.” In response, an option to decline was made available and about 10% of the Icelandic population chose to opt out. In contrast, the United Kingdom embarked on a massive public 3-year consultation campaign about its proposed national databank. The national diallog included town hall meetings, focus groups, interactive workshops, and published reports. The exchange of information helped inform policy-making to ensure that public concerns were addressed while at the same time raised public awareness about the project. In addition, public comments were requested on procedural and governance documents. Other countries such as Japan, Canada, Estonia, United States, Tonga, and Latvia are considering or have already established national biobanks. Their approach to public consultation has fallen in between Iceland’s communication approach and the UK’s partnership approach. The different approaches utilized are reflective of different cultures, attitudes towards scientific research, history, and government and healthcare systems. Data Disclosure Research results can be disclosed in a variety of formats including individual or aggregate reports. Aggregate reports would be the only way to inform participants about study outcomes if the samples are collected anonymously or pooled. However, aggregate reports may dilute the value of results to individuals because participants will be left with uncertainty regarding the significance of the findings for themselves specifically (e.g., whether they were found to be at high risk versus low risk) (Shalowitz and Miller, 2005). A number of studies have examined the attitudes and responses of research participants with respect to access to clinical research results, though not specifically to results from genetic studies. In general, studies have found that disclosure of research results to participants, even those of a serious nature, do not result in negative psychological impacts in the majority of research participants (Bunin et al., 1996; Schulz et al., 2003; Snowdon et al., 1998; Shalowitz and Miller, 2008). Participants indicated they
392
CHAPTER 33
■
From Sequence to Genomic Medicine: Genome Policy Considerations
would want access to research results (Wendler and Pentz, 2007), even if upsetting (Schulz et al., 2003) or if not considered clinically valid or useful (Wendler and Emanuel, 2002). Despite current evidence that research participants would like the option of requesting research results even if of little clinical utility, the practice does appear to be common. An analysis of consent forms in the Children’s Oncology Group found that only two of 202 studies offered the option to participants to receive research results and only 10 of 202 studies offered to provide new information after the study was completed (Fernandez et al., 2003a). Although almost 80% of clinical investigators surveyed agreed that clinical trial results should be offered to research participants, about 60% of investigators only offered results less than 20% of the time (Partridge et al., 2004). Moreover, more than 60% did not believe that the disclosure of research results would have a negative impact on participants (Partridge et al., 2004). The major reasons given for not returning results include potential harm of research participants and time constraints and cost related to development of research summaries or individual reports and re-contacting of participants (Fernandez et al., 2003b; Rigby and Fernandez, 2005). While several policies on data disclosure have been developed, few exist at the federal level. Of the guidelines specific to genetic studies, the importance of clinical relevance and utility is emphasized. For example, the NHLBI guidelines identified the availability of a clinical intervention (treatment or preventative) as a key criterion if results were to be returned to study participants (Bookman et al., 2006). It has been suggested that investigators have an ethical responsibility to offer participants access to research results (MacNeil and Fernandez, 2006). Providing an option to disclose research data shows respect to the research participant (Shalowitz and Miller, 2005). Those investigators who choose not to provide access to research data should have to justify this decision (Shalowitz and Miller, 2005). Given concerns about the potential harms of unvalidated or inconclusive data, it has been suggested that the option for access to research results should be provided but not necessarily encouraged and that the uncertain nature of the results may be emphasized (Shalowitz and Miller, 2005). Because of these potential risks, it has also been suggested that a second consent should be obtained prior to disclosure of research results (Fernandez et al., 2003c, d). Data-sharing Publicly accessible genomic databases employ a range of data access policies. Initially, the leaders of the Human Genome Project determined that all data generated from the project should be placed in a public repository and accessible to all interested users (The Wellcome Trust, 1996, 2003). Rapid deposit of sequence data was required from each of the sequencing centers. This policy decision was influenced by concerns that access to basic data would be impeded if traditional methods of scientific data disclosure were followed. For new publicly funded, large-scale international collaborations, similar policies may need to be developed in advance to address issues of datasharing, intellectual property and use of specimens (if applicable)
(Chokshi et al., 1999). The 2006 NIH proposal specifying policies on data-sharing for NIH-supported or conducted genomewide association studies is a step in this direction (National Institutes of Health, 2006a). Several studies have examined data-sharing and datawithholding practices of US biomedical scientists and, in particular, geneticists. According to the most recent study, geneticists were more likely to engage in data-withholding practices than other life scientists (Blumenthal et al., 2006). Among academic geneticists, at least one-third of requests for additional information, data, or materials and 10% of post-publication requests were denied (Campbell et al., 2002). Twenty-eight percent stated that data-withholding impacted their ability to confirm research findings (Campbell et al., 2002). Reasons for withholding included concerns about additional time and effort, protection of junior investigators, and protection of data intended for future publications (Campbell et al., 2002). Delay in publication may be caused by factors such as IP or desire to maintain a competitive advantage (Blumenthal et al., 1997). These three surveys combined have assessed data-sharing practices from the 1980’s until today and are among the few sources of information about these scientific practices in genetics or, by extension, genomics. Little research has been conducted to assess public attitudes regarding sharing of biomedical research data and specifically genomic research data. The privacy of genomic information, even if de-identified or stored anonymously, is tenuous, since as few as 30 single nucleotide polymorphisms (SNP) may uniquely identify an individual (Lin et al., 2004). Therefore, participants may need to be informed about the potential risks due to datasharing practices.
INTEGRATING GENOMIC MEDICINE APPLICATIONS IN HEALTHCARE While human genetics has focused traditionally on testing for Mendelian and rare diseases through established techniques such as karyotyping, fluorescence in situ hybridization, and restriction fragment length polymorphism analysis, genomic medicine builds on new technologies such as genome-wide analysis of SNP and copy number variation, metabolomics and whole genome sequencing. These new technologies have led to testing for a broader range of diseases including common, complex conditions such as Type II diabetes. However, the expanded scope of clinical genomic applications necessitates consideration of the adequacy of current policies to ensure the safe, effective, and appropriate use of the genomic applications. In particular, changes may be need to be made to the regulatory system, education of health professionals involved in the provision of these new applications, and patient protections. Oversight In the United States, it is difficult to determine the actual number of traditional genetic tests performed annually given the
Integrating Genomic Medicine Applications in Healthcare
privatized health care system; the only published survey data estimated that over 175,000 tests were performed in 1996 (McGovern et al., 1999). In contrast, the Clinical Molecular Genetics Society conducts annual audits of the UK National Health Service laboratories. In 2005–2006, 78,600 postnatal genetic tests were performed, almost 20% more than in 2004–2005 (Clinical Molecular Genetics Society, 2004–2005, 2005–2006). More than 1500 prenatal genetic tests were also performed, representing a 12% increase from the previous year (Clinical Molecular Genetics Society, 2004–2005, 2005–2006). Of particular interest, the number of predictive and confirmatory tests almost doubled over the prior year. About 85 conditions were tested for overall, although the majority of tests were for familial cancers such as breast cancer (BRCA1/2) and colorectal cancer (HNPCC and FAP). In the European Union, it is estimated that more than 700,000 tests are performed annually (Ibarreta et al., 2003). In the United States, the Clinical Laboratory Improvement Amendments (CLIA) program administered by the Centers for Medicare and Medicaid oversees laboratory operations including quality control, quality assurance and analytical test validity. Laboratories are subject to inspection every two years by a state CLIA inspector or an accredited organization such as the College of American Pathologists. In 2006, 198,000 laboratories were registered by the CLIA program. Tests that are packaged and marketed as kits are considered medical devices and require clearance or approval by the FDA. However, as the majority of genetic tests are offered as a clinical laboratory service, otherwise known as a “home-brew” test, they are not currently subject to FDA regulations. In addition to analytical validity, FDA also reviews the clinical validity of tests. Over the past decade, two federal committees have explored the adequacy of current regulatory oversight mechanisms of genetic tests. In 1997, a NIH-Department of Energy Task Force indicated that an assessment of current review processes of the validity and utility of genetic tests may be warranted, but stopped short of providing specific suggestions (Holtzman and Watson, 1997). In 2000, the Secretary’s Advisory Committee on Genetic Testing recommended that the FDA regulate all genetic tests and that oversight of genetic testing laboratories under the CLIA be strengthened (Secretary’s Advisory Committee on Genetic Testing, 2000). To date, these recommendations have not been implemented, although the creation of a genetic testing specialty under CLIA has been proposed and debated (Centers for Disease Control and Prevention, 2000). Several guidance documents have been issued by FDA, resulting in a gradual increase in oversight of new genomics applications. In 2005, a guidance document was issued encouraging sponsors to voluntarily submit pharmacogenomics data to help prepare reviewers on the type and complexity of genomics-based data (FDA, 2005b). Two relevant draft guidances were also released in 2005 – a draft guidance on nucleic acid-based in vitro diagnostic devices for detection of microbial pathogens was released (FDA, 2005f) and another on the co-development process of drug-device
■
393
combination products as evident from their concept paper (FDA, 2005e). In addition, other guidance documents have been developed focusing on drug metabolizing enzymes genotyping system (FDA, 2005c), multiplex technology (FDA, 2005d), analyte specific reagents (FDA, 2006b), algorithm-based genetic tests (FDA, 2006c), genetic toxicology studies (FDA, 2006d), and pharmacogenetic tests (FDA, 2007). Direct-to-Consumer Marketing Genetic services and applications are often promoted directly to consumers, similar to drugs and other healthcare services. However, for many of these services and products, the involvement of a health care professional is not required, and consumers can purchase goods directly from the laboratory or manufacturer (an “over-the-counter” test). Laboratory services include testing for paternity, lifestyle, health, and ancestry. Products include DNA-tailored cosmetics and nutritional supplements. Consumers who purchase these services may decline to share the results with their health practitioner due to fear of genetic discrimination. Direct-to-consumer (DTC) provision of genetic services and products, particularly those that are not health-related, represent an emergent industry. However, at this early stage of research and development, the complex genetic etiology and role of environmental factors for many phenotypes and traits are not clearly understood. Therefore, many of the early tests promoted directly to consumers likely lack robust validity and utility. As these tests are not subject to FDA oversight, the evaluation of the validity of genetic tests and products may be challenging for some consumers. Consumers may be vulnerable to ambiguous and exaggerated claims. In addition, the complex nature of genetic information related to health may warrant consultation with professionals trained in genetics or other specialist to ensure appropriate test interpretation and intervention. General advertising of genetic tests accessible only through a physician has also increased. For example, Myriad Genetics launched an advertising campaign in two cities in 2002 for the BRCA1/2 breast cancer tests. A study conducted the following year to assess the impact of the campaign on women’s awareness of the tests found that consumers were substantially more likely to have heard of the test and seen an advertisement (Jacobellis et al., 2004). However, despite the increased awareness, the perceived knowledge and interest in testing did not differ between the pilot and control cities. Physician knowledge of breast cancer testing did not differ between the pilot and control cities, although physicians reported ordering a significantly higher number of tests in the pilot cities (Myers et al., 2006). DTC genetic services and products have raised concerns among a number of groups. In 2002, the UK Health and Science Ministers requested an investigation by the Human Genetics Commission into DTC genetic testing. The commission identified two potential harms from DTC genetic testing: the impact of inaccurate or misinterpreted test results on consumer health and the absence of an adequate informed consent process (UK Human Genetics Commission, 2003). As a result, the
394
CHAPTER 33
■
From Sequence to Genomic Medicine: Genome Policy Considerations
Commission recommended that stricter controls be put in place for DTC genetic tests and that predictive genetic tests should be restricted from DTC sale. In the United States, the American College of Medical Genetics (2004 and 2008) recommends “a knowledgeable healthcare professional should be involved in the process of ordering and interpreting a genetic test” to prevent potential harms such as test misinterpretation and inappropriate follow-up. In 2006, an investigation conducted by the General Accounting Office concluded that DTC nutrigenetic tests were medically unproven and ambiguous (Government Accountability Office, 2006). To educate consumers about home genetic test kits, the Federal Trade Commission, the Centers for Disease Control and Prevention, and the Food and Drug Administration developed a fact sheet (Federal Trade Commission, 2006). And, in 2007, the American Medical Association’s House of Delegates passed Resolution 522, which requested further study of the practice of DTC advertising of genetic tests and, in particular, at the existing oversight of this field (American Medical Association, 2007).
Access/Reimbursement Coverage of genomic applications is intimately linked to uptake of a new test and integration into clinical practice. The demonstration of the validity and utility of a genomic application as well as cost-effectiveness is essential to securing reimbursement. More than 150 economic analyses of genetic services have been conducted, many for adult conditions such as cancer (Carlson et al., 2005; Phillips and Van Bebber, 2004). While many tests have been shown to be cost-effective, others still have questionable utility despite demonstrated clinical validity. For example, a commercially available molecular profiling test known as Allomap has been shown to be cost-effective to identify heart transplant patients at risk for rejection compared to endomyocardial biopsy to detect rejection, mainly due to the costsavings gained from not having to perform a biopsy (Evans et al., 2005). But several tests developed to guide therapy decisions are considered by some to be of questionable value (Matchar et al., 2006). The increased involvement of FDA review of genomic tests and more focused translational research has encouraged the coverage of these applications. For example, in 2005, the Blue Cross Blue Shield Association’s Technology Evaluation Center (TEC) assessed four gene expression profiles for use in the management of breast cancer. Based on the available evidence at the time, the report concluded that none of the profiles met the TEC criteria (Blue Cross Blue Shield Association Technology Evaluation Center, 2005). The TEC criteria include approval by federal regulatory bodies and evidence demonstrating an improved outcome and comparable benefits to existing alternative treatments or applications. However, in 2007, a second review of the Oncotype DX test (one of the four tests evaluated in 2005) concluded that the test now met the TEC criteria and was considered to be useful specifically regarding adjuvant
chemotherapy for women with estrogen receptor-positive, node-negative, tamoxifen-treated breast cancer (Blue Cross Blue Shield Association Technology Evaluation Center, 2007). In addition to evaluating both clinical validity and utility, pharmacogenetic testing policies must also consider coverage of the drugs identified as being most likely to perform well or least likely to cause adverse side effects. For tests that are co-developed with drugs, the cost of the testing and drug could be determined by a single policy decision. But in the event that a test indicates a different drug, payors may be forced to revise their coverage policies for drugs that are not listed on their formularies. Health Professional Education As the applications of genetic and genomic information have expanded beyond the traditional specialties of medical genetics, pediatrics, and obstetrics into other fields such as oncology, cardiology, neurology, and psychiatry, the education of health professionals needs to evolve as well. While a number of surveys have documented the variable level of physician knowledge of genetics (Metcalfe et al., 2002), none has assessed knowledge of the newer field of genomics. The uptake and use of new applications will be stalled until health professionals gain some understanding about the appropriate use of these tests and the interpretation and application of test results. As a result, the continuum of health education from graduate or professional school to continuing education should aim to increase awareness and understanding of these new tools and therapies across virtually all medical specialties (Challen et al., 2005; Gurwitz et al., 2003). In addition, it will be important for health professionals to understand other issues related to the use of these technologies such as implications for family members and privacy and confidentiality issues. Re-defining Roles of Health Professionals Due to the expected widespread use of genomic applications, some re-organization of health professional roles may be required, given the limited number of genetics specialists and new applications such as pharmacogenetic testing (see Table 33.2) (Guttmacher et al., 2001). For example, the pharmacist may play a much greater role in the integration of pharmacogenetic testing by helping to ensure that the drug and dosage are safe based on the patient’s genotype (Clemerson et al., 2006). Monitoring pharmacogenetic information to assure appropriate drug dosing would appear to be a natural extension of the role of pharmacists. To fulfill this expanded role, pharmacists would need greater access to patient medical records (Clemerson et al., 2006), although physicians and patients have expressed concerns regarding patient confidentiality if full access were permitted (Porteous et al., 2003). In contrast, a board-certified genetics professional (physician or genetic counselor) may have a revised, lesser or even non-existent role in the provision of genomic testing such as
Integrating Genomic Medicine Applications in Healthcare
T A B L E 3 3 . 2 Changing roles of health professionals in genomic medicine Health professional
New or revised role
Primary care Physician
Increased focus on family-history assessment, identification of appropriate patients for testing, discussion of risks and benefits of testing, interpretation of test results, revision of medical management based on test results, focus on preventative strategies
Nurse
Enhanced family-history taking and counseling skills; enhanced education in corresponding areas of genetics (e.g., cancer genetics for oncology nurses); awareness of ethical, social, and legal issues associated with testing
Genetic or Genomic Specialist (Genetic Counselor or Physician)
Consultation reserved for complex genetic cases; interpretation of wholegenome data, including sequence; increased role in professional education development
Pharmacist
Increased understanding of genetic factors in drug safety and efficacy; enhanced education of genetic etiology of drug response; consideration of pharmacogenetic test results prior to drug treatment; increased collaboration with prescribing physician; increased patient education with respect to pharmacogenetic testing; added consideration of pharmacogenetic test result to determine appropriate dose adjustment or drug selection
a pharmacogenetic profile as these tests become routine. The need for a genetics specialist may be determined by several factors including complexity in determining appropriate candidates for testing and interpretation of test results, the extent of the informed consent process, reimbursement policies, familial implications and the risks and benefits of the test. However, given the relatively small size of the genetics workforce, any new requirements for their services may not be feasible. Consultation by a board-certified genetics specialist may need to be reserved for only the most complicated cases. To increase the number of physicians with training in genetics or genomics, one possibility may be to increase the number of dual board-certified specialties with medical genetics. For example, an individual dual-certified in medical genetics and pharmacology would have expertise in genetics and pharmacology and
■
395
be able to advise the treating physician about the best choice of treatment based on the patient’s pharmacogenetic profile and other clinical information. However, the training requirements to fulfill the medical genetics certification should be substantially revised to provide direct knowledge relevant to the individual’s primary profession with greater focus on test appropriateness, interpretation and test utility. In addition, genetics certification should not lengthen the overall training program. Pilot development studies have found that new roles associated with genetic testing were mostly fulfilled by non-genetics specialists, in particular nurses (Bennett et al., 2007). For example, the cancer family nurse specialist is trained to identify patients with a family history of cancer that would place them at increased risk and indicate testing (Bennett et al., 2007). Indeed, genetic competencies have been developed and training programs have been accordingly revised to meet these new needs and to enable nurses to adequately provide genetics services and guidance (Jenkins and Calzone, 2007; Lewis et al., 2006). Studies have demonstrated the equivalence of care provided by nurses trained in genetics compared to board-certified genetic counselors (Torrance et al., 2006). Several groups have developed professional genetic competencies including the National Coalition for Health Professional Education in Genetics (2005), and the National Health Service’s National Genetics Education and Development Centre and the Skills for Health Service (2007) have developed general competences for non-genetics practitioners. Privacy and Confidentiality The issue of medical privacy is a major concern in the United States. In 2005, the National Consumer Health Privacy Survey reported that 66% of Americans were very or somewhat concerned about medical privacy (California Health Care Foundation, 2005). Due to concerns that employers or other groups may have access to a medical record, 11% of survey respondents indicated paying for a cancer test or procedure out-of-pocket. The predictive nature and often deterministic views associated with genetic information have likely contributed to the heightened concerns linked to testing. In addition, as the number of health professionals with access to a patient’s genomic test results increases, the ability to safeguard the information increases becomes more challenging. Genetic Discrimination Genetic discrimination involves the inappropriate use of genetic information, primarily with respect to health insurance and employment, but including other areas such as life insurance, long-term care insurance, adoption services, egg and sperm banks or educational admissions programs. While the fear of genetic discrimination has existed prior to the start of the Human Genome Project, the solutions have been slow to come by and, at best, provide a patchwork of protections in the United States and around the world. For example, about two-thirds of states have enacted legislation to prohibit or limit the use of
396
CHAPTER 33
■
From Sequence to Genomic Medicine: Genome Policy Considerations
genetic information for risk selection and risk classification for health insurance purposes. Passage of national legislation would provide universal coverage against health insurance and employment discrimination on the basis of genetic information for all. But despite all of the interest and support shown by federal lawmakers including the past two US presidents, national legislation has only just been enacted. Unbelievably, it has taken longer to pass genetic anti-discrimination legislation than it actually took to sequence the human genome. Since the first bill on genetic discrimination in health insurance was introduced in the United States in 1995, at least 20 other bills have been introduced, but only one has passed providing partial protection against genetic discrimination for the group insurance market. In 2007, the Genetic Information Nondiscrimination Act was re-introduced to “prohibit discrimination on the basis of genetic information with respect to health insurance and employment.” In contrast to previous years, it swiftly moved through the three House committees who have jurisdiction and passed by a floor vote of 420 to 3, thus moving further along than in any previous Congress. The bill was also passed by the Senate and signed into law by President Bush in May 2008. Despite universal condemnation against genetic discrimination (UNESCO, 1997, 2003), the development and implementation of national protections has also been slow outside the United States. For example, although “genetic features” is included in the European Union’s Charter of Fundamental Rights’ section on non-discrimination, countries have had variable success in providing national protections (European Parliament, 2000). Austria, Belgium, France, the Netherlands, Luxembourg, Greece, and Italy have enacted legislation prohibiting access of genetic information without consent. Various initiatives have been undertaken in the United Kingdom, Australia, and Canada to protect against genetic discrimination by life insurers and employers. Similar to the United States, a patchwork of provincial legislative protections in Canada exist. A recent review of the Canadian Human Right Act by the Department of Justice recommended that the definition of disability include genetic predispositions (Department of Justice, Canada, 2005). Australia has been particularly active in investigating the protections of genetic information. In 2000, the Attorney-General of Australia and the Minister for Health and Aged Care called for an inquiry into genetic discrimination. An extensive public consultation was undertaken including widespread dissemination of two consultation papers (Australian Law Reform Commission, 2001, 2002), 15 public forums around the country, and 185 meetings with key stakeholders and interested parties. In 2003, a final report was released containing 144 recommendations directed at local, regional, and national governments as well as statutory authorities (Australian Law Reform Commission, 2003). And in the United Kingdom, a moratorium is in effect banning the use of genetic testing information for insurance underwriting purposes until 2014 (Association of British Insurers, 2008).
Intellectual Property Intellectual property (IP) of genes and genetic material has been a major point of controversy since the early years of the Human Genome Project. Nearly 20% of the human genome is under US patents, comprising about 4400 genes (Jensen and Murray, 2005). More than 60% of these patents are owned by private entities (Jensen and Murray, 2005). In addition, almost a dozen complete genomes have been patented or are pending patent approval (O’Malley et al., 2005). A constellation of issues pertaining to ownership has garnered significant attention and concern from various stakeholders. From the single laboratory to national advisory bodies to international organizations, gene patenting has become a controversial topic, particularly with respect to the limitation of access to gene-related inventions and benefit-sharing (World Health Organization, 2005; Canadian Biotechnology Advisory Committee, 2006). The patenting of genomes has significantly broadened the scope of IP, and its impact remains to be seen. Although there has been a significant amount of debate about the patenting of genes, concern also exists about the licensing practices of patent holders, as exemplified by BRCA1/BRCA2 and Canavan’s disease. To address some of these concerns, the NIH has also taken steps to promote open access to its large-genomic datasets and encourage fair licensing practices through IP user agreements and guidance documents (National Institutes of Health, 2005). In addition, an amendment to a Request for Applications for genome sequencing centers requires applicants to develop an “IP Management Plan” to ensure wide availability of sequencing data generated by awardees (National Institutes of Health, 2006b).
CONCLUSION While advances stemming from the genome sciences are certain to significantly impact the practice of medicine and our understanding of human development, biology, biochemistry, and physiology, the safe and appropriate use of this new information and its associated technologies will warrant new policies and an improved level of understanding by patients, consumers, and providers. New policies should not obstruct or prevent scientific advancements from moving forward, but rather should aim to facilitate the expansion of the field in a manner that addresses concerns about the direction, pace, and application of the science. In addition, while not explicitly discussed, the ethical aspects of genome research and medical applications will substantially guide and influence policy-making related to the development and use of these new tools. Given the wide range of stakeholders involved in the translation of genomic medicine applications from bench to bedside, it will be critical to gather their perspectives as new policies are being considered and developed.
References
■
397
REFERENCES American College of Medical Genetics Board of Directors (2008). ACMG Statement on Direct-to-Consumer Genetic Testing. Available at http://www.acmg.net/StaticContent/StaticPages/ DTC_Statement.pdf. American College of Medical Genetics Board of Directors (2008). ACMG Statement on Direct-to-Consumer Genetic Testing. Available at http://www.acmg.net/AM/Template.cfm?Section=Policy_ Statements&Template=/CM/ContentDisplay.cfm&ContentID= 2975. American Medical Association (2007). Resolution 522: Direct to Consumer Advertising and Provision of Genetic Testing. Available at http://www.ama-assn.org/ama1/pub/upload/mm/467/522.doc. Association of British Insurers (2008). Insurance genetics moratorium extended until 2014. Available at http://www.abi.org.uk/ Newsreleases/ViewNewsRelease.asp?nrid=16274. Australian Law Reform Commission and Australian Health Ethics Committee of the National Health and Medical Research Council (2001). Protection of Human Genetic Information (IP 26). Available at http://www.austlii.edu.au/au/other/alrc/publications/issues/26/ index.html. Australian Law Reform Commission and Australian Health Ethics Committee of the National Health and Medical Research Council (2002). Protection of Human Genetic Information (Discussion paper 66). Available at http://www.austlii.edu.au/au/other/alrc/ publications/dp/66/. Australian Law Reform Commission and Australian Health Ethics Committee of the National Health and Medical Research Council (2003). Essentially Yours: The Protection of Human Genetic Information in Australia (ALRC 96). Available at http://www.austlii.edu. au/au/other/alrc/publications/reports/96/. Bennett, C., Burton, H. and Fardon, P. (2007). Competences, education and support for new roles in cancer genetics services: outcomes from the cancer genetics pilot projects. Fam Cancer 6, 171–180. Blue Cross Blue Shield Association. Technology Evaluation Center. Gene Expression Profiling for Managing Breast Cancer Treatment. Assessment Program 2005, 20 (3). Available at http://www.bcbs. com/tec/Vol20/20_03.pdf. Blue Cross Blue Shield Association’s Technology Evaluation Center (2007). Actions Taken by the Medical Advisory Panel. Available at http://www.bcbs.com/betterknowledge/tec/press/actions-takenby-the-medical-1.html. Blumenthal, D., Campbell, E.G., Anderson, M.S., Causino, N. and Louis, K.S. (1997).Withholding of research results in academic life science: Evidence from a national survey of faculty. JAMA 277, 1224–1228. Blumenthal, D., Campbell, E.G., Gokhale, M., Yucel, R., Clarridge, B., Hilgartner, S. and Holtzman, N.A. (2006). Data withholding in genetics and the other life sciences: Prevalences and predictors. Acad Med 81(2), 137–145. Bookman, E.B., Langehorne, A.A., Eckfeldt, J.H., Glass, K.C., Jarvik, G.P., Klag, M., Koski, G., Motulsky, A., Wilfond, B., Manolio, T.A. et al. (2006). Reporting genetic results in research studies: Summary and recommendations of an NHLBI working group. Am J Med Genet A 140A, 1033–1040. Bunin, G.R., Kazak, A.E. and Mitelman, O. (1996). Informing subjects of epidemiologic study results. Children’s Cancer Group. Pediatrics 97(4), 486–491.
Burchard, E.G., Ziv, E., Coyle, N., Gomez, S.L., Tang, H., Karter, A.J.,Mountain, J.L., Perez-Stable, E.J., Sheppard, D. and Risch, N. (2003). The importance of race and ethnic background in biomedical research and clinical practice. N Engl J Med 348, 1170. California Health Care Foundation and Forrester Research, Inc. (2005). National Consumer Health Privacy Survey. Available at http://www. chcf.org/documents/ihealth/ConsumerPrivacy2005Slides.pdf. Campbell, E.G., Clarridge, B.R., Gokhale, M., Birenbaum, L., Hilgartner, S., Holtzman, N.A. and Blumenthal, D. (2002). Data-withholding in academic genetics: Evidence from a national survey. JAMA 287, 473–481. Canadian Biotechnology Advisory Committee (2006). Human Genetic Materials, Intellectual Property and the Health Sector. Available at http:// strategis.ic.gc.ca/epic/internet/incbac-cccb.nsf/en/ah00578e.html. Carlson, J.J., Henrikson, N.B., Veenstra, D.L. and Ramsey, S.D. (2005). Economic analyses of human genetics services: A systematic review. Genet Med 7, 519–523. Centers for Disease Control and Prevention. Notice of Intent: Genetic testing under the Clinical Laboratory Improvement Amendments. 65 Fed Reg 25928, May 4, 2000. Challen, K., Harris, H.J., Julian-Reynier, C., Ten Kate, L.P., Kristoffersson, U., Nippert, I., Schmidtke, J., Benjamin, C. and Harris, R.GenEd Research Group (2005). Genetic education and non genetic health professionals: Educational providers and curricula in Europe. Genet Med 7(5), 302–310. Chokshi, D.A., Parker, M. and Kwiatkowski, D.P. (1999). Data sharing and intellectual property in a genomic epidemiology network: Policies for large-scale research collaboration. Bull World Health Organ 84(5), 382–387. Clemerson, J.P., Payne, K., Bissell, P. and Anderson, C. (2006). Pharmacogenetics, the next challenge for pharmacy?. Pharm World Sci 28(3), 126–130. Clinical Molecular Genetics Society. Final 2004–2005 Audit Report. Available at http://www.cmgs.org/new_cmgs/Laboratory%20 Information/FinalCMGSAudit0405.pdf. Clinical Molecular Genetics Society. Final 2005–2006 Audit Report. Available at http://www.cmgs.org/new_cmgs/Laboratory%20 Information/FinalCMGSAudit0506.pdf. Collins, F.S. (May 22, 2003). Testimony Before the Subcommittee on Health Committee on Energy and Commerce, United States House of Representatives. Available at http://www.genome. gov/11007447. Collins, F. and Galas, D. (1993). A new five-year plan for the U.S. Human Genome Project. Science 262, 43–46. Collins, F.S., Patrinos, A., Jordan, E., Chakravarti, A., Gesteland, R. and Walters, L. (1998). New goals for the U.S. Human Genome Project: 1998–2003. Science 282, 682–689. Collins, F.S., Green, E.D., Guttmacher, A.E. and Guyer, M.S. (2003). U.S. National Human Genome Research Institute. A vision for the future of genomics research. Nature 422, 835–847. Cooper, R.S., Kaufman, J.S. and Ward, R. (2003). Race and genomics. N Engl J Med 348, 1166. European Commission (2002). The Sixth Framework Programme in Brief. Available at http://ec.europa.eu/research/fp6/pdf/fp6-inbrief_en.pdf.
398
CHAPTER 33
■
From Sequence to Genomic Medicine: Genome Policy Considerations
European Commission (2006). FP7 – Tomorrow’s Answers Start Today. Available at http://ec.europa.eu/research/fp7/pdf/fp7factsheets_en.pdf. Department of Justice, Canada (2005). Promoting Equality: A New Vision. [online] http://canada.justice.gc.ca/chra/en/toc.html. European Medicines Agency (1998). ICH Topic E 5 (R1) Ethnic Factors in the Acceptability of Foreign Clinical Data. Available at http://www.emea.europa.eu/pdfs/human/ich/028995en.pdf. European Medicines Agency (2005). Guideline on Risk Management Systems for Medicinal Products for Human Use. Available at http://www.emea.europa.eu/pdfs/human/euleg/9626805en.pdf. European Parliament (2000). Charter of fundamental rights of the European Union. Article 21 (Non-discrimination). Available at http://www.europarl.europa.eu/charter/pdf/text_en.pdf. Evans, R.W., Williams, G.E., Baron, H.M., Deng, M.C., Eisen, H.J., Hunt, S.A., Khan, M.M., Kobashigawa, J.A., Marton, E.N., Mehra, M.R. et al. (2005). The economic implications of noninvasive molecular testing for cardiac allograft rejection. Am J Transplant 5, 1553–1558. Federal Trade Commission (2006). Facts for Consumers: At-Home Genetic Tests: A Healthy Dose of Skepticism May Be the Best Prescription. Available at http://www.ftc.gov/bcp/edu/pubs/consumer/health/hea02.htm. Fernandez, C.V., Kodish, E., Taweel, S., Shurin, S. and Weijer, C. (2003a). Disclosure of the right of research participants to receive research results: An analysis of consent forms in the Children’s Oncology Group. Cancer 97, 2904–2909. Fernandez, C.V., Kodish, E., Shurin, S. and Weijer, C.Children’s Oncology Group (2003b). Offering to return results to research participants: Attitudes and needs of principal investigators in the Children’s Oncology Group. J Pediatr Hematol Oncol 25, 704–708. Fernandez, C.V., Kodish, E. and Weijer, C. (2003c). Informing study participants of research results: An ethical imperative. IRB 25(3), 12–19. Fernandez, C.V., Kodish, E. and Weijer, C. (2003d). Importance of informed consent in offering to return research results to research participants. Med Pediatr Oncol 41, 592–593. Food and Drug Administration (FDA), “Investigational New Drug Applications and New Drug Applications, Final Rule,” Federal Register, February 11, 1998; 63(8):6854-62 (codified at 21 CFR 312.33(a)(2) and 21 CFR 314.50(d)(5)). Food and Drug Administration (FDA). Innovation or Stagnation? Challenge and Opportunity on the Critical Path to New Medical Products. Available at http://www.fda.gov/oc/initiatives/criticalpath/whitepaper.html (2004). Food and Drug Administration (FDA) (2005a). Guidance for Industry – Collection of Race and Ethnicity in Clinical Trials. Available at http://www.fda.gov/cber/gdlns/racethclin.pdf. Food and Drug Administration (FDA) (March 2005b). Guidance for Industry – Pharmacogenomic Data Submissions. Available at http://www.fda.gov/cder/guidance/6400fnl.pdf. Food and Drug Administration (FDA) (March 2005c). Guidance for Industry and FDA Staff. Class II Special Controls Guidance Document: Drug Metabolizing Enzyme Genotyping System . Available at http://www.fda.gov/cdrh/oivd/guidance/1551.pdf. Food and Drug Administration (FDA) (March 2005d). Guidance for Industry and FDA Staff. Class II Special Controls Guidance Document: Instrumentation for Clinical Multiplex Test Systems. Available at http://www.fda.gov/cdrh/oivd/guidance/1546.pdf. Food and Drug Administration (FDA) (April 2005e). Draft – DrugDiagnostic Co-Development Concept Paper. Available at http:// www.fda.gov/cder/genomics/pharmacoconceptfn.pdf.
Food and Drug Administration (FDA) (December 28, 2005f). Nucleic Acid Based In Vitro Diagnostic Devices for Detection of Microbial Pathogens-Draft Guidance for Industry and FDA Staff. Available at http://www.fda.gov/cdrh/oivd/guidance/1560.pdf. Food and Drug Administration (FDA) (March 2006a). Innovation or Stagnation? Critical Path Opportunities List. Available at http:// www.fda.gov/oc/initiatives/criticalpath/reports/opp_list.pdf. Food and Drug Administration (FDA) (September 2006b). Draft Guidance for Industry and FDA Staff-Commercially Distributed Analyte Specific Reagents (ASRs): Frequently Asked Questions. Available at http://www.fda.gov/cdrh/oivd/guidance/1590.pdf. Food and Drug Administration (FDA) (September 2006c). Draft Guidance for Industry, Clinical Laboratories, and FDA Staff-In Vitro Diagnostic Multivariate Index Assays. Available at http:// www.fda.gov/cdrh/oivd/guidance/1610.pdf. Food and Drug Administration (FDA) (June 2006d). Guidance for Industry and Review Staff. Recommended Approaches to Integration of Genetic Toxicology Study Results. Available at http://www.fda.gov/cder/guidance/6848fnl.pdf. Food and Drug Administration (FDA) (June 2007). Guidance for Industry and FDA Staff. Pharmacogenetic Tests and Genetic Tests for Heritable Markers. Available at http://www.fda.gov/cdrh/ oivd/guidance/1549.pdf. Foster, M. (2005). Genetics and social identity after the HapMap. Nat Rev Genet 6, 879. Godard, B., Marshall, J., Laberge, C. and Knoppers, B.M. (2004). Strategies for Consulting with the Community: The cases of four large-scale genetic databases. Sci Eng Ethics 10, 457–477. Government Accountability Office (2006). Nutrigenetic Testing: Tests Purchased from Four Web Sites Mislead Consumers (GAO-06977T). Available at http://www.gao.gov/new.items/d06977t.pdf. Gurwitz, D., Weizman, A. and Rehavi, M. (2003). Education: Teaching pharmacogenomics to prepare future physicians and researchers for personalized medicine. Trends Pharmacol Sci 24, 122–125. Guttmacher, A.E., Jenkins, J. and Uhlmann, W.R. (2001). Genomic medicine: Who will practice it? A call to open arms. Am J Med Genet 106(3), 216–222. Haga, S.B. and Beskow, L.M. (2008). Ethical, legal, and social implications of biobanks for genetics research. Adv Genet 60, 505–544. Haga, S.B. and Ginsburg, G.S. (2006). Prescribing BiDil: Is it black and white?. J Am Coll Cardiol 48(1), 12–14. Haga, S.B. and Venter, J.C. (2003). FDA races in wrong direction. Science 301, 466. Haga, S.B. and Willard, H.F. (2006). Defining the spectrum of genome policy. Nat Rev Genet 7(12), 966–972. Hahn, R.A., Truman, B.I. and Barker, N.D. (1996). Identifying ancestry: The reliability of ancestral identification in the United States by self, proxy, interviewer, and funeral director. Epidemiology 7, 75–80. Hall, D., Ybazeta, G., Destro-Bisel, G., Petzl-Erler, M.L. and Di Renzo, A. (1999). Variability at the uridine diphosphate glucuronosyltransferase 1A1 promoter in human populations and primates. Pharmacogenetics 9, 591–599. Helgadottir, A., Manolescu, A., Helgason, A., Thorleifsson, G., Thorsteinsdottir, U., Gudbjartsson, D.F., Gretarsdottir, S., Magnusson, K.P., Gudmundsson, G., Hicks, A. et al. (2006). A variant of the gene encoding leukotriene A4 hydrolase confers ethnicity-specific risk of myocardial infarction. Nat Genet 38, 68–74. Holtzman, N.A. and Watson, M.S. (1997). Promoting Safe and Effective Genetic Testing in the United States: Final Report of the Task Force on Genetic Testing (eds). Johns Hopkins University Press, Baltimore.
References
Ibarreta D, Bock, AK, Klein C, Rodriguez-Cerezo E. Towards quality assurance and harmonization of genetic testing services in the EU. Available at ftp://ftp.jrc.es/pub/EURdoc/eur20977en.pdf (2003). Institute of Medicine (2003). Unequal Treatment: Confronting Racial and Ethnic Disparities in Healthcare. National Academy Press, Washington, DC. International HapMap Consortium (2005). A haplotype map of the human genome. Nature 437, 1229–1320. Ioannidis, J.P., Ntzani, E.E. and Trikalinos, T.A. (2004). Racial differences in genetic effects for complex diseases. Nat Genet 36, 1312–1318. Jacobellis, J., Martin, L., Engel, J., VanEenwyk, J., Bradley, L.A., Kassim, S., Jorgensen, C., Litch, J.A. and Myers, M.F. (2004). Genetic testing for breast and ovarian cancer susceptibility: Evaluating direct-to-consumer marketing – Atlanta, Denver, Raleigh-Durham, and Seattle, 2003. MMWR Morb Mortal Wkly Rep 53, 603–606. Jenkins, J. and Calzone, K.A. (2007). Establishing the essential nursing competencies for genetics and genomics. J Nurs Scholarsh 39(1), 10–16. Jensen, K. and Murray, F. (2005). Intellectual property landscape of the human genome. Science 310, 239–240. Johnson, C.E. (1974). U.S. Bureau of the Census, Technical Paper No. 31, Washington, D.C. Kahn, J.D. (2005). From disparity to difference: How race-specific medicines may undermine policies to address inequalities in health care. So Calif Interdiscipl Law J 15, 105–129. Lewis, J.A., Calzone, K.M. and Jenkins, J. (2006). Essential nursing competencies and curricula guidelines for genetics and genomics. MCN Am J Matern Child Nurs 31(3), 146–155. Lin, Z., Owen, A.B. and Altman, R.B. (2004). Genomic research and human subject privacy. Science 305, 183. MacNeil, S.D. and Fernandez, C.V. (2006). Offering results to research participants. BMJ 332, 188–189. Matchar, D.B., Thakur, M.E., Grossman, I., McCrory, D.C., Orlando, L.A., Steffens, D.C., Goldstein, D.B., Cline, K.E., Gray, R.N. (2006). Testing for Cytochrome P450 Polymorphisms in Adults With Non-Psychotic Depression Treated With Selective Serotonin Reuptake Inhibitors (SSRIs). Evidence Report/ Technology Assessment No. 146. (Prepared by the Duke Evidencebased Practice Center under Contract No. 290-02-0025.) AHRQ Publication No. 07-E002. Rockville, MD: Agency for Healthcare Research and Quality. McGovern, M.M., Benach, M.O., Wallenstein, S., Desnick, R.J. and Keenlyside, R. (1999). Quality assurance in molecular genetic testing laboratories. JAMA 281, 835–840. Metcalfe, S., Hurworth, R., Newstead, J. and Robins, R. (2002). Needs assessment study of genetics education for general practitioners in Australia. Genet Med 4, 71–77. Monaghan, G., Ryan, M., Seddon, R., Hume, R. and Burchell, B. (1996). Genetic variation in bilirubin UPD-glucuronosyltransferase gene promoter and Gilbert’s syndrome. Lancet 347, 578–581. Mothelet, V.G. and Herrera, S. (2005). Mexico launches bold genome project. Nat Biotech 23, 1030. Myers, M.F., Chang, M.H., Jorgensen, C., Whitworth, W., Kassim, S., Litch, J.A.,Armstrong, L., Bernhardt, B., Faucett,W.A., Irwin, D. et al. (2006). Genetic testing for susceptibility to breast and ovarian cancer: Evaluating the impact of a direct-to-consumer marketing campaign on physicians’ knowledge and practices. Genet Med 8, 361–370. O’Malley, M.A., Bostanci, A. and Calvert, J. (2005). Whole-genome patenting. Nat Rev Genet 6, 502–506. National Cancer Institute (2006). The TAILORx Breast Cancer Trial. Available at http://www.cancer.gov/ clinicaltrials/digestpage/TAILORx.
■
399
National Coalition for Health Professional Education in Genetics (2005). Core Competencies in Genetics Essential for All HealthCare Professionals. Available at http://www.nchpeg.org/core/ Corecomps2005.pdf. National Health Service’s National Genetics Education and Development Centre and the Skills for Health (2007). Competences for Genetics in Clinical Practice. Available at http:// www.geneticseducation.nhs.uk/develop/index.asp?id44. National Human Genome Research Institute (2007). Medical Sequencing Program and Current Initiatives. Available at http:// www.genome.gov/15014882#2. National Institutes of Health Revitalization Act. Pub. L. No. 103-43 107 Stat. 122 (1993). National Institutes of Health (2005). Best Practices for the Licensing of Genomic Inventions: Final Notice. Fed Regist 70, 18413–18415. National Institutes of Health (2006a). Request for Information (RFI): Proposed Policy for Sharing of Data obtained in NIH supported or conducted Genome-Wide Association Studies (GWAS) (NOTOD-06-094). Available at http://grants.nih.gov/grants/guide/ notice-files/NOT-OD-06-094.html. National Institutes of Health (2006b). Amendments and Corrections to RFA-HG-06-001 “Genome Sequencing Centers.” Available at http://grants2.nih.gov/grants/guide/notice-files/NOT-HG-06010.html. NIH Chemical Genomics Center (2005). Available at http://www. ncgc.nih.gov/. Parra, E.J., Marcini, A., Akey, J., Martinson, J., Batzer, M.A., Cooper, R., Forrester, T., Allison, D.B., Deka, R., Ferrell, R.E. et al. (1998). Estimating African American admixture proportions by use of population-specific alleles. Am J Hum Genet 63, 1839–1851. Partridge, A.H., Hackett, N., Blood, E., Gelman, R., Joffe, S., BauerWu, S., Knudsen, K., Emmons, K., Collyar, D., Schilsky, R.L. et al. (2004). Oncology physician and nurse practices and attitudes regarding offering clinical trial results to study participants. J Natl Cancer Inst 96(8), 629–632. Phillips, K.A. and Van Bebber, S.L. (2004). A systematic review of cost-effectiveness analyses of pharmacogenomic interventions. Pharmacogenomics 5, 1139–1149. Porteous, T., Bond, C., Robertson, R., Hannaford, P. and Reiter, E. (2003). Electronic transfer of prescription-related information: comparing views of patients, general practitioners, and pharmacists. Br J Gen Pract 53(488), 204–209. Rankin, J. and Bhopal, R. (1999). Current census categories are not a good match for identity. BMJ 318, 1696. Rigby, H. and Fernandez, C.V. (2005). Providing research results to study participants: Support versus practice of researchers at the American Society of Hematology annual meeting. Blood 106, 1199–1202. Rosenberg, N.A., Pritchard, J.K., Weber, J.L., Cann, H.M., Kidd, K.K., Zhivotovsky, L.A. and Feldman, M.W. (2002). Genetic structure of human populations. Science 298, 2381–2385. Sankar, P., Cho, M.K., Condit, C.M., Hunt, L.M., Koenig, B., Marshall, P., Lee, S.S. and Spicer, P. (2004). Genetic research and health disparities. JAMA 291, 2985–2989. Schulz, C.J., Riddle, M.P., Valdimirsdottir, H.B., Abramson, D.H. and Sklar, C.A. (2003). Impact on survivors of retinoblastoma when informed of study results on risk of second cancers. Med Pediatr Oncol 41(1), 36–43. Secretary’s Advisory Committee on Genetic Testing (2000). Enhancing the Oversight of Genetic Testing: Recommendations of the SACGT. Available at http://www4.od.nih.gov/oba/sacgt/reports/ oversight_report.htm.
400
CHAPTER 33
■
From Sequence to Genomic Medicine: Genome Policy Considerations
Senior, P.A. and Bhopal, R. (1994). Ethnicity as a variable in epidemiological research. BMJ 309, 327–330. Shalowitz, D.I. and Miller, F.G. (2005). Disclosing individual results of clinical research: Implications of respect for participants. JAMA 294, 737–740. Shalowitz, D.I. and Miller, F.G. (2008). Communicating the results of clinical research to participants: attitudes, practices, and future directions. PLoS Med 5(5), e91. Shields, A.E., Fortun, M., Hammonds, E.M., King, P.A., Lerman, C., Rapp, R. and Sullivan, P.F. (2005). The use of race variables in genetic studies of complex traits and the goal of reducing health disparities: A transdisciplinary perspective. Am Psychol 60, 77–103. Shriver, M.D., Parra, E.J., Dios, S., Bonilla, C., Norton, H., Jovel, C., Pfaff, C., Jones, C., Massac, A., Cameron, N. et al. (2003). Skin pigmentation, biogeographical ancestry and admixture mapping. Hum Genet 112, 387–399. Sinha, M., Larson, E.K., Elston, R.C. and Redline, S. (2006). Selfreported race and genetic admixture. N Eng J Med 354, 421–422. Snowdon, C., Garcia, J. and Elbourne, D. (1998). Reactions of participants to the results of a randomised controlled trial: Exploratory study. BMJ 317(7150), 21–26. Tang, H. (2006). Confronting ethnicity-specific disease risk. Nat Genet 38, 13–15. Tang, H., Quertermous, T., Rodriguez, B., Kardia, S.L., Zhu, X., Brown, A., Pankow, J.S., Province, M.A., Hunt, S.C., Boerwinkle, E. et al. (2005). Genetic structure, self-identified race/ethnicity, and confounding in case-control association studies. Am J Hum Genet 76, 268–275. Torrance, N., Mollison, J., Wordsworth, S., Gray, J., Miedzybrodzka, Z., Haites, N., Grant, A., Campbell, M., Watson, M.S., Clarke, A. et al. (2006). Genetic nurse counsellors can be an acceptable and costeffective alternative to clinical geneticists for breast cancer risk genetic counselling. Evidence from two parallel randomised controlled equivalence trials. Br J Cancer 95(4), 435–444. U.K. Human Genetics Commission (2003). Genes Direct: Ensuring the effective oversight of tests supplied directly to the public. Available at http://www.hgc.gov.uk/UploadDocs/Contents/Documents/ Genes%20direct%20-%20FULL%20REPORT%20FINAL.pdf.
United Nations Educational, Scientific and Cultural Organization (UNESCO) (1997).Universal Declaration on the Human Genome and Human Rights. Available at http://portal.unesco.org/shs/en/ ev.php-URL_ID1881 & URL_DODO_TOPIC & URL_ SECTION201.html. United Nations Educational, Scientific and Cultural Organization (UNESCO) (2003). A declaration on human genetic data. Available at http://portal.unesco.org/shs/en/ev.php-RL_ID1882&URL_ DODO_TOPIC&URL_SECTION201.html. The ENCODE Project Consortium (2007). Identification and analysis of functional elements in 1% of the human genome by the ENCODE pilot project. Nature 447, 799–816. Wall, A. (July 16, 2007). Races of Mexico and the Mexican Genome Project. Available at http://www.mexidata.info/id1442.html. Weigmann, K. (2006). Racial medicine: Here to stay? The success of the International HapMap Project and other initiatives may help to overcome racial profiling in medicine, but old habits die hard. EMBO Rep 7(3), 246–249. The Wellcome Trust (1996). Summary of principles agreed at the International Strategy Meeting on Human Genome Sequencing. Available at http://www.gene.ucl.ac.uk/hugo/bermuda.htm. The Wellcome Trust (2003). Sharing Data from Large-scale Biological Research Projects: A System of Tripartite Responsibility. Available at http://www.genome.gov/Pages/Research/ WellcomeReport0303.pdf. Wendler, D. and Emanuel, E. (2002). The debate over research on stored biological samples: What do sources think?. Arch Intern Med 162(13), 1457–1462. Wendler, D. and Pentz, R. (2007). How does the collection of genetic test results affect research participants. Am J Med Genet A. DOI 10.1002/ajmg.a.31823. World Health Organization (2005). Genetics, genomics and the patenting of DNA: Review of potential implications for health in developing countries. Available at http://www.who.int/genomics/FullReport.pdf. Zerhouni, E. (2003). The NIH Roadmap. Science 302, 63–72. Zerhouni, E.A. (2005). Translational and clinical science – time for a new vision. N Engl J Med 353, 1621–1623.
CHAPTER
34 Educational Strategies in Genomic Medicine Charles J. Epstein
INTRODUCTION Education for what? Education for whom? What is there about genomic medicine that requires us to consider matters of education? The answers to these questions reside, of course, in what we mean by genomic medicine (Epstein, 2006), and for the purposes of introducing this article I shall cite two definitions: 1. the use of industrialized methods of data acquisition and analysis to improve medical care, including prognostics, diagnostics, preventive intervention, therapeutic selection, and individualized treatment based on the complex interaction between inherited and acquired elements of human variation (Harvard-MIT Division of Health Sciences and Technology, 2004), and 2. an approach that will build on the comprehensive nature of the genome sciences … As a clinical paradigm, genomic medicine will provide global, comprehensive, and multidimensional treatment and management strategies based on the science now emerging from the study of genomes (Willard, 2004) … Having access to the entire human sequence is a necessary but insufficient prerequisite for genomic medicine. What is equally important is having the technology at hand to reliably visualize individual genomes (and their derivatives, the transcriptome, proteome and metabolome) for health information that, in combination with clinical data, can Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
contribute to assessment of individual risks and guide clinical management and decision-making. [This makes] the prospect for developing truly individualized care … even more real (Willard et al., 2005). These two definitions are quite broad. They certainly encompass the “medicine” side of “genomic medicine,” but there is also emphasis in the first definition on “industrialized methods of data acquisition and analysis” and in the second on “global, comprehensive, and multidimensional … strategies” and “the technology to reliably visualize individual genomes.” It is clear that, as the table of contents of this handbook would suggest, what is really new in genomic medicine is the application of genomic technology on a large scale to virtually every problem of medicine. We might call this the process of genomic medicine. However, these definitions also make it clear that an, if not the principal, intent of genomic medicine is what has been variously described as preventive intervention, individualized treatment, prospective medicine (Snyderman and Williams, 2003) or personalized medicine (Ginsburg et al., 2005) and medicines – the last referring, of course, to pharmacogenetics (Royal Society, 2005). Weston and Hood (2004) combined all of these together as the 3 P’s: predictive, preventative, and personalized medicine. Since the term genomic medicine certainly has considerably wider connotations, it might be more accurate to refer to the 3 P’s as applied genomic medicine, and this is what the following discussion will be concerned with. Copyright © 2009, Elsevier Inc. All rights reserved. 401
402
CHAPTER 34
■
Educational Strategies in Genomic Medicine
Key to all of this, of course, will be risk assessment based on testing, and there are indeed many forms of testing already in use and being developed – biochemical, genetic, genomic, proteomic, metabolomic, and pharmacogenetic (or pharmacogenomic). From the point of view of public perception, testing is where things are headed when we talk about genomic medicine. Indeed, there appears to be pervasive belief in both scientific and public circles that genetic testing or profiling is going to be the cornerstone of much, if not all of what genomic medicine – in fact, all of medicine – holds for the future (Epstein, 2004). [Genetic testing, risk assessment, and profiling already are and will, to a much greater extent in the future, be based on information gathered by genomic approaches and will utilize genomic technologies. However, for the sake of convenience and in conformity with current usage, I shall use the terms “genetic testing,” “genetic risk assessment,” and “genetic profiling” throughout.] There has been an extensive debate swirling around this notion of the centrality of genetic testing in the future of medicine, which I shall not repeat here, but there a few important points that are worth noting: ●
●
●
In most instances, genetic risk assessment will attain sufficient predictive power to be of use if – and only if – analyses of many genetic loci are combined with evaluations of non-genetic life-style and environmental factors (Janssens et al., 2006). As a result, complicated algorithms, the nature of which may not, for proprietary reasons, always be revealed, will often be required to compute the risks (Food and Drug Administration, 2006). The results of such risk assessments will not be absolutely definitive. The information will be probabilistic, and we shall always be dealing with estimates of risk, not predictions of certain outcomes. Even though risk assessment or profiling can ultimately be done does not mean that it actually will or, indeed, should be done. This will ultimately be determined by whether genetic testing really offers more than currently available forms of risk assessment in terms of predictive power and securing compliance – in other words, by whether it will have greater clinical utility and would enhance a person’s motivation to comply with recommendations for therapy and changes in life-style. [For a discussion of clinical utility and its definitions, see Burke et al., 2006; Grosse and Khoury, 2006; Scheuner and Rotter, 2006.] My assumption is that genetic profiling and risk assessment will ultimately be done, and that the pressures from both consumers and the developers of tests – and perhaps from the medical profession as well – will lead to its being done quite extensively. The announcement of a $10 million prize to the first group that can “successfully map 100 human genomes in 10 days … to usher in a new era of personalized preventative medicine” (X Prize Foundation, 2006) bears dramatic testimony to this, as does the NIH Request for Applications for the $1000 genome (National Institutes of Health, 2004).
The definitions and issues that have just been outlined describe the what in the first question – education for what?
More specifically, the what is the ability for the whom to understand and participate in the process of genetic risk assessment and those things related to it, and the whom consists of both the providers and the recipients of genetic information and of the recommendations that are based on it, as well as those who make and implement the public policies that ultimately determine how genomic medicine will be practiced. The educational needs of all of these groups were extensively discussed in Assessing Genetic Risks: Implications for Health and Social Policy, the report of the Institute of Medicine Committee on Assessing Genetic Risks (Andrews et al., 1994). Remarkably, the issues raised in that report are as valid today as they were then. The only real difference is that we are much farther along the pathway of genetic testing and risk assessment, all of which was predicted in the report, than we were when the report was written.
GENETIC AND GENOMIC LITERACY OF THE PUBLIC AND MAKERS OF PUBLIC POLICY The central thesis of the Institute of Medicine report with regard to public education in genetics is the need to develop a genetically literate public (Andrews et al., 1994, p. 185 et seq.). In this context, the term public encompasses both the policy makers and the consumers or recipients of genetic testing and risk assessment, as well as the public at large, and genetically literate translates to being able to understand basic biological research and the personal and health implications of genetics and to participate effectively in the public debate of relevant public policy issues. Of particular importance with regard to genomic medicine, the report placed emphasis on attaining an understanding of the limits of genetics so as not to fall into the trap of genetic determinism – the belief that genetics is completely predictive. Another formulation of the reasons for improving education in basic genetics suggests that the goals are to allow individuals to understand general genetic concepts, applications, and social and ethical issues and to become informed users of genetic and genomic technologies and their applications (Haga, 2006). Both of these formulations are certainly reasonable for the public at large, consumers and policy makers alike, but it is the last goal in each formulation that is particularly germane to the consumers of the genetic testing and risk assessment applications of genomic medicine. The probabilistic nature of risk assessment and pharmacogenetic testing, whether for the chance of developing diabetes or coronary artery disease or for benefiting from a particular drug, will require a level of genetic and genomic understanding that has not heretofore been required on the part of the general population of patients and consumers of medical services. It is true, of course, that genetic counseling, as it is presently being carried out for many conditions, does involve risk assessment based on genetic testing and requires the presentation and handling of probabilistic information. The best
Genetic and Genomic Literacy of the Public and Makers of Public Policy
example of this in current medical practice is risk assessment for breast and ovarian cancer based on the testing for mutations in BRCA1 and BRCA2. However, there are two major differences between this type of risk assessment and the more widespread risk assessment for common (complex) conditions and pharmacogenetic testing that will constitute genomic medicine. One is that cancer risk assessment is directed at a targeted and restricted population – those already suspected to be at risk because of prior medical or family history. This makes it possible for a significant educational component to be provided by a genetic counselor or other highly trained (in genetics and genomics) health professional – a luxury that may not be possible for the population at large. Therefore, as will be discussed below, the burden of communicating and interpreting the results of risk assessments will likely fall to individuals less expert in genetics and genomics and with less time to spend with individual patients. It is this fact that makes it essential that the members of the general public have a basic knowledge and appreciation of genetic and genomic principles if they are to participate intelligently in the risk assessment process. And, as has already been noted, one principle that it is essential for the public to understand is that probabilities are not certainties. I have used the term “probabilistic” several times in the preceding paragraphs, and a few amplifying comments are warranted. The result of a presymptomatic genetic test or set of tests performed before the onset of disease will be certain only when the penetrance of the condition(s) that are being tested for is complete (100%) within a specific time frame. In all other situations, risks are presented as probabilities. These probabilities may be expressed quantitatively, usually as a percentage (e.g., if a particular mutation of BRCA1 is present, the risk of developing breast cancer in the next X years is Y %) or qualitatively (from very low to very high). While the quantitative expression of probabilities is more precise, not all individuals are able to fully comprehend them. Sometimes, especially when the risks are relatively low, the probabilities may be inverted and given in terms of not developing disease (e.g., the odds are 10 to 1 that breast cancer will not occur). Sometimes risks are given in relative rather than absolute terms (e.g., the risk of developing disease Z is Y times greater than that of the general population), an approach that is frequently lacking because the “general population” risks are often not stated. However, what is most important is that the interpretation of risks can be greatly influenced by precisely how the risks are formulated and presented and by the preconceptions of the recipients (Tversky and Kahneman, 1981). These factors will determine the actions that will ultimately be taken. For this reason, there needs to be sufficient understanding on both sides of how risks, the probabilistic information to which I have been referring, should be presented and interpreted, and this will require education of all concerned. Returning to the rationale for improved public genetic and genomic literacy, one other thing that can be added is the increase in direct-to-consumer advertising and promotion of genetic tests (Wolfberg, 2006). Just at the time that this chapter was being written, a new home gene test for schizophrenia
■
403
was being touted (SureGene, 2006), and the Federal Trade Commission felt it necessary to issue a warning against at-home genetic testing in general (Federal Trade Commission, 2006). Clearly, an educated public will be essential if the potential problems likely to accrue from unbridled genetic testing and risk assessment are to be prevented. The Institute of Medicine report (Andrews et al., 1994) lamented the relatively low level of genetic literacy in the public, and the same still appears to the case today (Haga, 2006). Since it is likely that the solution to this problem ultimately resides in the education of children, when their flexibility and scientific interest are at a maximum (Andrews et al., 1994), the 2005 report of the National Assessment of Educational Progress with regard to science in American schools is particularly sobering (Grigg et al., 2006). Since 1996, there has been an increase in the average science score (as well as in the life science subscore, which would include genetics) in grade 4, no change in grade 8, and a decline in grade 12. In grade 12, nearly half (46%) of students scored below the basic level. [An example of a basic question is to identify the relationship between genes and enzymes.] How to go about increasing the amount of genetics and genomics in the curricula of students from pre-school through college and the quality of genetics and genomics instruction at all levels is beyond the scope of this chapter, but the subject was dealt with extensively in the Institute of Medicine report (Andrews et al., 1994, p. 189 et seq.), Two summary recommendations are worth citing: Variation and kinship in the context of the environment should be the fundamental concepts of genetics/genomics education for the public [note the emphasis is on genetics/genomics and the environment, not just genetics/genomics alone], and genetics/genomics education should include the ethical, legal, and social issues stemming from science and technology. To these two recommendations may be added a third, that the concept of using family health histories is a way to motivate students to explore genetics in a way that is personal and has direct meaning for them. Generating a personal family health history by individuals as the first step in learning about potential health risks has recently attained prominence with the launching of the US Surgeon General’s Family History Initiative (United States Department of Health and Human Services, 2006). Professionally obtained family histories have already been recognized as a comprehensive risk assessment method for the common diseases of adulthood (Scheuner et al., 1997). Useful teaching resources in genetics and genomics were recently reviewed by Haga (2006), who cites numerous online resources for genetic and genomic education. However, it will take more than on-line resources to accomplish the goal. What will ultimately be required is a systematic integration of genetics and genomics into all levels of education, both vertically (through the years) and horizontally (across subject areas) (Haga, 2006), and this will have to be done by the educational establishment. Although it will ultimately be the educators who will have to accomplish the task, human geneticists and genetic counselors and their professional societies (such as the American Society of Human Genetics, the American College of Medical
404
CHAPTER 34
■
Educational Strategies in Genomic Medicine
Genetics, and the National Society of Genetic Counselors) can and should serve as advocates and catalysts for change and as sources of expertise in genetics and genomics and related areas. To this end, the American Society of Human Genetics has established a Mentor Network to work with teachers to create in-depth and personal learning opportunities for students, has created a database of education standards for all grades between kindergarten and 12th grade, and commissioned a review of the genetics content in introductory biology courses for nonscience majors (Hott et al., 2002). The consumers of genetic services and the makers of public policy have been lumped together in this discussion of genetic literacy, on the assumption that they all need to become genetically literate. However, more than just the basic elements of genetic and genomic literacy will certainly be required for those who make policy and those who advise them. And, when I speak of policy makers, I am referring to an extremely broad range of individuals who will need to be educated about genomics and genomic medicine when they are students in universities and professional schools and particularly when they have taken up their professional careers. Included in this group, at the United States federal level, are members of Congress and their staffs and executives and members of the Department of Health and Human Services and many of its constituent organizations (including but not limited to the Food and Drug Administration, the Centers for Medicare and Medicaid Services, the Agency for Health Care Research and Quality, the Centers for Disease Control and Prevention, the Health Resources and Services Administration, and the National Institutes of Health with its many Institutes). Also at the federal level are the White House Office of Science and Technology Policy and the Domestic Policy Council. To this list can be added the parallel group of individuals operating at the state government level, along with officials of insurance companies, health maintenance organizations, and hospital centers. Individuals from all of these groups and organizations may (many certainly will) directly or indirectly influence how genetic testing and genomic medicine will be practiced, financed, and regulated. Many approaches, ranging from lectures and seminars to courses in schools of law, government, and public policy, will be required to educate these people.
EDUCATION OF THE PROVIDERS OF HEALTH CARE As genetically and genomically literate as the consumers of genomic medicine may become, the burden for conducting and interpreting testing will fall, of course, on the providers of health care – a diverse group of professionals involved in the delivery of preventive and therapeutic medical services to the public. These would include, in particular, primary care physicians, physician specialists, medical geneticists, genetic counselors and nurses, disease-specific providers and counselors, the public health community, pharmacists, and other allied health professionals.
The choice of educational strategies to bring health care providers into the world of genomic medicine will depend, of course, on what the various types of professionals will be doing. Once again, this subject was discussed extensively in the 1994 Institute of Medicine report. The basic premise of the discussion was the following: Historically, genetic tests have been administered and interpreted by highly trained health professionals working in academic health settings, usually with a strong genetics research and service record. In the future, however, genetic tests will become available for a growing variety of monogenic and complex diseases and for susceptibility to more common disorders such as breast, colon, and other cancers. Testing on such a broad scale will necessarily move us beyond the models of service delivery and professional roles that have characterized genetic testing and screening in the past. Increasingly, genetic tests will be offered and interpreted within the context of the mainstream of medicine in primary care practice including pediatrics, obstetrics, internal medicine, and family practice in a variety of individual and group practice settings. This exciting and challenging prospect for the future involves a large pool of potential personnel for genetic testing, screening, education, and counseling, but will they be prepared to play this role? (Andrews et al., 1994, p. 202 et seq.)
Others have echoed the idea that primary care will be the principal locus for genetic testing. We have to anticipate that every health care provider is going to become a genetic counselor in the next 10 years or they’re not going to be doing a good job (Jones and Collins, 2003). Clearly, the day-to-day genetic health providers of the future will be the primary care clinicians (Caskey, 2001).
Explicit or implicit in these statements is the assumption that there will not be enough genetic professionals to handle the burden that wide-scale genetic testing and risk assessment would entail and that primary care practitioners will be have to assume the responsibility. In a sense, this is a negative reason for doing so. However, a Genetics White Paper published by the British National Health Service presents the argument in positive terms: GPs, practice nurses and the primary care practitioner will all be able to help their patients benefit from the new genetic knowledge and its applications. They already understand the long term, psychosocial aspects of illness. They work with individuals in the context of their families over time. They are adept at identifying health problems and making appropriate referrals. They co-ordinate the care of the affected patient. And, they are at the forefront of health promotion and prevention. [Therefore,] the roles for primary care in genetics [will include] managing patients’ concerns and expectations, identifying genetic conditions, assessing risk, managing risk, screening, [and] testing … (Department of Health, 2003).
Education of the Providers of Health Care
There is certainly no question that the supply of trained and certified medical geneticists and genetic counselors is very limited. At the latest tally, which covers through the 2007 certifying examinations, there were, in the United States, 1253 certified physician clinical geneticists, 900 certified PhD geneticists, most working as laboratory directors, and 2512 certified genetic counselors (American Board of Genetic Counseling, 2008; American Board of Medical Genetics, 2008). Since these figures date back as far as the first certifying examinations in 1982, it is clear that the number of geneticists and counselors actually in practice in substantially less than the figures cited. Furthermore, whereas the number of individuals seeking certification in genetic counseling has been increasing over the past decade, the number of new physician clinical geneticists has been declining. Therefore, it is fair to say that there is no way that genetic professionals could handle all of the genetic testing and risk assessment likely to come in the future, even if they wanted to. So, who will do what, and how should they be trained to do it? In a broad assessment of the roles of primary care physicians, specialists, and medical geneticists, Korf (2002a) divided medical disorders in which genetics is involved into three not mutually exclusive three groups – single gene or chromosomal, major gene multifactorial (such as breast cancer), and complex multifactorial (otherwise referred to as common or common complex). The first group is principally within the realm of medical genetics as now practiced, with primary care and specialist physicians being responsible for referral and general management and the medical geneticists and counselors for diagnosis, counseling, and, for the inborn errors of metabolism at least, treatment. In fact, it has been suggested by some that this is just where the focus of medical genetics should remain (Guttmacher et al., 2001; Khoury, 2003). The second and particularly the third groups fall within the purview of genomic medicine, as I have defined it here. For these, the medical geneticist has been assigned a more advisory role with regard to test interpretation and counseling, with primary care and specialist physicians arranging for genetic testing to guide prevention, treatment, and, if necessary, diagnosis. Given their present small numbers, reality dictates that medical geneticists, and genetic counselors for that matter, would certainly not be in a position to become the principal providers of genetic testing and risk assessment for the greater population (Epstein, 2006). Nevertheless, medical geneticists and counselors still possess, albeit with different areas of expertise, knowledge of genetics and genomics and human disease that should be brought to bear on the provision of these services. This can be accomplished through their playing a role in the education of other physicians and health providers (Caskey, 2001), but it also involves serving as designated referral sources for problems that may be too complex for those without substantial knowledge of genetics and genomics to handle alone. The summary of a roundtable discussion of the future of genomic medicine states that: The different responsibilities of the primary care physician, genetic counselor, and medical geneticist need to be examined and perhaps redefined. Currently, there are only
■
405
2,000 genetic counselors in the United States. If genetic counselors are to play a central role as genomic medicine is integrated into clinical practice, there is a need for more genetic counselors. At the same time, other health care providers will need to learn more genomics so they can better inform and treat their patients. The standard practice of care for doctors with regard to genomic medicine needs to be defined (NHGRI Policy Roundtable, 2005).
Although what this statement says is true, it unfortunately gives short shrift to medical geneticists. Given the breadth of their knowledge and expertise, medical geneticists should certainly also have a central role in genomic medicine. Medical geneticists know genetics and how testing is done, are comfortable with family histories and probabilities and with counseling and decision-making, and are already doing genetic testing and risk assessment (Epstein, 2004). A primary care physician has been defined as a generalist physician who provides definitive care to the undifferentiated patient at the point of first contact and takes continuing responsibility for providing the patient’s care (American Academy of Family Physicians, 2006a). In the context of the various statements cited above, family practice medicine and general internal medicine would be regarded as the principal primary care specialties, as would general pediatrics for infants and children. For some women, obstetricians and gynecologists also serve as primary care physicians. Primary Care Physicians However the precise balance is struck between primary care physicians and other potential participants in the provision of genetic testing and risk assessment (or, as previously mentioned, applied genomic medicine), it is clear that all groups will have to become knowledgeable about genetics and genomics and their applications. Although the formal educational process for all physicians begins in medical school, medical students should be expected to be no less genetically and genomically literate than what has been proposed for the public at large, and, if possible, more so. It would certainly be helpful if they could master the principles of human genetics and genomics during their premedical education, but this may not be absolutely necessary. However, what will be essential is that the principles and skills that are necessary to understand the role of genetics and genomics in all of medicine are taught both in formal courses and during the various clinical clerkships. The only hope of having primary care physicians who will be able to practice genomic medicine in the future is to begin training them in medical school now. The Medical School Curriculum At the present time, the teaching of genetics and genomics in medical schools is quite variable in terms of when it is taught, how much time is allocated, and how it is taught, and there is certainly no consistent curriculum (see, e.g., the listings in the online Health Sciences Library: Genetics Curriculum Resource, 2006). A medical school core curriculum in genetics, described
406
CHAPTER 34
■
Educational Strategies in Genomic Medicine
TABLE 34.1 What medical students should appreciate about genetics and genomics prior to graduation from medical school (Genetics Education MSOP Expert Panel, 2004) ●
the potential for genetics to contribute to the development of new approaches to prevention, diagnosis, and treatment of disease
●
the potential for genetics to expand understanding of the basic pathophysiology of all human disease
●
the possibility of using a genetic approach to provide personalized health care with a much greater focus on prevention
●
current limitations in the existing knowledge base
●
that the principles for use of genetic information in decision-making are largely the same as for other areas of medicine
●
the rapidity of the advancing front of knowledge
●
that genetic information may have implications not only for an individual patient, but also for a family, and in some cases an entire community
●
the potentially disconcerting nature of genetic information particularly as it relates to interpretation of predictive tests
●
the need to reduce public fear and misinformation about genetics information
●
the diversity in public understanding of genetic information and evaluation of information sources
●
the need for continued learning and receptivity to advances in knowledge and changes in medical practice
as a work in progress, was developed by the Association of Professors of Human and Medical Genetics and the American Society of Human Genetics (APHMG/ASHG). With a total of 74 requirements presented in terms of essential general medical competencies and specific knowledge, skills, and behaviors, this curriculum covers all of the traditional aspects of medical genetics, and, as might be expected, has a heavy focus on mendelian genetics and monogenic and chromosomal disorders (Association of Professors of Human and Medical Genetics and the American Society of Human Genetics, 2001; see also Korf, 2002b). However, in a more recent formulation of curricular objectives published under the aegis of the Association of American Medical Colleges (Genetics Education MSOP Expert Panel, 2004), which appears to be based on the APHMG/ASHG report, the balance, at least in terms of a set of attitudes (Table 34.1) has shifted quite considerably in the direction of genomic medicine (both overall and applied). The knowledge base, skill set, and set of educational strategies that follow are generally consonant with these attitudes. Of particular note are an understanding of multifactorial inheritance and the role of genetic factors in common disease, concepts of genetic linkage and association, the role of genetics in determining response of individual to environmental factors or pharmacological agents, and the role of genetics in modification of risk for common multifactorial disorders. However, what is not explicitly stated, but definitely should be, is an in depth knowledge, summarized in Table 34.2, of how the whole process of genetic testing and risk assessment, both disease-related and pharmacogenetic, will actually work, at least as far as it can now be visualized. I have outlined these items in detail because I am concerned that in real practice the process of genetic testing and risk assessment could well become an automated operation. A source of DNA would go into a genotyping
robot, the output of which would be sent to a computer from which, after the genotype was combined with personal information, again computer generated, would issue a printout of probabilities and recommendations – essentially a “black box” operation (Epstein, 2004). It can be argued that it will not be necessary for the physician to know everything that goes on inside the black box, just as we do not need to understand how a combustion engine works in order to drive a car, and I would agree. Nevertheless, if a physician is to be able to explain intelligently the basis on which risk assessment is performed and how its outcome is to be interpreted, he or she really needs to understand what goes into the process. Residency Training Although training in genetics and genomic medicine must be an essential component of the medical school curriculum, it is only during the years of residency and specialty training that physicians develop the attitudes and skills that will determine how well they will integrate genomic medicine into their practice of medicine. The Association of American Medical Colleges report (Genetics Education MSOP Expert Panel, 2004) mentioned earlier contains a table entitled “What do physicians have to know, and when do they need to know it?” The table contains a list, part of which is summarized in Table 34.3, of what it is predicted that physicians will need to know when in practice 10 years later (i.e., the year 2014). This projection is based on current knowledge, but it could of course be quite different when the 10 years actually passes and things are possible that we cannot even imagine at present. Unfortunately, acquiring the practice skills and attitudes to make it possible for physicians to know these things in the future is, at the present time, more easily said than done. There are
Education of the Providers of Health Care
TABLE 34.2
407
What needs to be understood to understand genetic testing and risk assessment
●
how susceptibility genes or haplotypes are identified
●
how susceptibility genes confer susceptibility
●
how susceptibility genes are tested for
●
which susceptibility genes should be tested for
●
why it will most often be necessary, for most complex traits, to look at several susceptibility genes simultaneously and how algorithms will be developed to do so
●
how environmental and family history data will become part of the calculation
●
what is the significance of the probabilities or likelihoods that will constitute the output of the calculations
●
how these probabilities form the basis for preventative interventions
TABLE 34.3 Panel, 2004)
■
What physicians will need to know in 2014 about genomic medicine (from Genetics Education MSOP Expert
●
routine use of proteomic screens for very early detection of common cancers
●
increasing use of screening for risk for common disorders to achieve risk stratification and implement prevention strategies
●
use of panels of molecular tests to stratify common disorders such as asthma or hypertension
●
routine molecular characterization of tissues in pathology
●
use of panels of tests to achieve precise diagnosis of monogenic and chromosomal disorders inborn errors of metabolism and other disorders
●
routine use of pharmacogenetic profiling
●
stratification of common disease and selection of specifically targeted therapies
●
use of expression arrays to determine treatment strategies
two factors that have conspired together to make this difficult to do. The first is the tremendous competition for the resident’s time, with the demands of learning how to diagnose and treat a multitude of disorders taking precedence over concerns about genetic risk assessment and testing, which, at least at present, are more hypothetical than real. Somewhat ironically, it is to a great extent exactly the same set of conditions – the common disorders of adulthood – that both compete for the residents’ time and are the targets of testing and risk assessment.There is no easy solution to the problem of competition for time in training programs. Therefore, it will be necessary that the requirements for residency training programs and for specialty certification, as specified by the relevant residency review committees and specialty boards, respectively, insure that genomic medicine is appropriately covered during residency training. The second factor interfering with the integration of genomic medicine into residency training is that those who provide the role models for the residents in training – the more senior residents and fellows, as well as the attending faculty – are themselves, for the most part, not well versed in genetics and genomic medicine. However, this will gradually change with
time, as medical students trained in genomic medicine enter the residency programs and as faculty members themselves become more experienced in the use of genetics and genomics in medical practice (see below). Medical geneticists, when they are available, can and should play an important role in the education of both faculty and residents. The Genetics in Primary Care project (Burke et al., 2002) was an early effort to integrate genetics (not genomics) into primary care residencies that was based on training the faculty first, certainly a reasonable approach. It is not yet known how well this approach has worked. Continuing Medical Education For those physicians currently in practice, the only means for entering the world of genomic medicine will be through participation in various forms of postgraduate medical education. The primary care specialty societies, such as the American Academy of Family Physicians, with its 2005 Annual Clinical Focus on Genomics (American Academy of Family Physicians, 2006b), have already begun to make major efforts in this direction. However, even though such efforts are to be encouraged, I am concerned that they will not prove to be a highly effective
408
CHAPTER 34
■
Educational Strategies in Genomic Medicine
means of education, again because of the competition for the physician’s time and interest and the lack of a substantial background in the principles of genetics and genomics. Medical Specialists and Subspecialists Everything that has been said about the training of primary care physicians applies to specialist physicians as well. However, nonprimary care specialists and subspecialists will need to acquire a more in depth knowledge about those areas of genomic medicine that apply particularly to their areas of practice. It is reasonable that this should occur during their residency and fellowship training, and I anticipate that the necessary instruction will be built into the training programs. Again, specialty societies have recognized that specialists need to be appropriately trained in genetics and genetic testing. For example, the American Society for Clinical Oncology (ASCO), which had earlier developed a curriculum in cancer genetics (American Society of Clinical Oncology, 1997), affirmed the importance of training oncology practitioners to become more knowledgeable with regard to clinical genetics and called upon a broad range of organizations to work in concert to address the important issues related to credentialing of clinical cancer genetics practitioners (American Society of Clinical Oncology, 2003). Although probably not exactly what ASCO had in mind, the statement concerning credentialing raises another possibility for the training of specialists and subspecialists in genomic medicine, and that is the creation of specialty and subspecialty training programs in genetics and genomic medicine (Korf, 2005). Within the American Board of Medical Specialties certifying system, this could take the form of individual or conjoint (between two specialty boards) specialty training programs and certificates. The logical board to partner in conjoint programs would be the American Board of Medical Genetics, and a series of subspecialities such as cancer genetics, cardiovascular genetics, and genetic testing and risk assessment could be envisioned. Medical Geneticists Mention has already been made of the role that might, and I believe should, be played by medical geneticists in the genomic medicine of the future. Medical genetics is the primary specialty recognized by the American Board of Medical Specialties that deals with both the clinical and laboratory aspects of genetics as applied to human health and disease. Clinical medical geneticists complete a 2 year residency program that generally follows 2 or more years in another medical specialty, primary care or otherwise. The current formulation of the training in clinical genetics is summarized in the Program Requirements (Accreditation Council for Graduate Medical Education, 2007), which describe what medical geneticists do and specify how medical genetics training programs are to operate. These descriptions and specifications are, of course, quite broad, and not every medical geneticist is expert in or practices all of the components listed. In fact, whether formally or informally defined, there are areas of specialization within clinical genetics, the three most important
being dysmorphology, metabolic diseases, and prenatal diagnosis. Nevertheless, all medical geneticists are expected to have a general knowledge of all of the areas specified in the program requirements. However, despite allusions to genetic risk assessment and the interpretation of clinical and laboratory genetic tests, there is relatively little in the requirements that addresses the issues most relevant to genomic medicine. As a result, the training presently being provided to clinical genetics residents is, in most instances, not sufficient to allow them to assume a major role in genomic medicine, whether as a provider, consultant, or teacher (Epstein, 2005). What is lacking is greater and more intensive instruction in population genetics and epidemiology (genetic and otherwise), genomics, pharmacogenetics, bioinformatics, the principles of risk assessment, and the genetics of the common adult diseases and complex traits (Epstein, 2006). Therefore, training programs in medical genetics need to be reformulated to incorporate these subjects, and discussions about how this might be done are currently under way in organized medical genetics (Korf, 2005). In addition to training medical geneticists for their future responsibilities, such a refocusing of medical genetics training could also have the effect of attracting more physicians to the field, which is essential if the manpower deficit mentioned earlier is to be alleviated. Genetic Counselors As was noted earlier, there is a general presumption that genetic counselors will be heavily involved in genetic testing and risk assessment in the future. The training and certification of genetic counselors is carried out under the aegis the American Board of Genetic Counseling, and the specified curricular requirements, like those for medical geneticists, are quite broad (American Board of Genetic Counseling, 2006). Of specific relevance to genomic medicine/genetic testing and risk assessment, the curriculum specifies didactic training in the principles of human genetics (including, among a longer list of topics, population and quantitative genetics, the basis of human variation and susceptibility, and family history and pedigree analysis), the applicability of related sciences (including molecular and cancer genetics) to medical genetics, and the principles and practice of clinical/ medical genetics (including the clinical features and natural history of a broad range of genetic diseases, complex common disorders and syndromes of unknown etiology, genetic testing of all types, utilizing risk assessment skills, and the use of genetics literature, databases, and computerized tools). The training that genetic counselors now receive is already well along the way to enabling them to play a major role in applied genomic medicine, as they now do in cancer risk assessment, and it should be relatively easy to modify the curriculum in the future as genetic testing and risk assessment expand. However, what is less clear at the present time is just what the role of genetic counselors will and should be. The traditional model for genetic counseling has been for the counselors to work with and under the supervision of clinical geneticists. But, whereas this model generally applies to conventional forms of counseling, it seems to have become less applicable to genetic risk assessment, as for breast and
Education of the Providers of Health Care
ovarian cancer, and even to prenatal diagnosis, where the supervising physician may be an oncologist or obstetrician who is much less knowledgeable about genetics and genomics than the counselor. And, beyond this, there is a movement within the genetic counseling community to obtain the legal authority through licensure to operate wholly independently of physicians and medical geneticists. A discussion of whether or not this would be desirable is not germane to the purposes of this chapter. Nevertheless, from the educational point of view, whether or not such independence is attained, both the formal and the postgraduate/continuing education of counselors will have to insure that they possess the requisite knowledge and skills to enable them to function effectively. Genetic Nurses In recent years, nurses in the United States with genetic training have begun to aspire to and assume roles very similar to those of genetic counselors. “In a few short years genetic nursing practice has been transformed from a nearly hidden specialty to a recognized specialty practice with formal recognition, publication of scope and standards of practice, and most recently the availability of credentialing for genetic nurses” (Greco and Mahon, 2003). The International Society of Nurses in Genetics (ISONG) defines genetic nurses as licensed professional nurses with special education and training in genetics who perform risk assessment, analyze the genetic contribution to disease risk and discuss the impact of risk on health care management for individuals and families (International Society of Nurses in Genetics, 2006). Similar to what is occurring in genetic counseling, Greco and Anderson (2002) have proposed a transdisciplinary model that is in contrast to the traditional hierarchical model of genetic services, in which medical geneticists oversee genetic counselors and nurses. In the transdisciplinary model, genetic nurses work in parallel with genetic counselors as core team members of genetic service programs for multifactorial (complex) disorders such as cancer, diabetes, Alzheimer disease, cardiac diseases, with medical geneticists and medical and non-medical specialists serving as consultative team members. In addition to performing what are seen as the usual responsibilities of genetic counselors and genetic nurses (which include taking family history, constructing pedigree, risk assessment/calculation, and test results disclosure), advanced practice nurses and clinical nurses certified in genetics are also seen as potentially being able, by virtue of the terms of their licenses, to perform or participate in certain functions, such as ordering tests and prescribing medication, that are usually assigned to medical geneticists. Training programs in genetic nursing have been established at the masters degree level for the training of genetics clinical nurses and advanced practice genetic nurses, sometimes in association with training in other nursing specialties. Essential competencies and curricular guidelines have been established (American Nursing Association, 2001). The Genetic Nursing Credentialing Commission carries out certification, and the requirements for certification encompass both clinical experience and a modest amount (45–50 hours over 3–5 years) of academic or continuing education courses (Genetic Nursing
■
409
Credentialing Commission, 2006a,b). To date, the number of nurses actually certified by the Commission is very small, only 39 as of 2005 (Ormond and Faucett, 2005). However, there are certainly many more who are currently involved in genetics without being certified, and the pool of nurses who could potentially be trained in genetics is quite large. I have described genetic nursing in some detail because it seems likely that nurses, by virtue of the major roles that they already play in specialty practices, will be in a position to assume a much greater role in genomic medicine than they have to this point. They are already involved in cancer risk assessment, and other forms of risk assessment and especially pharmacogenetic profiling will undoubtedly afford attractive opportunities. It has also been suggested, in a British study of cancer risk counseling, that genetic nurses may prove to be more cost effective than clinical geneticists as providers of genetic risk assessment (Torrance et al., 2006). Therefore, it would seem reasonable to expand training opportunities in genetics for nurses and to enhance the content of such training in the area of genomic medicine. Pharmacists In the discussion above, all forms of genetic testing and risk assessment have been lumped together. However, one form of testing and risk assessment that might conceivably be implemented more rapidly than that for common disease susceptibility is pharmacogenetic testing to look for variants that might make a person more or less sensitive to the therapeutic and toxic effects of drugs. There has certainly been enormous press on the subject, and even some of the strongest critics of susceptibility testing have seen promise in this approach (Holtzman and Marteau, 2000). Nevertheless, even if inheritance does influence the effect of a drug, the relatively simple, one-to-one relationship observed for of some drugs will not always be obvious (Weinshilboum, 2003). Given this complexity, multiple genes that simultaneously affect the metabolism and effectiveness of individual drugs will have to be looked at simultaneously (Johnson, 2003), and other factors beyond the genetic ones that also have an influence will have to be taken into account (Food and Drug Administration, 2003, 2006; Nuffield Council on Bioethics, 2003). Therefore, pharmacogenetic (pharmacogenomic) testing may turn out to be more complex than previously thought, and Malinowski (2004) has suggested that primary care physicians and nurses will not be in a position to shoulder the burden alone. His solution to this problem is for pharmacists and non-physician clinicians to assume an expanded role in the health care process. It is certainly not clear how this suggested division of labor will actually occur or that pharmacists should play a role in the decision-making process. Although I personally believe that such decisions should remain within the province of physicians, primary care or otherwise, the American Association of Colleges of Pharmacy suggests otherwise: “The potential is enormous for pharmacogenomics to yield a powerful set of molecular diagnostics that will become routine tools by which pharmacists and physicians select the proper medications and doses for each individual patient … The results of these tests will not be simply a list of
410
CHAPTER 34
■
Educational Strategies in Genomic Medicine
gene SNPs, but rather a report formatted and interpreted according to the patient’s diagnosis and treatment options … to which the patient’s future physicians and pharmacists would be granted access as needed to make treatment decisions” (Johnson, et al., 2004). However it eventually plays out, it is clear that pharmacists, as well as medical geneticists and genetic counselors, will all have to become knowledgeable about pharmacogenetic testing.To this end, schools of pharmacy are beginning to introduce courses in genetics and genomics, and pharmacogenetics and bioinformatics, but it is likely, even if the pharmacist’s role is only explanatory, that much more extensive exposure will be required than is currently being provided (Latif and McKay, 2005). Laboratory Geneticists Although the principal focus of this chapter is on those who work directly with patients, there is, of course, another side to genetic testing and risk assessment – the actual performance of the tests in laboratories. For all practical purposes, this part of the process is invisible to the patient or client, but very often not to the persons, whether physicians, counselors, or nurses, who are involved in ordering the tests. Therefore, the directors of such laboratories need to be well trained not only in the technical aspects of the tests they are performing, but more broadly is how these tests are being used and interpreted in the risk assessment process. Laboratory geneticists in the United States are certified by either the American Board of Medical Genetics (as clinical molecular geneticists) or the American Board of Pathology (as molecular genetic pathologists). In general, the training required for these certifications is quite germane to genomic medicine, but again it will need to be focused and enhanced as the number and volume of tests increases. Although less than 500 individuals were certified as of 2007 (American Board of Medical Genetics, 2008), a single certified laboratory geneticist can handle testing for a very large number of people, so his/her impact is greater than might first appear. Genetic testing is carried out in both academic and commercial settings, with the balance rapidly shifting to the latter. Furthermore, many directors of testing laboratories are neither TABLE 34.4
board-certified medical geneticists nor molecular pathologists, although many are certified in pathology, which includes both anatomic and clinical (laboratory) pathology. It is in the latter area that exposure to molecular genetics occurs and includes basic molecular techniques and methods, human genetic principles, types of mutations and their detection, and gene structure and function and the application of these techniques and principles to inherited, neoplastic, and infectious diseases and to histocompatibility (American Board of Pathology, 2005). Given the large numbers of areas in which training is required, the exposure to molecular genetics cannot be very intensive. Several other boards, none of which are specifically oriented to genetics and genomics (except for the American Board of Medical Genetics, which is included in the list) are also authorized under the Clinical Laboratory Improvement Amendments (CLIA) to certify laboratory directors of so-called high complexity testing (Centers for Medicare and Medicaid Services, 2005). The import of all of this is that the education of laboratory directors required for their participation in genomic medicine will have to be enhanced and carried on over many fronts. Other Health Professionals In addition to physicians, nurses, genetic counselors, pharmacists, and laboratory geneticists, there are many other health professionals who may be involved directly or indirectly in genomic medicine. These range from dentists and optometrists to psychologists, physician assistants, and physical and occupational therapists. The National Coalition for Health Education in Genetics (NCHPEG), which includes health professional societies, certifying boards, consumer groups, foundations, commercial entities, government agencies, and hospital and health care systems among its membership (NCHPEG, 2006), has prepared a list of core competencies in genetics that are considered essential for all health professionals (NCHPEG, 2005). In addition to several knowledge items, the skills listed in Table 34.4 have been identified. It is clear from this list that health professionals who possess only the skills enumerated would not be in a position to perform genetic risk assessments, but they would certainly be in
Skills considered essential for all health professionals (NCHPEG, 2005)
The ability to: ●
gather genetic family history information, including an appropriate multigenerational family history
●
identify clients who would benefit from genetic services
●
explain basic concepts of probability and disease susceptibility, and the influence of genetic factors in maintenance of health and development of disease
●
seek assistance from and refer to appropriate genetics experts and peer support resources
●
obtain credible, current information about genetics, for self, clients, and colleagues
●
use effectively new information technologies to obtain current information about genetics
●
educate others about client-focused policy issues
●
participate in professional and public education about genetics
References
a position to be knowledgeable about what is involved and who should be referred.
CONCLUSION In presenting the competencies just listed, the NCHPEG committee that prepared them made some observations that bear directly on the issue of exactly who will be doing what that I alluded to earlier in this chapter: Some in the genetics community were concerned that broad adoption of the competencies would result in the transfer of all genetic services to providers not formally trained in genetics, thereby obviating medical genetics and genetic counseling as specialties. Others in the genetics community viewed the competencies as unrealistic and unattainable by non-geneticists. Some non-geneticists concurred and wondered whether the intent of the competencies was to turn all health-care providers into genetic specialists. Those questions became part of the larger debate about the future of genetically based health care, one more variable in the complex mix of factors influencing a health-care system in rapid flux. There is, of course, no final resolution at this time to the questions of who will provide genetic services, in what settings, and under what financial structures. Indeed, there is as yet no clear answer to the question of how genetic services will be defined in the future, and that answer is unlikely to emerge any time soon. It is crucial, however, that all health professionals incorporate genetic and genomic knowledge so they are well informed about the ways in which that information can enhance patient outcomes (my emphasis) (NCHPEG, 2005).
The uncertainties expressed in this statement with regard to how genetic services, which include applied genomic medicine, will be incorporated into the health care system make it difficult to develop any definitive educational strategies for preparing for the future. Nevertheless, it seems reasonable to proceed on the premise that many health care professionals are likely to be involved in genomic medicine, with physicians in primary care and in those specialties that are principally involved in the prevention and treatment of common diseases will being central
■
411
to the process. However, it is also likely that many of the nonmedical aspects of genetic risk assessment may and perhaps should be delegated to genetically trained counselors and nurses, Although their numbers are presently quite small, training programs for these two groups could be expanded relatively quickly. When pharmacogenetic testing is involved, pharmacists, whose numbers are considerably greater, may be in a position to play an important role. What are less certain are the roles beyond the consultative and educational that clinical medical geneticists will play. It is conceivable that they could become directly involved in risk assessment, but again their numbers are small and are unlikely to increase very rapidly. In this chapter, I have discussed the education of the several professional groups just mentioned, and the conclusion is the same for each – more intensive training in genetics and genomics beginning earlier in their careers will be required. The fact that all health care professionals are themselves part of the public at large, for which genetic literacy will be essential, means that this training must begin as early as elementary and high school and continue into their professional training and practice. It is not clear at this time who will actually take responsibility for formulating the curricula in genetics and genomics for the many groups that have been considered in this chapter. Several different professional organizations have already been alluded to along the way, and it is likely that they and many others like them will have a part to play. However, as a practical matter, it would seem that the promulgation of curricula will ultimately have to be done by the organizations that are most directly involved with particular target audiences. Lastly, the discussion of training in genetics and genomics for the public and health care professionals has been cast principally in terms of science and technologies. However, it is likely that many of these individuals will be drawn into discussions of the policy issues that will certainly develop and revolve around genomic medicine. Therefore, consideration of the social, ethical, and legal issues that will form the basis for the making of policy should be included in the educational curricula at all levels. For health professionals, this would best done during the time of their formal professional training. For the public, high school and university would be most appropriate.
REFERENCES Accreditation Council for Graduate Medical Education (2007). Program requirements for graduate medical education in Medical Genetics. http://www.acgme.org/acWebsite/RRC_130/130_prIndex.asp American Academy of Family Physicians (2006a). AAFP Policies: Primary Care: Definitions. http://www.aafp.org/online/en/home/policy/ policies/p/primarycare.html American Academy of Family Physicians (2006b). AAFP Annual Clinical Focus: 2005 Genomics. http://www.aafp.org/online/en/ home/clinical/acf/genomics.html American Board of Genetic Counseling (2006). Specific requirements for accreditation in genetic counseling. http://www.abgc. iamonline.com/view.asp?x1642&mid110
American Board of Genetic Counseling (2008). Certification. http:// www.abgc.net/english/view.asp?x1418 American Board of Medical Genetics (2008). Numbers of certified specialists in genetics. http://www.abmg.org/pages/ resources_certspecial.shtml American Board of Pathology (2005). Description of examinationsanatomic pathology and clinical pathology. http://www.abpath. org/CandBkltDescrAPCP.htm American Nurses Association (2006). Essential nursing competencies and curriculum guidelines for genetics and genomics, established by consensus panel, September 21–22, 2005. http://www.isong. org/resources/genetics_competencies_092206.pdf
412
CHAPTER 34
■
Educational Strategies in Genomic Medicine
American Society of Clinical Oncology (1997). Resource document for curriculum development in cancer genetics education. J Clin Oncol 15, 2157–2169. American Society of Clinical Oncology (2003). American Society of Clinical Oncology Policy Statement Update: Genetic testing for cancer susceptibility. J Clin Oncol 21, 2397–2406. Andrews, L.B., Fullarton, J.E., Holtzman, N.A. and Motulsky, A.G. (1994). Assessing Genetic Risks. Implications for Health and Social Policy. National Academy Press,Washington, D.C. Association of Professors of Human and Medical Genetics and the American Society of Human Genetics (2001). Medical school core curriculum in genetics. http://www.ashg.org/pages/pubs_curriculum.shtml Burke, W., Acheson, L., Botkin, J., Bridges, K., Davis, A., Evans, J., Frias, J., Hanson, J., Kahn, N., Kahn, R. et al. (2002). Genetics in primary care: A USA faculty development initiative. Comm Genet 5, 138–146. Burke, W., Khoury, M.J., Stewart, A. and Zimmern, R.I. (2006). The path from genome-based research to population health: Development of an international public health genomics network. Genet Med 8, 451–458. Caskey, C.T. (2001). Foreword. The future of medical genetics. In Genetics in the Clinic. Clinical, Ethical, and Social Implications for Primary Care (M.B. Mahowald, V.A. McKusick, A.S. Scheurle and T.J. Aspinwall, eds), Mosby, St. Louis, pp. ix–xi. Centers for Medicare and Medicaid Services (2005). Certification boards for clinical consultants and laboratory directors of high complexity testing. http://www.cms.hhs.gov/CLIA/16_Certification_Boards_ Clinical_Consultants_&_Laboratory_Directors.asp#TopOfPage Department of Health (2003). Our Inheritance, our Future. Realizing the Potential of Genetics in the NHS, London. Epstein, C.J. (2004). Genetic testing: Hope or hype? Am J Med Genet 6, 165–172. Epstein, C.J. (2005). Medical geneticists in the 21st century. Genet Med 7, 375–379. Epstein, C.J. (2006). Medical genetics in the genomic medicine of the 21st century. Am J Hum Genet 79, 434–438. Federal Trade Commission (2006). At-home genetic tests: A healthy dose of skepticism may be the best prescription. http://www.ftc. gov/bcp/edu/pubs/consumer/health/hea02.htm Food and Drug Administration (2003). Guidance for industry. Pharmacogenetic data submissions. Draft guidance. November, 2003. http//www.fda.gov/cder/guidance/index.htm Food and Drug Administration (2006). Draft guidance for industry, clinical laboratories, and FDA Staff on in vitro diagnostic multivariate index assays; Availability. http://www.fda.gov/OHRMS/DOCKETS/ 9 8 f r / c h 0 6 4 1 . p d f w w w. f d a . g ov / O H R M S / D O C K E T S / 98fr/ch0641.pdf Genetic Nursing Credentialing Commission (2006a). Advanced practice nurse in genetics. http://www.geneticnurse.org/advancedpracticeapng.html Genetic Nursing Credentialing Commission (2006b). Genetics clinical nurse. http://www.geneticnurse.org/geneticsnursegcn.html Genetics Education MSOP Expert Panel (2004). Contemporary Issues in Medicine: Genetics Education. Association of American Medical Colleges Medical School Objectives Project,Washington, D.C. Ginsburg, G.S., Donahue, M.P. and Newby, L.K. (2005). Prospects for personalized cardiovascular medicine. The impact of genomics. J Am Coll Cardiol 46, 1615–1627. Greco, K. and Anderson, G. (2002). Redressing policy in cancer genetics: Moving toward transdisciplinary teams. Policy Polit Nurs Pract 3, 129–139.
Greco, K.E. and Mahon, S.M. (2003). Genetics nursing practice enters a new era with credentialing. The Internet JAdv Nurs Pract 5(2). http://www.ispub.com/ostia/index.php?xmlFilePathjournals/ ijanp/vol5n2/genetics.xml Grigg, W., Lausko, M. and Brockway, D. (2006). The Nation’s Report Card: Science 2005. (NCES-2006-466). U.S. Department of Education, National Center for Education Statistics. U.S. Government Printing Office,Washington, D.C. Grosse, S.D. and Khoury, M.J. (2006). What is the clinical utility of genetic testing?. Genet Med 8, 448–450. Guttmacher, A.E., Jenkins, J. and Uhlmann, W.R. (2001). Genomic medicine: Who will practice it? A call to open arms. Am J Med Genet (Semin Med Genet) 106, 216–222. Haga, S.B. (2006). Teaching resources for genetics. Nat Rev Genet 7, 223–229. Harvard-MIT Division of Health Sciences & Technology (2004). Genomic medicine. HST 512/513 https://www.hstdev.mit.edu/ servlet/ControllerServlet?handlerPublicHandler&actionview Course&courseIDHST512&term2004SP Health Sciences Library: Genetics Curriculum Resource (2006). http:// www.library.nymc.edu/informatics/genetics.cfm Holtzman, N.A. and Marteau, T.M. (2000). Will genetics revolutionize medicine?. N Engl J Med 343, 141–144. Hott,A.M.,Huether, C.A.,McInerney, J.D.,Christianson, C.,Fowler, R., Bender, H., Jenkins, J., Wysocki, A., Markle, F.G. and Karp, R. (2002). Genetics content in introductory biology courses for non-science majors: Theory and practice. BioScience 52, 1024–1035. International Society of Nurses in Genetics (2006). What is a genetic nurse? www.isong.org/resources/what_is.pdf Janssens, A.C.J.W., Aulchenko, Y.S., Elefante, S., Borboom, G.J.J.M., Steyerberg, E.W. and van Duijn, C.M. (2006). Predictive testing for complex diseases using multiple genes: Act or fiction?. Genet Med 8, 395–400. Johnson, J.A. (2003). Pharmacogenetics: Potential for individualized drug therapy through genetics. Trends Genet 19, 660–668. Johnson, J.A., Bootman, J.L., Evans, W.E., Hudson, R.A., Knoell, D., Simmons, L., Straubing, R.M. and Meyer, S.M. (2004). Pharmacogenomics: A scientific revolution in pharmaceutical sciences and pharmacy practice. Final report of the 2001/02 Academic Affairs Committee, American Association of Colleges of Pharmacy. http://www.aacp.org/site/page.asp?VID1&CID1039&DID 6100&TrackIDwww.google.com/search?q Jones, S. and Collins, F. (2003). Genomics in medicine: Hype or real promise? Interview by Ed Rabinowitz. Healthplan 44, 20–24. Khoury, M.J. (2003). Genetics and genomics in practice: The continuum from genetic disease to genomic information in health and disease. Genet Med 5, 261–268. Korf , B.R. (2002a). Genetics in medical practice. Genet Med 4(6, Supplement), 10S–14S. Korf , B.R. (2002b). Integration of genetics into clinical teaching in medical school education. Genet Med 4(6, Supplement), 33S–38S. Korf , B.R. (2005). Genetics training in the genomic era. Curr Opin Pediatr 17, 747–750. Latif, D.A. and McKay, A.B. (2005). Pharmacogenetics and pharmacogenomics instruction in colleges and schools of pharmacy in the United States. Am J Pharm Educat 69, 152–156. Malinowski, M.J. (2004). Law, policy, and marketing implications of genetic profiling in drug development. http://www.hofstra.edu/ academics/law/law_faculty_workshop_archives.cfm
References
National Institutes of Health (2004). Revolutionary genome sequencing technologies – the $1000 genome. http://www.grants.nih. gov/grants/guide/rfa-files/RFA-HG-05-004.htmlhttp://grants. nih.gov/grants/guide/rfa-files/RFA-HG-05-004.html NCHPEG (2005). Core competencies in genetics essential for all health-care professionals. http://www.nchpeg.org/content.asp? dbsectionbasic&dbid1 NCHPEG (2008). NCHPEG member listing. http://www.nchpeg. org/core/Corecomps2005.pdf NHGRI Policy Roundtable (2005). The future of genomic medicine: Policy implications for research and medicine. http://www. genome.gov/17516574 Nuffield Council on Bioethics (2003). Pharmacogenetics: Ethical issues. London: 2003. Ormond, K. and Faucett, A. (2005). Working group report. http:// www.nsgc.org/client_files/news/SACGHS_Feb05.pdf Royal Society (2005). Personalised medicines: Hopes and realities. http: www.royalsoc.ac.uk/displaypagedoc.asp?id15874 Scheuner, M.T. and Rotter, J.I. (2006). Quantifying the health benefits of genetic tests: A clinical perspective. Genet Med 8, 141–142. Scheuner, M.T., Wang, S.J., Raffel, L.J., Larabell, S.K. and Rotter, J.I. (1997). Family history: A comprehensive genetic risk assessment method for the chronic conditions of adulthood. Am J Med Genet 71, 315–324. Snyderman, R. and Williams, R.S. (2003). Prospective medicine: The next health care transformation. Acad Med 78, 1079–1084. SureGene (2006). AssureGene test for schizophrenia susceptibility. http://suregene.net/AssureGeneTest.aspx
■
413
Torrance, N., Mollison, J., Wordsworth, S., Gray, J., Miedzybrodzka, Z., Haites, N., Grant, A., Campbell, M., Watson, M.S., Clarke, A. and Wilson, B. (2006). Genetic nurse counsellors can be an acceptable and cost-effective alternative to clinical geneticists for breast cancer risk genetic counselling. Evidence from two parallel randomised controlled equivalence trials. Br J Cancer 95, 435–444. Tversky, A. and Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science 211, 453–458. United States Department of Health and Human Services (2006). U.S. Surgeon General’s family history initiative. http://www.hhs. gov/familyhistory/ Weinshilboum, R. (2003). Inheritance and drug response. N Engl J Med 348, 529–537. Weston, A.D. and Hood, L. (2004). Systems biology, proteomics, and the future of heath care: Toward predictive, preventative, and personalized medicine. J Proteome Res 3, 179–196. Willard, H.F. (2004). Message from the director. http://www.genome. duke.edu/genomelife/glarchive/issue9/dirmessage Willard, H.F., Angrist, M. and Ginsburg, G.S. (2005). Genomic medicine: Genetic variation and its impact on the future of health care. Phil Trans Roy Soc B 360, 1543–1550. Wolfberg, A.J. (2006). Genes on the web – direct-to-consumer marketing of genetic testing. N Engl J Med 355, 543–545. X Prize Foundation (2006). Archon Genomics X Prize http://genomics.xprize.org/
.
CHAPTER
35 Federal Regulation of Genomic Medicine Janet Woodcock
INTRODUCTION Regulation of the medical uses of genomic technologies is an evolving and controversial field. In the ideal case, medical product regulation strikes an appropriate balance between protecting patients and fostering innovation. Such balance can be difficult to achieve during the dynamic evolution of a new area of technology. Currently, the fields of genomic testing and pharmacogenomics are experiencing very active growth and corresponding regulatory and policy interest, and there is ongoing debate about the appropriate level of regulation of these technologies (Hudson, 2006; Javitt and Hudson, 2006). This chapter provides an introduction to both the regulatory framework and the ongoing policy challenges for genomic medicine. Commercial introduction of genetic tests began in the late 1970s; however, the pace of introduction was slow at first and has rapidly accelerated over the last decade. It is estimated that over 1300 genetic tests are currently available in the United States (Gene Tests, www.genetests.org). Most of these are diagnostic tests for genetically based disorders. Recently, new types of in vitro diagnostics (IVDs) – for example, assays evaluating gene expression – are being introduced. Based on the current amount of activity in the field, an explosion of new assays can be anticipated over the next several years. The landscape of genetic test regulation is somewhat confusing. The vast majority of available genetic tests are
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 414
“laboratory-developed tests.” Such assays are not FDA approved; rather, they are developed and conducted within a specific laboratory and are offered as a service by the laboratory. In contrast, all marketed test kits go through an FDA regulatory process of some type. Laboratories that develop an in-house test may submit an application for marketing clearance to the FDA; however, few have chosen to do so. This framework has resulted in a perception of an “uneven playing field” for the two assay categories. All laboratories reporting out human results require Clinical Laboratory Improvement Amendments (CLIA) certification. The CLIA program is administered by the Centers for Medicare and Medicaid Services (CMS). The respective roles of FDA and CMS have not been clear to all, especially newcomers, in the field and have resulted in considerable confusion. Recently FDA issued several “Guidance” documents intended to specifically clarify FDA regulation of newer diagnostic tests, including genomic tests. These will be discussed in more detail below. Researchers interested in developing a specific assay for use in clinical medicine will need a basic understanding of the various regulatory regimes, as described further in this section. Exploratory assay development using human samples – for example, tissues or body fluids – can usually be conducted under the general human subject protection and privacy provisions implemented by academic institutions. Prior to undertaking analytical validation or clinical studies, however, it would be prudent for a developer to determine the potential regulatory
Introduction
status of the test, and the types of studies and regulatory clearances needed to bring the test to market successfully. Further evolution of the current regulatory framework for genomic tests is likely if the new discipline of genomic medicine is as successful as is currently anticipated. In contrast to genetic diagnostic testing, the field of pharmacogenomics is in fairly early development. Genomic technologies have been used extensively in drug discovery for over a decade, but application to drug development is a more recent phenomenon. Pharmacogenetics can be defined as the study of the genetic contributions to drug responses (see Chapter 27). Drug metabolizing enzyme polymorphisms are the best understood examples of pharmacogenetics because correlations between phenotypes and drug concentrations have been well worked out for a number of polymorphisms. Despite decades of study, however, metabolizing enzyme polymorphisms are rarely taken into account in drug development or clinical practice, in part because convenient assays have not been available. Recently, genetic tests for common drug metabolizing enzyme polymorphisms have been introduced and several have been approved by the FDA; however, as of this writing, no drug label contains specific requirements for dose adjustment based on metabolizing enzyme test results. Routine use of such tests awaits demonstration of added value. Pharmaceutical developers generally seek to eliminate, prior to clinical testing, candidate drugs that are subject to polymorphic metabolism or that may be prone to drug– drug interactions based on metabolic factors. Traditional clinical drug development processes have not included individualized dose adjustments, and there are strong commercial and logistical disincentives to incorporating such individualization. The first pharmacogenetically directed, dose-adjusted drug development program will likely involve: ●
●
●
a promising molecule that developers are unable to engineer to avoid an undesirable metabolic pathway (Hudson, 2006); an indication for a very serious or life-threatening illness (Javitt and Hudson, 2006); and an expected narrow therapeutic index (e.g., cancer therapeutics).
Much of the current scientific interest in drug metabolism involves marketed drugs, many of which are metabolized by polymorphic enzymes. Clinical practitioners have long been accustomed to performing empirical dose adjustments based on clinical parameters, for example, INR for warfarin, leukocyte counts for 6-mercaptopurine, or side effects for many drugs. There is considerable skepticism within the clinical and reimbursement communities about the value of up-front genetic testing and dose adjustment. Resolution of these issues will require randomized clinical outcome trials that demonstrate added safety or effectiveness, or improved patient convenience/adherence and/or cost-effectiveness as a result of testing and dose adjustment. From a scientific perspective, it is unquestionable that the many-fold variation in drug exposure often introduced by polymorphic metabolism has important clinical
■
415
consequences in both drug development and in patient care, particularly for drugs with narrow therapeutic indices. However, ingrained practices change slowly, and, in any case, there is a need for evidence-based dosing algorithms for practitioners in instances where dose adjustment is found to be important (see for example www.warfarin-dosing.org). Although drug metabolism is the best understood example, the promise of pharmacogenomics for drug and vaccine development and use is much more expansive. Pharmacogenomic tests are being explored for selecting patient subgroups for treatment (based on probability of response, prognostic category, or disease subtype), identifying patients with a high probability of an adverse response to therapy, and for monitoring response. Additionally, the field of toxicogenomics, that evaluates the genomic responses to toxicants, is expected to contribute much more sensitive toxicity assays and greatly enhanced understanding of the mechanisms of drug toxicity. All these applications are in the early stages of development. The most extensive clinical use occurs in the field of antiretroviral therapy, where HIV drug resistance testing is routinely performed by assaying the viral genome, and therapy is directed by the results. Both FDA-approved and laboratory-developed HIV drug resistance assays are commercially available. With respect to the human genome, the most developed examples occur in cancer, where various genomic assays are employed to provide information about tumor aggressiveness and prognosis, or about likelihood of response to a targeted therapy (e.g., imatinib, trastuzumab). NIH, FDA and the pharmaceutical industry are also pursuing efforts to understand the genetic basis of drug-related adverse events. An historical example of a genetically based drug adverse event is drug-induced hemolysis in people with G6PD deficiency (metabolism based); a non-metabolic, recent example is the abacavir hypersensitivity reaction (Rauch et al., 2006). It is expected that some, but not all, drug-related adverse events will have a genetic contribution. Because of the potential for pharmacogenomics to improve drug safety and effectiveness, the FDA has been vigorously involved in efforts to foster this new field. In November 2003, the Agency established the “voluntary genomic data submission process,” a kind of safe harbor where pharmaceutical developers can share genomic experiments and data with the FDA in a non-regulatory context (www.fda.gov/cder/genomics/GDS. htm). Multiple data submissions have been shared with FDA and this process is expected to help ease the transition of pharmacogenomic technologies into mainstream preclinical and clinical drug development. Nevertheless, numerous policy questions related to the use and regulation of pharmacogenomics in drug development remain to be answered. For example, what regulatory standards should apply during co-development of a pharmacogenomic test and drug to be used in combination? What data and studies are needed to qualify genomic biomarkers for regulatory use? “Regulation of Genomic Tests” explores the regulatory framework for the use of pharmacogenomic technologies in drug development and therapy, and discusses various unresolved policy areas.
416
CHAPTER 35
■
Federal Regulation of Genomic Medicine
REGULATION OF GENOMIC TESTS Researchers intending to translate basic genomic discoveries into new diagnostic or prognostic assays will need to have a basic understanding of the regulatory framework governing such tests. FDA regulations define IVDs products as “those reagents, instruments and systems intended for use in the diagnosis of disease or other conditions, including a determination of the state of health, in order to cure, mitigate or prevent disease or its sequelae.” The regulations go on to say: “These products are devices as defined in Section 201(h) of the Federal Food Drug and Cosmetic Act …” (21 CFR 809.3) Therefore reagents, systems, and assays for human use are considered medical devices under the law and are subject to FDA regulation. However, medical device regulation has considerable flexibility, and this has been exercised by FDA in its oversight of IVDs. Additionally, the FDA Modernization Act passed by Congress in 1997 called upon FDA to use an approach to device regulation that is “least burdensome” on manufacturers while still achieving public health objectives, and this charge is taken into account when implementing regulatory standards. Regulation of In Vitro Diagnostics as Medical Devices Prior to embarking on a development program for a genomic test, researchers and developers should strongly consider consulting with the FDA staff about the development plan. Most genomic assays regulated as medical devices will be overseen within the Office of In Vitro Diagnostic Device Evaluation and Safety (OIVD) in the Center for Devices and Radiological Health (CDRH) at FDA. CDRH offers assistance to researchers, developers, and device manufacturers, who can request a “preIDE” meeting with OIVD to discuss the proposed development plan. Such consultations can save time and prevent wasted efforts. Developers can submit their planned protocols for evaluating the performance of the IVD and reach agreement with FDA staff on the extent and nature of the studies. Contacts can be reached through the CDRH web page (www.fda.gov/cdrh/index.html). FDA has also recently published a relevant draft Guidance entitled “Pharmacogenetic Tests and Genetic Tests for Heritable Markers” that contains advice on product development (www. fda.gov/cdrh/oivd/guidance/1549.html). Regulatory Classification of In Vitro Diagnostics Medical devices are classified into categories, based on risk, that govern the intensity of their regulation. For IVDs, the risk determination is primarily a function of the intended use of the device, based generally on the claims sought by the manufacturer. Many types of IVDs have been formally classified by the FDA in regulations. Class I devices are considered low risk and are subject to the fewest regulatory controls. These so-called “general” controls include, for example, facility registration, requirements for label information and format, manufacturing procedures, and reporting of adverse events. Many (non-genetic) IVDs have been assigned to Class I and are exempted from submitting applications for marketing to the FDA. Class II devices are moderate
risk and require additional controls. Many genomic tests will be considered Class II IVDs. As a relevant example, drug metabolizing enzyme genotyping systems are Class II. In making this determination, FDA published the document “Class II Special Controls Guidance Document: Drug Metabolizing Enzyme Genotyping System,” in 2005 (www.fda.gov/cdrh/oivd/guidance/1551.html). This document explains what data to submit to FDA in a Premarket Notification Submission, also known as a 510(k) (see below) for a diagnostic of this type. By submitting this information, and also complying with the “general” controls, a manufacturer can meet the requirements for marketing a drug metabolizing enzyme genotyping system. Special controls vary with the category of IVD. In some cases, clinical studies may need to be submitted as part of a 510(k) submission for a Class II IVD. Class III IVDs are considered high risk and often require submission of a Premarket Approval Application (PMA). PMAs usually include data from clinical studies. If a device category has not been classified, it is considered Class III. Applications for Marketing In Vitro Diagnostic Tests The Premarket Notification Submission or 510(k)
When submitting a traditional 510(k) application, a manufacturer intends to demonstrate that its IVD is “substantially equivalent” (the legal terminology) to a previously legally marketed IVD, called the “predicate device.” If a predicate IVD exists, clinical data in the 510(k) for a genomic test would include information on the analytic validation of the test, plus comparisons with the predicate IVD performance, using appropriate clinical samples (Figure 35.1). If no predicate exists, the FDA can, when appropriate, use a process know as a de novo 510(k) to establish the special controls needed for a Class II IVD, as discussed for the drug metabolizing enzyme genotyping test systems mentioned above. In these instances, the medical literature or submitted data can be used by FDA to help determine the needed extent of special controls. The de novo 510(k) may contain clinical data consisting of protocol-based testing of clinical samples, may sometimes also include data gathered in prospective clinical trials, and in some cases may include data derived from the clinical literature and/ or information found in recognized clinical practice standards. In February 2007 FDA approved Agendia’s MammaPrint® as a de novo 510(k). This test is intended to evaluate the likelihood of breast cancer recurrence within 5–10 years in patients with Stage I or Stage II disease. It is the first FDA-cleared molecular test using genetic profiling to predict breast cancer prognosis and is indicated as a prognostic marker only. Total FDA review time for this application was 30 days; the entire process (which necessitated the firm submitting a petition for down-classification) took less than 180 days. The special controls guidance document for these types of assays, “Class II Special Controls Expression Profiling Test System for Breast Cancer Prognosis,” has been published (www.fda.gov/cdrh/oivd/guidance/1627.pdf). The clinical data for this application was based on analyses of stored samples from various studies (www.fda. gov/cdrh/reviews/K062694.pdf).
Regulation of Genomic Tests
417
Application review
IDE Submission/review (If Required)
Investigational phase pre-IDE or IDE meeting as Appropriate
■
PMA or 510(k) application
510(k) – 90 Day Review PMA – 180 Day Review
Figure 35.1 In vitro diagnostic development process. The requirement for an Investigational Device Exemption (IDE) filing is based on risk: most clinical studies of IVDs do not need to be conducted under an IDE. Sponsors may request a meeting with FDA at any time during the development process to obtain regulatory guidance on the development program or on how to submit an application: request a “pre-IDE” meeting if no IDE exists, or an “IDE” meeting if an IDE has been filed. The timeframe for FDA review depends on whether a 510(k) or PMA has been filed.
FDA has published extensive guidance on submission of 510(k)s for IVDs, including the documents “Format for Traditional and Abbreviated 510(k)s” (www.fda.gov/cdrh/ode/ guidance/1567.pdf) and “Points to Consider for Collection of Data in Support of in vitro Device Submissions for 510(k) Clearance” (www.fda.gov/cdrh/ode/95.pdf) FDA has also recognized dozens of voluntary standards (most developed by the Clinical Laboratory Standards Institute or CLSI) for use in establishing test performance. Published guidances and FDArecognized standards can be found on the CDRH web page (www.fda.gov/cdrh/index.html). The Premarket Approval Application
The claims attached to certain genomic tests may cause them to be classified into Class III (high risk). An example of such claims could include prognostic information linked to therapeutic decision-making (e.g., more extensive radiation or chemotherapy for cancer). PMA submissions contain clinical data that substantiate these claims, usually derived from clinical trials. Developers often meet with the FDA staff to reach agreement on the design of these trials prior to their initiation. Investigational In Vitro Diagnostics Investigational device exemptions (IDEs) must be submitted to the FDA prior to beginning clinical studies of certain devices. IDEs are rarely needed for IVDs. In clinical studies where the results of the test will not be used in patient management, an IDE is not needed; however, the usual requirements for informed consent and IRB approval remain, where applicable. In those instances where the test results will determine management (e.g., patients are randomized to further interventions based on test results), an IDE should be submitted. OIVD staff may be consulted on the need for an IDE filing. If investigational IVDs are shipped for research testing, for example, analytical validation, they should be labeled “For
Research Use Only. Not for use in diagnostic procedures” (21 CFR 809.10 (c)(2)(i)). If they are shipped for use in comparative testing or as part of a clinical trial, they should be labeled “For Investigational Use Only. The performance characteristics of this product have not been established” (21 CFR 809.10 (c)(2)(ii)). FDA recently issued a document entitled “Guidance on Informed consent for In Vitro Diagnostic Device Studies Using Leftover Human Specimens that are Not Individually Identifiable” that offers advice on the use of de-identified human specimens for testing (www.fda.gov/cdrh/oivd/1588.pdf). Laboratory-Developed Tests As stated above, the vast majority of currently marketed genetic tests are laboratory-developed tests that have not undergone the FDA review process. These tests are either fully developed by the laboratory or utilize a purchased “analyte-specific reagent” (ASR, see below) that is then configured into an assay by the laboratory, which carries out the test as a service. Such laboratory developed tests are not shipped for use outside the originating site. These tests are considered medical devices under the Food, Drug and Cosmetic Act (and the laboratories are considered device manufacturers), but FDA has not required marketing applications (using what is legally termed “enforcement discretion”), considering that the controls provided by the ASR regulations and the certifications for high complexity laboratories under CLIA (see below) to be sufficient. In September 2006 FDA issued a draft guidance document entitled “In Vitro Diagnostic Multivariate Index Assays” explaining that certain complex test systems, including certain genomic test systems, are Class II or Class III medical devices requiring 510(k) submissions or PMAs, respectively (www.fda. gov/cdrh/oivd/1610.pdf). The document defines the difference between in vitro diagnostic multivariate index assays (IVDMIAs) and the types of tests generally performed as laboratorydeveloped tests. This document and the related policy issues around laboratory-developed tests are generating considerable
418
CHAPTER 35
■
Federal Regulation of Genomic Medicine
discussion and debate, and the outcome of these debates will have consequences for developers of genomic tests. Analyte-Specific Reagents In 1997, FDA defined a group of reagents known as ASRs, and classified them (with some exceptions that are not germane to genomics) as Class I medical devices. This meant that ASRs were subject to “general controls” such as manufacturing and labeling requirements, but did not have to have applications cleared or approved by FDA to be put on the market. FDA accomplished this by publishing three regulations that defined and classified ASRs (21 CFR 864.4020), imposed restrictions on their sale, distribution and use (21 CFR 809.30) and established requirements for ASR labels (21 CFR 809.10(e)). In the regulations, ASRs were defined as “antibodies, both polyclonal and monoclonal, specific receptor proteins, ligands, nucleic acid sequences, and similar reagents which, through specific binding or chemical reactions with substances in a specimen, are intended for use in a diagnostic application for identification and quantification of an individual chemical substance or ligand in biological specimens” (21 CFR 864.4020). Thus, an ASR is a building block for an assay, but is not a test system. Subsequent to publication of these regulations, FDA observed a broadening of the implementation of ASRs beyond the stated parameters. Therefore, in September 2006, FDA issued a draft guidance entitled “Commercially Distributed Analyte Specific Reagents (ASRs): Frequently Asked Questions” (www.fda.gov/cdrh/oivd/guidance/1590.html). An important point in this document, for the purposes of this discussion, was that FDA did not consider “bundled” moieties (e.g., sets of primers), or extensively processed (e.g., arrayed on beads) reagents, or microarrays, to be ASRs. These distinctions are important to the genomic testing community and have elicited extensive comment. Undoubtedly the boundary between ASRs and test systems will continue to be clarified in the coming years. Clinical Laboratory Improvement Amendments Congress passed the CLIA in 1998. This law established quality standards for all laboratory testing to ensure the accuracy, reliability and timeliness of patient test results regardless of where the test is performed. The CLIA program is administered by the CMS (www.cms.hhs.gov/CLIA). The ASR regulations discussed above stipulated that the only clinical laboratories to which ASRs could be sold were those qualified under CLIA to perform high complexity testing (or alternatively, were regulated under the Veteran’s Health Administration Directive 1106) (21CFR 809.30 (a)(2)). Currently, most genetic testing is performed by such laboratories. The oversight under CLIA relates to the quality of performance of laboratory testing. It does not extend to evaluation of the clinical utility of a given assay. There is ongoing controversy about the degree of oversight of genetic testing under CLIA (Hudson et al., 2006); specifically, whether there should be a specific genetic testing specialty area that would incorporate proficiency testing for genetic tests. To summarize the current Federal regulation of genomic testing: most genetic tests are on the market as laboratory-developed
tests. Such laboratories are subject to quality standards under CLIA. This is possible outside of FDA marketing clearance as a result of the ASR regulations and enforcement discretion on the part of FDA. FDA has approved a number of genetic tests, including tests for drug metabolizing enzyme polymorphisms, generally based on information available in the scientific literature. Recently, FDA issued draft guidances pointing out, among other things, that reagents such as micoarrays and sets of primers are not ASRs, and that IVDMIAs are not among the types of assays considered to be appropriate for laboratory-developed tests. These guidances have resulted in ongoing discussion and controversy, and the policy issues they raise have yet to be definitively resolved.
PHARMACOGENOMICS IN DRUG DEVELOPMENT AND CLINICAL MEDICINE: THE ROLE OF REGULATION Researchers interested in translating genomic discoveries into new approaches to drug therapy will need to understand the challenges involved in incorporating these new scientific approaches into traditional drug development and use patterns. While genomic technologies are used extensively and routinely during drug discovery and in early drug development phases (e.g., lead optimization) routine utilization of genomics drops precipitously as candidates enter the animal safety and clinical testing phases. As a result, few investigational drugs now in the clinical pipeline are being studied in concert with pharmacogenomic assays (although most development programs include collection of human samples for potential genetic testing). Likewise, although extensive information exists about the contribution of drug metabolizing enzyme polymorphisms to variability in drug exposure, as of early 2007, no marketed drug label contains an explicit requirement (as opposed to recommendations) for dose modification based on genetic test results. (See “Genomics at FDA” (www.fda.gov/cder/ genomics/genomic_biomarkers_table.htm) for an up-to-date list of pharmacogenomic tests utilized in drug labels.) These facts, when considered in light of the enormous potential for improvement in drug therapy that pharmacogenomics represents, demonstrate that the field is in its nascent stages. Many barriers to rapid uptake of genomics in drug development and clinical medicine exist, beyond the purely scientific and technical challenges. Drug development is an enormously risky endeavor. Only about one in nine drug candidates entering clinical investigation reach the market, and every year promising candidates encounter catastrophic, highly public late-stage failures. Some compounds that reach the market fail to achieve commercial success. Additionally, drug development programs are extremely expensive. Furthermore, the clinical investigation and marketing application stages of development are subject to extensive regulatory scrutiny worldwide. These factors result in very conservative approaches on the part of pharmaceutical developers. Developing an investigational drug using a relatively untested pharmacogenomic-directed approach is perceived
Pharmacogenomics in Drug Development and Clinical Medicine: The Role of Regulation
as adding additional risk. The fact that drug development is a highly competitive activity also restricts the flow of information among developers, so that those who instigate new approaches may not share the successes or failures with others. Under these circumstances, the regulators may be the only group possessing a broad perspective on the state of the field. Commercial considerations are also important in development. Developers normally pursue the broadest possible market for a new intervention. Many pharmacogenomic approaches (e.g., targeting therapy to a specific-identified population), would limit the indicated population. Complexity of use (e.g., dose adjustments, need for pre-testing) is also commercially undesirable and in fact can limit clinical uptake. For these reasons, along with the natural human optimism about the performance of a new discovery, pharmacogenomic tests have often been considered as a salvage approach when one-size-fits all development fails. Concerns about restrictions on markets are slowly being mitigated by the success of targeted therapies but still limit enthusiasm for pharmacogenomic approaches during drug development (Bernstein, 2006). Once a drug is approved by the FDA, a manufacturer has a limited amount of time to recoup its investment before patent and exclusivity protections expire and generic competition occurs (generic copies are permitted for small molecule drugs but not for therapeutics approved as biological products). Therefore there is limited time to explore refinements of an approved therapy using pharmacogenomics. Once generic competition occurs there is no commercial incentive for further development. The promise of pharmacogenomics and the barriers to its adoption led FDA to aggressively promote investigations of the new technology. Senior FDA staff published a 2002 paper calling for progress in the field and pointing out the Agency’s interest (Lesko and Woodcock, 2002). In May 2002, a workshop between FDA and the pharmaceutical industry was held to explore barriers to information sharing (Lesko et al., 2003). At the workshop, industrial sponsors stated reservations about providing genomic information on compounds in development to the Agency because they feared uninformed questions and delays in product development; however, all agreed on the desirability of information sharing. FDA subsequently established a process for sharing data and results in a non-regulatory context, the “Voluntary Genomic Data Submission” (VGDS) (www.fda.gov/cder/genomics/GDS.htm). Subsequently, over thirty submissions of genomic data involving drugs in development, encompassing topics such as animal studies, preventive indications, identification of responsive subgroups, and genetic links to adverse events, have been discussed with FDA scientists. The goals of VGDS process include (1) encouraging the development and sharing of genomic data linked to therapeutic and preventive indications; (2) helping familiarize FDA scientific staff with the emerging uses of genomic information in drug development; (3) developing data standards and analytical methods suitable for regulatory use; and (4) enabling the transition of genomic studies from exploratory research to regulatory submissions. (See www.fda.gov/cder/genomics/presentations/ Webinar.pdf for a detailed presentation of the VGDS process.)
■
419
Subsequent workshops on drug-diagnostic co-development issues and on genomic data submission standards have been held (www.fda.gov/cder/genomics/biomarkers.htm, Hinman et al., 2006). The increased level of dialog between industry and the FDA has mitigated some of the concerns about regulatory barriers. The European Medicines Agency (EMEA) also instituted pharmacogenomics “Briefing Meetings” in 2003 (The European Agency for the Evaluation of Medicinal Products, 2003). In May 2005 FDA and EMEA held their first joint VGDS meeting, and additional joint meetings are occurring. Discussions of regulatory aspects of pharmacogenomics are now being held among regulators and industry worldwide under the auspices of the International Conference on Harmonization (ICH) of the Technical Requirements for Pharmaceuticals. Translation from the laboratory into clinical development is only one of the translational barriers faced by new technologies. Achieving uptake in clinical practice and obtaining reimbursement are also major challenges: the utility and value of new technologies must be explicitly demonstrated. Ironically, the tight regulation of pharmaceuticals will tend to enhance acceptance of FDA-approved pharmacogenomic test-directed drugs. The regulatory standards for drug approval include the scientific demonstration of effectiveness and safety. When a new drug is approved in conjunction with a pharmacogenetic test, the likelihood of use as recommended in the drug label is very high – in part because of confidence in the approval standard, and in part because promotion (which is often extensive for a new therapy) is limited by law to the label uses. Conversely, “retrofitting” a pharmacogenetic test into established practice patterns can be very difficult. Both the clinical and reimbursement communities have made it very clear that, in the majority of cases, randomized outcome trials will be the standard that will govern acceptance – even for dose adjustment with drug metabolizing enzyme assays. From a regulatory standpoint, the current issues in pharmacogenomics applied to drug development and clinical use include the following: development of an investigational drug incorporating a pharmacogenomic test; use of a pharmacogenomic test to enhance the performance of an approved drug; drug safety pharmacogenomics and toxicogenomics; and qualifying genomic biomarkers for regulatory use. These issues are discussed in more detail in the sections below. Development of an Investigational Drug with a Pharmacogenetic Test to Select Patients for Therapy or for Monitoring Use of diagnostics to discriminate among individuals who present with similar symptoms is routine clinical practice. Genomic tests are conceptually no different; however, these are new types of tests, and they will yield novel insights into the sources of human variability that are currently not utilized in decision-making. Pharmacogenomic tests may be used during drug development to identify patients with different disease subsets (that differ by prognosis, response to therapy, etc.), to identify individuals with disease responsive to an particular intervention (i.e., targeted therapies), to stratify individuals to different dose
420
CHAPTER 35
■
Federal Regulation of Genomic Medicine
regimens, to identify individuals with a high probability of an adverse reaction to treatment, or to monitor treatment response. In all these cases, the use of the diagnostic is intended to enhance drug safety or effectiveness. One major challenge in using pharmacogenomic tests during drug development is that often the assay itself must be developed in parallel. This raises the level of complexity significantly. A full discussion of the issues involved in a drug-diagnostic co-development program is beyond the scope of this chapter. The points below indicate the major areas that must be considered. FDA is planning to issue a guidance document covering this topic in detail (Woodcock, 2005). Regulatory Considerations for Pharmacogenetic-TestDirected Therapy Development The types of data that need to be generated for an investigational drug paired with a pharmacogenomic test depend on the intended label claims for the drug and the test. If use of the drug requires results from the test, demonstration of the safety and effectiveness of the drug-test combination in the definitive trials will be required. If mention of the test in the drug label is strictly informational, a lesser standard of evidence applies. For example, data from retrospective analysis of samples might lead to a suggestion in the label that patients with result x on test y may have a lower probability of response. However, in this scenario, the drug would be indicated for the entire population, absent test y, and would need to be shown effective and safe in the entire population. Most pharmacogenomic information in current drug labels is of this latter type. Exceptions include trastuzumab and imatinib (Bernstein, 2006). Often, developers do not know, at the start of clinical trials, which label scenario they will end up with. In these cases, design of an adaptive program that could support either result may be desirable (see trial designs, below). Drug and Diagnostic Development Considerations Development of the Diagnostic Test
A reliable, reproducible test configuration should be achieved prior to use of the diagnostic in prospective drug trials. Stored samples are frequently used for this purpose, and methods for analytical validation of diagnostics are well worked out. A more significant challenge involves setting cut-off values for the test results. Genomic assays often contain multiple analytes, which are sometimes continuously variable, giving rise to a nearly infinite range of results. Use to direct drug therapy will require achieving unambiguous, often dichotomous (i.e., yes/no) results. These cut-offs often will be constructed via retrospective analysis of clinical samples from the investigational drug trials. This exercise will create an hypothesis: test x will segregate population y into distinct subgroups. This hypothesis can then be verified using additional stored samples or in prospective studies. Depending on the clinical implications of the subgroups, such data might suffice for a stand-alone claim for the diagnostic. However, a claim for use in combination with the drug will require trials demonstrating the clinical utility of the diagnostic as well as drug safety and effectiveness. If the results of such trials
require retrospective analyses that readjust diagnostic cut-offs to achieve significant results, they will be regarded as hypothesis generating rather than confirmatory. Trial Design Considerations for Investigational Drug Combinations
Major design issues arise in this context. These involve ways to integrate the developing information on drug and diagnostic performance when it is unclear whether the diagnostic adds value to the drug therapy (and it is still unclear if the drug has value). At the end of Phase 2 trials, developers may know that the diagnostic can discriminate subgroups but will not have information on how much this impacts performance of the drug. For example, suppose a genomic test appears to identify a patient subset more responsive to the therapy. One approach might be to perform an “enriched” trial, randomizing test-positive patients to investigational drug versus control. This trial answers the question: does the drug work in test-positive patients? It does not yield information on drug performance in test-negative patients (so it does not tell you if the test discriminates anything meaningful), or whether test-positive patients have generally better or worse prognoses. A more elaborate trial might enroll all patients with the condition and then stratify both treatment and control arms by genomic test result. This design can answer many questions but risks answering none definitively because of the multiple comparisons that arise. Simon (2005) and Temple (2005) have proposed various methodological approaches including adaptive designs to address these problems. A number of sequential designs could also be useful. The appropriate design in a given case will be influenced by both mechanistic (how plausible is the test-drug link?) and clinical considerations – for example, how much information on the test-negative subgroup is needed when a claim for effectiveness in the test-positive subgroup is sought? In cases where the mechanistic link is well understood – for example, HIV drug resistance – studying a test-defined subgroup alone is standard practice. In other cases, information on results from the test-negative group may be crucial. Developers should consult with FDA early in such development programs. Dosing Adjustments
Incorporating pharmacogenetically driven dose adjustments into Phase 2 and 3 clinical trials is straightforward. Typically, dose finding is performed in the second phase of drug development. These trials evaluate the dose-response relationships for safety/tolerance and efficacy (which may be defined by a pharmacodynamic marker). “Dose-response” is more properly understood as “concentration response”; and there are excellent pharmacometric methods for analyzing the impacts of variables such as test results. When the effect of a polymorphism is large, the drug developer may then plan to adjust dosing in the efficacy trials for alleles with known effects on concentration. Questions will arise about default dosing for individuals with non-evaluable alleles, as well as how to extrapolate the results to populations with different allele frequencies. For life-threatening diseases, for example some cancer indications, where development programs may incorporate
FDA Efforts to Advance Genomic Product Development
a combined Phase 2–3 development program and may lack pharmacodynamic endpoints, opportunities to understand the impact of variation in drug metabolism are fewer, and evaluation should start as early as possible in the development program. Pharmacogenomic Safety Tests
Drug toxicity is a major issue in development. Beyond toxicity attributable to relative overdosing from variable metabolism, genetically related adverse events can arise from drug target (or related pathway) polymorphisms, immunologic predisposition (e.g., abacavir hypersensitivity), genetically based disease variability, and genetically based organ vulnerabilities (White et al., 2006). Sorting out the mechanistic bases of drug adverse events will be straightforward in some cases and highly complicated in others. The size of most drug development programs will only allow elucidation of the most clear-cut genetic association. Regulatory issues do not complicate such investigations, except in the rare instance where the drug is not safe enough to be approved absent the test. In these cases, sponsors should discuss with FDA the amount of evidence needed to demonstrate that use of the test confers an adequate level of safety. Pharmacogenomic Tests for Approved Drugs It is likely that successful implementation of pharmacogenomic testing would significantly improve therapeutic outcomes of drug therapy. In particular, drug toxicity exacts a severe human and economic toll in the United States, being estimated as the 4th–6th leading cause of death, causing tens of thousands of hospitalizations, and costing billions of dollars (Budnitz et al., 2006; Lazerou et al., 1998). Surveys suggest that drugs with polymorphic metabolism contribute a disproportionate share of harm (Phillips et al., 2001). Nevertheless, as detailed above, there are significant barriers to even initiating the work that needs to be done. Once opportunities are identified out of the basic science, funding for controlled clinical outcome trials needs to be obtained. Pharmaceutical companies may sponsor trials for marketed drugs that are not off-patent. Trials of pharmacogenomic tests for off-patent drugs may require public funding or sponsorship by consortia of consisting of government, health care systems, the pharmaceutical industry, etc. For example, currently a number of groups are considering or initiating trials of pharmacogenetic-directed dosing of warfarin. Issues beyond funding include how to actively involve in consortia diagnostic companies and researchers who have developed tests. FDA has been actively seeking, as an initial step, to incorporate established information on drug metabolizing enzymes into drug labels without stipulating testing and dose adjustment. Examples include 6MP (6-mercaptopurine) (the enzyme TPMT), atomoxitine (Strattera) (CYP2D6) and irinotecan (UGT1A1) (www.fda.gov/cder/genomics/genomic_biomarkers_table.htm) It is possible that in the future, pharmacogeneticdirected dosing could be explored during studies of these agents that are being conducted for other purposes. For a good discussion of the issues involved in developing pharmacogenetic information on a marketed drug (see Ratain, 2006).
■
421
Many rare, serious adverse drug reactions occur across multiple drug classes and may have a genetic association (e.g., Stevens Johnson syndrome). These reactions occur so infrequently, however, that it is likely that international consortia will be required to assemble adequate cases and controls to perform genetic analysis. Other serious drug side effects occur more commonly, within a drug class (e.g., tardive dyskinesia). Public private partnerships between government and industry are being explored as ways to gather cases and carry out the needed analyses. Toxicogenomics When the toxicant is a drug, the areas of toxicogenomics and pharmacogenomics overlap. Many drugs share mechanisms of toxicity with non-pharmaceutical chemicals. Mechanistic understanding of drug toxicity will greatly improve the science of drug safety. The current protocols for preclinical animal toxicology testing during drug development are highly empirical and offer few ways to evaluate the relevance of a given toxic finding to humans. Toxicogenomics may offer methods to discriminate by comparing cellular responses in various animal species and in human cells. In some cases, a toxic finding in an animal species may not be mechanistically relevant in humans. Toxicogenomics is also anticipated to be a source of more sensitive toxicity assays than currently those currently used in traditional animal and human testing. Under the auspices of the C-Path Institute, a nonprofit organization in Tucson, Arizona, a consortium of twelve pharmaceutical companies are pooling, evaluating and validating nontraditional toxicity assays (some of them genomic), on an organ system basis, with the intent of submitting the best performing assays to FDA for regulatory assessment (www. c-path.org/programs/SafePath/tabid/61/Default.aspx). It is expected that toxicogenomic tests will be generally relevant to drug development, rather than linked to a particular drug.
FDA EFFORTS TO ADVANCE GENOMIC PRODUCT DEVELOPMENT Genomic medicine promises to bring new, mechanistic insights into health and disease, to improve drug development, and to move towards more individualized therapeutic interventions that enhance both safety and effectiveness. In recognition of the potential benefits of new sciences, including genomics, in 2004 FDA launched its “Critical Path Initiative,” which is intended to accelerate use of new science in medical product development and regulation (www.fda.gov/oc/initiatives/criticalpath/). This initiative focuses on the use of consortia and public–private partnerships to accomplish the significant translational work needed to fully utilize new technologies in development and in the clinic. In March 2006 the Agency published “The Critical Path Opportunities Report and List,” which identifies biomarker development, modernization of clinical trials, and bioinformatics as top priorities for improvement (www.fnih.org/news/June_Press_2007.shtml). The document also listed 76 separate critical path projects that,
422
CHAPTER 35
■
Federal Regulation of Genomic Medicine
if accomplished, could significantly improve the evaluation and delivery of medical products to the public. A major focus of the Critical Path Initiative is how to translate existing basic science knowledge into valid new biomarkers that can be used reliably to guide medicine and regulation. Pharmacogenomic tests, like most diagnostics, are biomarkers. The process of establishing the utility of a biomarker for a given use (e.g., selecting patients who should receive, or should not receive, a specific drug) is termed “biomarker qualification.” Until recently, the need for qualification of biomarkers for various uses has been confused with the extremely difficult process of validating a biomarker as a surrogate endpoint for drug approval. As part of the Critical Path Initiative, FDA has been assessing the conceptual framework needed to understand the evidence needed to “qualify” a biomarker for use in various types of regulatory decision-making, and hopes to publish guidance on this topic. The major barrier to adoption of new biomarkers for regulatory use, and in clinical medicine, has been the lack of solid evidence about their value. Developing this evidence usually requires some type of clinical studies. Many of the currently marketed genetic tests have relied on pre-existing evidence derived from epidemiologic studies or clinical investigations of disease syndromes. Such data will rarely be available for genomic tests, and often diagnostic developers will lack the resources to conduct the necessary clinical trials. For this reason, FDA has focused on the creation of consortia that bring together interested stakeholders, enable pooling of resources, and can accept philanthropic contributions. For example, “The Biomaker Consortium,” a public–private partnership among NIH, FDA, PhRMA, BIO, and others, convened at the Foundation for NIH, has as its goal the discovery and qualification of new biomarkers (www.fnih.org/news/June_Press_2007.shtml). It is expected that genomic biomarkers will be prominent among these. Other consortia, either industrial or based at foundations or academic sites, are being formed in order to accomplish other biomarkerrelated work. Such groups can leverage ongoing trials being conducted by one of the partners (e.g., NIH or the pharmaceutical
industry) for other purposes by sponsoring substudies aimed at qualification of biomarkers relevant to the disorder being investigated in the trial. Clearly, the design and conduct of clinical trials, particularly trials during drug development, will need to be modified to incorporate exploratory use of genomic markers. Regulatory acceptance of various Bayesian, adaptive, and sequential designs will be needed. As part of the Critical Path Initiative, FDA plans to issue draft guidance on these topics, and engage in an effort to reach consensus within the scientific community on methodological approaches that are sufficiently robust to use in regulatory decision-making. New, multianalyte genomic tests are creating both opportunities and challenges in bioinformatics. FDA is developing experience in review of the resulting large datasets through its VGDS project. The agency is using the “Array Track” software developed at FDA’s National Center for Toxicologic Research to enable data review and storage. With further experience, the Agency plans to participate in data standards organizations to enable development of standards in this area. FDA is an active participant in the MACQ (microarray quality control) consortium that is aimed at standardizing microarray platforms (see http://www. fda.gov/nctr/science/centers/toxicinformatics/maqc/).
CONCLUSIONS Federal regulation of new technologies is intended to strike an appropriate balance between protecting the public and promoting innovation. This balance can be difficult to achieve during periods of rapid evolution. The FDA is seeking to both stimulate the field of genomics and also institute reasonable, scientifically based regulation. The next decade will bring significant advances in genomics, many new marketed genomic products, and a continuing evolution in the regulatory framework. Collaboration of the biomedical, regulatory, policy and legal communities will be necessary to ensure that the promise of genomic medicine is delivered in an orderly, responsible, timely and effective manner.
REFERENCES Bernstein, K. (2006). Targeted Narrow. BioCentury 14, 1–9. Budnitz, D.S., Pollock, D.A., Weidenbach, K.N., Schroeder, T.J. and Annest, J.L. (2006). National surveillance of emergency department visits for outpatient adverse drug events. JAMA 296, 1858–1866. Hinman, L.M., Huang, S-M., Hackett, J., Koch, W.H., Love, P.Y., Pennello, G., Torres-Cabassa., A. and Webster, C. (2006). The drug diagnostic co-development paper. Commentary from the 3rd FDA-DIA-PWG-PhRMA-BIO pharmacogenomics workshop. Pharmacogenomics J 6, 375–380. Hudson, K.L. (2006). Genetic testing oversight. Science 313, 1853. Hudson, K.L., Murphy, J.A., Kaufman, D.J., Javitt, G.H., Katsanis, S.H. and Scott, J. (2006). Oversight of US genetic testing laboratories. Nat Biotech 24, 1083–1090. Javitt, G.H. and Hudson, K.L. (2006). Federal neglect. Federal regulation of genetic testing. Issues SciTechnol XXII, 59–66.
Lazerou, J., Pomeranz, B. and Corey, P. (1998). Incidence of adverse drug reactions in hospitalized patients: A meta-analysis of prospective studies. JAMA 279, 1200–1205. Lesko, L.J., Salerno, R.A., Spear, B.B., Anderson, D.C., Anderson, T., Brazell, C., Collins, J., Dorner, A., Essayan, D., Gomez-Mancilla, B. et al. (2003). Pharmacogenetics and pharmacogenomics in drug development and regulatory decision making: Report of the first FDA-PhRMA-DruSAFE-PWG workshop. J Clin Pharmacol 43, 1–17. Lesko, L.J. and Woodcock, J. (2002). Pharmacogenomic-guided drug development: Regulatory perspective. Pharmacogenomics J 2, 20–24. Phillips, K.A., Veenstra, D.L., Oren, E., Lee, J.K. and Sadee, W. (2001). Potential role of pharmacogenomics in reducing adverse drug reactions. A systematic review. JAMA 286, 2270–2279. Ratain, J.M. (2006). From bedside to bench to bedside to clinical practice: An odyssey with irinotecan. Clin Cancer Res 12, 1658–1660.
Recommended Resources
Rauch, A., Nolan, D., Martin, A. et al. (2006). Prospective genetic screening decreases the incidence of abacavir hypersensitivity reactions in the Western Australia HIV Cohort Study. Clin Infect Dis 43, 99–102. Simon, R. (2005). Developing and validating genomic classifiers. http:// www.fda.gov/cder/genomics/biomarkers.htm Temple, R. (2005). Use of genomic biomarkers in a regulatory environment. http://www.fda.gov/cder/genomics.htm The European Agency for the Evaluation of Medicinal Products, London, 23 January, 2003. (CPMP/4445/03). White, T.J., Clark, A.G. and Broder, S. (2006). Genome-based biomarkers for adverse drug effects, patient enrichment and prediction of drug response, and their incorporation into clinical trial design. Pers Med 3, 177–185.
423
Woodcock, J. ( 2005). Importance of genomic biomarker validation in the context of pharmacogenomic initiatives at the FDA. http:// www.fda.gov/cder/genomics/biomarkers.htm 21 CFR 809.10 (c)(2)(i) 21 CFR 809.10 (c)(2)(ii) 21 CFR 809.10(e) 21 CFR 809.30 21 CFR 809.30 (a)(2) 21 CFR 809.3 21 CFR 864.4020
RECOMMENDED RESOURCES See presentations from FDA-DIA-PhRMA-BIO-PWG Workshop on “Application and Validation of Genomic Biomarkers for Use in Drug Development and Regulatory Decision Making”, Bethesda, MD, October 6–9, 2005. http://www.fda.gov/cder/genomics/ biomarkers.htm
■
Gene Tests. http://www.genetests.org/
CHAPTER
36 Economic Issues and Genomic Medicine David L. Veenstra, Louis P. Garrison, and Scott D. Ramsey
INTRODUCTION The era of genomic medicine offers significant promise for the development of novel health care technologies that ultimately improve patients’ quality of life and life expectancy. As with any new health care technology, questions arise as to their potential budgetary impact and cost-effectiveness. But because genomic technologies inherently involve diagnostic or prognostic testing, and because of the complexities of incomplete gene penetrance and multiple gene and environmental interactions, their assessment can be more challenging. In addition, perhaps more than in any other area of medicine, questions have arisen in regard to the economic incentives to develop these technologies. Formal health economics frameworks can be used to gain insights into these issues, and provide guidance for resource investment, technology appraisal, and policy development. In this chapter, we provide a brief introduction to the principles of health economics and discuss the unique aspects of the application of cost-effectiveness analysis to genomic medicine. We review drivers of the cost-effectiveness of genomic technologies and assess several recent state-of-the-art examples. We then examine the particular challenges in pharmacogenomics of economic incentives for developing genetic tests linked to therapeutics. Lastly, we propose an approach to help establish value-based reimbursement of genomic testing technologies. The content of this chapter will help readers understand the multidisciplinary
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 424
nature of the factors that will influence the value of genomic medicine, and provides a framework for assessing these factors in a systematic, quantitative fashion.
ECONOMIC EVALUATION AND COST-EFFECTIVENESS ANALYSIS Health Economics and Foundations of Economic Analysis Drawing on methods and concepts from economics, clinical epidemiology, psychology, and the decision sciences, the field of “cost-effectiveness” research has synthesized a set of tools and a theoretical framework for evaluating the complex issues in health care (Garber and Phelps, 1997; Weinstein et al. 1996). At a broad level, effectiveness estimates are generally derived from randomized clinical trials, disease progression estimates from long-term follow-up epidemiological studies, quality of life values from health state preferences studies, and costs from a wide variety of sources including payment schedules (i.e., Medicare) and attributable- and micro-costing studies. These factors, which are often considered either explicitly or implicitly in decisionmaking, can be synthesized using decision analytic techniques to inform formal economic analyses as discussed below (Detsky et al., 1997).
Copyright © 2009, Elsevier Inc. All rights reserved.
Economic Evaluation and Cost-Effectiveness Analysis
Methodological Approaches to Economic Evaluation in Health Care Several specific yet related methodological approaches are used in the economic evaluation of health care technologies: (1) costminimization, (2) cost-consequences analysis, (3) cost-benefit analysis, (4) cost-effectiveness analysis, and (5) cost-utility analysis (Flowers and Veenstra, 2004). These methods vary primarily in the way intervention effectiveness is valued. For example, for cost-minimization analysis, it is assumed there is no difference in effectiveness or side effects. In both cost-effectiveness and cost-consequences analysis, effectiveness is measured in natural, clinical units such as heart attacks or infections avoided. In cost-benefit analysis, a monetary value is assigned to effectiveness (e.g., a heart attack avoided might be “valued” at $100,000). And in cost-utility analysis, effectiveness is measured in qualityadjusted life years (QALYs), which account for improvements in both life expectancy and quality of life. Another characteristic that differentiates these methods is that cost-minimization and cost-consequences analyses do not involve the calculation of a ratio; for cost-minimization, only costs are presented, and in cost-consequences both costs and effectiveness are calculated, but not combined in a ratio. Of note, although “cost-effectiveness analysis” is a specific type of economic evaluation, the term is used generally to refer to all types of economic evaluation in health care. In some cases, cost-effectiveness studies can be based on a single randomized clinical trial. However, because clinical trials are usually in a controlled setting, the costs incurred are not representative of utilization in a real-world setting. The “efficacy” observed in a controlled setting can also be distinguished from the “effectiveness” that could be expected in practice. Finally, the timeframe of clinical trials for chronic conditions are generally not sufficient to evaluate long-term outcomes. Because of these reasons, as mentioned above, modeling techniques such as decision analysis are often used to extrapolate the results from clinical studies using primarily epidemiologic and economic data from other sources. In any economic evaluation, it is important that the technology being evaluated is compared to current medical practice. Weinstein and Stason, in 1977, defined the incremental costeffectiveness ratio (ICER) as ICER (C2 C1 ) (E2 E1 ) where C2 and E2 are the cost and effectiveness of the new intervention being evaluated, and C1 and E1 are the cost and effectiveness of the standard therapy (Weinstein and Stason, 1977). The costs and effects that are included in the equation depend on the perspective of the analysis. From a societal perspective, indirect costs and effects such as patient time away from work, downstream medical care costs years or decades after the intervention, and the quality of life of the patient and even their family need to be considered. Because of the all-encompassing nature of the societal perspective, it is generally better suited for
■
425
national health care plans. More relevant to health care plans or providers in the United States is the payer perspective, which addresses primarily direct medical care costs incurred by the payer (e.g., drug cost, professional fees, hospital stay). Resource Allocation Decision-Making How can cost-effectiveness information guide health policy decisions? The favored approach, from an academic perspective, is to utilize cost-utility analysis because it allows for comparisons across interventions and diseases, accounts for impact on life expectancy and quality of life, and has theoretical foundations in welfare economics. Medical interventions are considered to be cost-effective when they produce health benefits at a cost comparable to other commonly accepted treatments. A general guide is that interventions that produce one QALY (equivalent to 1 year of perfect health) for under $50,000 are considered costeffective, those between $50,000 and $100,000 per QALY are of intermediate cost-effectiveness, and above $100,000 per QALY generally is not considered cost-effective. The cutoff of $50,000 per QALY was derived loosely from the cost of providing dialysis for a patient for one year – a service paid for by Medicare for any US citizen. However, these criteria are somewhat arbitrary, and can vary across disease and indication-specific therapeutic areas, particularly in the United States (Neumann, 2007). The application of cost-effectiveness analysis has increased dramatically in the past decade as a result of increasing health care costs and the desire to deliver the greatest health value for the money. The formal application of cost-effectiveness analysis to drug coverage decisions has its origins in countries with single-payer health care systems (e.g., government sponsored). Recently, multiple countries and health care systems have begun to adopt requirements for such pharmacoeconomic information. These requirements formalize an otherwise implicit demand for health care technologies that are cost-effective, and will influence, to a certain extent, “go/no-go” decisions in drug research and development. The United Kingdom, Canada, and Australia all have formal requirements in place for cost-effectiveness information and programs in place for evaluating cost-effectiveness data – the National Institute for Health and Clinical Excellence (NICE, United Kingdom), the Canadian Agency for Drugs and Technologies in Health (CADTH, Canada), and the Pharmaceutical Benefits Advisory Committee (PBAC, Australia) (CADTH 2007; NICE 2007; PBAC, 2007), as well as several European countries such as the Netherlands (Drummond et al., 1999). In the United States, cost-effectiveness information is most often used in support of drug formulary listing in managed care settings. Many managed care organizations and pharmacy benefits managers utilize guidelines that require outcomes and economic information for formulary evaluation. In addition, the Academy of Managed Care Pharmacy (AMCP) has adopted guidelines for the submission of information, including outcomes and cost-effectiveness data, to support formulary consideration (Fry et al., 2003; Neumann, 2004; Spooner et al., 2007).
426
CHAPTER 36
■
Economic Issues and Genomic Medicine
EVALUATING GENOMIC TECHNOLOGIES Unique Aspects of Genomic Medicine The tools of health economics and cost-effectiveness analysis can be used to assess any health care technology, from immunization strategies, to new drug therapies, to educational programs. Genomic medicine is not so unique (i.e., exceptional) that fundamentally new methods in economic evaluation are needed. However, genomics can add layers of complexity to health care interventions such that particular care should be taken in their evaluation. Disease-Risk Testing The validity of a prognostic genomic test will have a significant impact on its potential cost-effectiveness. The evaluation of validity will be similar to other non-genomic prognostic tests, yet potentially more complex, due to challenges in validating genetic markers with small contributions from multiple genes and environmental interactions. Thus, when undertaking an economic study, care should be taken that the performance of the test has been evaluated by peer review, and perhaps more importantly, reproduced by independent investigators in different patient populations. An issue that will be more unique, to a degree, to genetic testing for disease risk is whether to test family members. The potential costs and clinical and patient outcomes should be considered. If potentially substantial, this strategy should be included in the analysis (Ramsey et al., 2001). Pharmacogenomics Pharmacogenomics tests face a similar issue with validation of a test’s ability to predict drug response and differentiating it from prognostic ability – that is, predicting disease prognosis rather than the effect of treatment on the disease. Furthermore, as discussed later in the chapter, the pairing of pharmacogenomic tests with drugs may lead to uncertainty about which – the test or the drug – provides the incremental value, and this can have implications for pricing or reimbursement decisions. Lastly, it is possible that genetic information obtained for one purpose may also provide information, for example, about disease risk. The potential impact of such ancillary information should be considered in a broad sense; for example, disease-risk information that is available after the test is conducted versus the use of CYP2D6 drug-metabolizing enzyme polymorphism data to help guide therapy for drugs other than the original one for which the test was obtained. Thus, there could be potential benefits but also harms (Henrikson et al., 2007). Cost-effectiveness studies of both disease-risk and pharmacogenomic tests in some cases may benefit from more complex analytical frameworks and simulation methods. However, it generally is not possible, due to statistical power limitations, to identify specific risks for individual patients. Rather, in genetic association studies, patients tend to be categorized into
larger subgroups to improve statistical power. For example, the breast cancer disease recurrence risk score (RS) derived from a 21-gene profile with the Oncotype Dx test could be utilized in a similar fashion to the Framingham models that predict the risk of a cardiovascular event based on non-genetic factors. In general, when more complex modeling approaches are employed, the added value should be explained in addition to justification of data sufficiency. Cost-Effectiveness Framework and Drivers We have previously developed a set of criteria, based on a formal cost-effectiveness framework, for evaluating the potential clinical and economic benefits of genomic tests (Table 36.1) (Flowers and Veenstra, 2004). The key aspects of these criteria are highlighted in the following paragraphs, and a more in-depth evaluation of the epidemiologic considerations in bringing pharmacogenomics to clinical practice is presented. Assessing the incremental cost-effectiveness of a pharmacogenomic strategy involves evaluating factors that are common to the evaluation of all screening strategies, as well as factors that are specific to genetic testing. These include factors associated with the genetic variation of interest, the genetic test, the disease state, and the treatment options. Questions to consider in assessing the cost-effectiveness of a pharmacogenomic treatment strategy include: ● ●
●
● ● ●
●
● ●
What is the frequency of the genetic variation? How closely is the variation linked to a consistent phenotypic drug response? Are there other significant influences on drug response such as diet, disease, or drug interactions? What are the sensitivity and specificity of the genomic test? How prevalent is the disease of interest? What are the characteristic outcomes associated with the disease with and without treatment? How does the pharmacogenomic strategy alter these outcomes? What alternative treatment options are available? How effective are current monitoring strategies for preventing severe adverse drug reactions and predicting drug response?
Pharmacogenomics is more likely to be cost-effective when: (1) the polymorphism under consideration is prevalent in the population and has a high degree of penetrance; (2) genetic testing is highly sensitive and specific, and less costly alternative tests that could be used to individualize therapy are not readily available; (3) the disease state involves outcomes with significant morbidity or mortality without treatment; and (4) the treatment involves significant outcomes and/or costs that can be impacted by genotype-individualized therapy. Case Study: Gene-Expression Profiling and Breast Cancer Treatment Breast cancer is the leading incident cancer among women of all major ethnicities in the United States and is the second highest
Evaluating Genomic Technologies
TABLE 36.1
Gene Test Disease
427
Factors that influence the cost-effectiveness of genomic testing strategies Factors to assess
Features that favor cost-effectiveness
Prevalence Penetrance
●
Sensitivity, specificity, cost
●
●
Variant allele is relatively common Gene penetrance is high
●
High specificity and sensitivity A rapid and relatively inexpensive assay is available
Prevalence
●
High disease prevalence in the population
Outcomes and economic impacts
●
High untreated mortality Significant impact on quality of life High costs of disease management using conventional methods
● ●
Treatment
■
Outcomes and economic impacts
● ● ● ●
Reduction in adverse effects that significantly impact quality of life or survival Significant improvement in quality of life or survival due to differential treatment effects Monitoring of drug response is currently not practiced or difficult No or limited incremental cost of treatment with pharmacogenomic strategy
Adapted from Flowers and Veenstra (2004).
source of cancer mortality. Adjuvant chemotherapy has been shown to increase recurrence-free and overall survival, but also may produce significant toxicity such as alopecia, nausea/vomiting, and myelosuppresion, and may lead to longer-term complications such as permanent ovarian failure in pre-menopausal patients (Shapiro and Recht, 1994, 2001). Current NIH clinical guidelines (Eifel et al., 2001) recommend adjuvant chemotherapy for women with tumors larger than 1 cm or lymph node involvement. Additionally, tumor markers such as HER2 and histologic grade are used for risk assessment (Bast et al., 2001; NCCN, 1999). Despite widespread use, these criteria are imprecise predictors of distant recurrence (Sauter and Simon, 2002). Gene-expression profiling (GEP) utilizing DNA microarrays (van de Vijver et al., 2002) or RT-PCR (Paik et al., 2004) has been proposed as an alternative approach to identify patients for adjuvant chemotherapy, (King and Sinha, 2001) potentially sparing low-risk patients from this treatment. There are two gene-expression profiles currently marketed for clinical use in breast cancer. One of these profiles, MammaPrint® (Agendia), was developed by van’t Veer and colleagues at the Netherlands Cancer Institute (van de Vijver et al., 2002). The other test, Oncotype DxTM, was developed by Genomic Health, Inc. in the United States (Paik et al., 2004). The assay marketed by Agendia utilizes a 70-gene microarray-based profile performed on fresh frozen tissue and is intended for patients younger than 55 years with Stage I invasive breast cancer or Stage II node-negative invasive breast cancer. In contrast, the test marketed by Genomic Health employs a 21-gene profile utilizing RT-PCR for expression analysis on paraffin embedded tissue and is intended for patients with node-negative, estrogen receptor positive (ER) disease. Although gene-expression
profiling has been proposed as an alternative to clinical guidelines to identify appropriate patients for adjuvant chemotherapy, the potential long-term clinical and economic outcomes associated with gene-expression profiling are not clear. In other words, “Do the quality of life benefits and cost savings of avoiding chemotherapy outweigh the potential increased risk of recurrence in women not given chemotherapy and cost of the test?” To address this question, we developed a decision analytic model to evaluate the incremental cost and QALYs of geneexpression profiling using MammaPrint versus NIH clinical guidelines in a hypothetical cohort of pre-menopausal earlystage breast cancer patients (Oestreicher et al., 2005). We assumed patients and doctors would follow test or guideline recommendations, and that chemotherapy response would be similar regardless of risk group, as no data to the contrary were available. Our findings suggested that use of MammaPrint gene-expression profiling could result in an absolute 5% decrease in the proportion of cases of distant recurrence prevented, 0.21 fewer QALYs, but a cost savings of $2800 (Table 36.2). Regardless of the test cutoff used to identify a poor prognosis tumor, the MammaPrint gene-expression assay did not produce equal or greater QALYs than NIH guidelines. These findings are fairly striking, and importantly, what do they imply? First, test performance is very important when a pharmacogenomic test is being used to recommend against treatment that is standard of care. In this case, because of test sensitivity of 84% for MammaPrint versus 98% for NIH criteria, some women who will progress will be categorized as low risk and not recommended for chemotherapy. Secondly, the efficacy of treatment in women categorized as low risk versus high risk can have a significant impact. In other words, prognostic ability
428
CHAPTER 36
■
Economic Issues and Genomic Medicine
TABLE 36.2 Performance, costs and outcomes of gene expression profiling versus NIH guidelines
Outcome
Gene Expression Profiling
NIH Guidelines
Difference
Sensitivity
84%
98%
14%
Specificity
51%
5%
46%
Proportion of women treated with chemotherapy
61%
96%
35%
Proportion of distant recurrences prevented
29%
34%
5%
Costs
$29,754
$32,636
$2882
Quality-adjusted life years
9.86
10.08
0.21
Adapted from Oestreicher et al. (2005).
to evaluate disease risk is important, but so is predictive ability for treatment response. In an evaluation of the clinical and economic utility of a test, these two aspects of test performance, disease-risk and treatment response, should be clearly delineated. In some cases, as illustrated here, prognostic (disease-risk) information alone may not be sufficient to justify utilization of a genomic test. The Oncotype Dx assay has been evaluated for its ability to predict chemotherapy outcomes in addition to disease recurrence risk. Paik et al. reported in a retrospective study that women with a low RS are also unlikely to respond to chemotherapy (relative risk, 1.31; 95% CI, 0.46–3.78) (Paik et al., 2006). The impact of this effect was incorporated in a cost-effectiveness analysis conducted by Hornberger and colleagues (2005). Using a similar methodological approach to Oestreicher et al., the author developed a Markov model and forecast overall survival, costs, and cost-effectiveness of using the Oncotype Dx RS in patients classified as having low or high risk of distant recurrence based on National Comprehensive Cancer Network (NCCN) clinical guidelines. In the analysis, 8% of patients were classified as having low risk of distant recurrence by NCCN guidelines, and the RS reclassified 28% of these patients to an intermediate/highrisk group. The remaining 92% of patients were classified at high risk of distant recurrence by NCCN guidelines and the RS reclassified 49% of these to a low-risk group. Overall, use of the Oncotype Dx test was predicted to increase quality-adjusted survival by 0.09 years and reduce overall costs by $2000. The costeffectiveness was most influenced by the propensity to administer chemotherapy based on test results, and by the proportion of patients at low risk as defined by NCCN guidelines. The author concluded that if applied appropriately, the test is predicted to increase quality-adjusted survival and save costs.
These cost-effectiveness studies served several valuable purposes. Most importantly, they clarified that without the ability to predict response to chemotherapy, these gene-expression profiles may not provide justifiable improvements (if any) in overall patient outcomes in a cost-effective manner. Thus, important areas for subsequent validation work have been highlighted. Furthermore, such analysis can help inform reimbursement decisions (Watkins et al., 2007). Thus, for example, with the Oncotype Dx test, assuming the prognostic and particularly the predictive validity of the test hold, the test may offer not only a cost-effective intervention, but potentially a cost-saving one. Another issue that remains to be addressed, preferably in “real-world” (non-controlled) settings, is the concordance of test results with actual treatments received. For example, a recent study indicated that a majority of sampled physicians and patients found the test results useful, but there was an absolute decrease in the proportion of patients choosing chemotherapy of only 2% (Lo et al., 2007). These treatment patterns may change as clinicians and patients gain experience in the use of genomic test results in decision-making. Of note, in an effort to validate these retrospective findings, the first US-based phase III randomized controlled clinical trial – The Trial Assigning IndividuaLized Options for Treatment (Rx), or TAILORx, comparing the efficacy of Oncotype Dx for choosing women receiving adjuvant chemotherapy – was launched on May 23, 2006, to examine whether genes that are frequently associated with risk of recurrence for women with early-stage breast cancer can be used to assign patients to the most appropriate and effective treatment [http://www.cancer. gov/clinicaltrials/digestpage/TAILORx, accessed 11/9/06]. This study will provide valuable prospective validation of previous retrospective studies, as well as the modeled outcomes from cost-effectiveness studies.
ECONOMIC INCENTIVES AND THE FUTURE OF GENOMIC MEDICINE Economic Incentives in Drug and Test Development Eight years have elapsed since the initial sequencing of the human genome, and the number of new genetic tests commonly used in clinical practice is quite small. For example, a recent report from the Royal Society cautions: “Pharmacogenetics is unlikely to revolutionize or personalize medical practice in the immediate future” (The Royal Society, 2005). Robert Califf has argued that achieving this promise will require a major overhaul of the US clinical research enterprise as well as substantial educational efforts (Califf, 2004). In two recently published papers, Garrison and Austin explored the lack of appropriate economic incentives that may in part be contributing to the current pace, in addition to the inherent challenges in genomics research (Garrison and Austin, 2006, 2007). The translation of the basic science of pharmacogenetics
Economic Incentives and the Future of Genomic Medicine
and other “-omics” biomarkers as applied to drug development and clinical care is occurring in a complex legal, regulatory, and reimbursement environment. Understanding and appropriately shaping this environment is vital for encouraging biomarker research and personalized health care. Suppliers of new medical technologies – both therapeutics and diagnostics – face a complex and heavily regulated commercial environment. Furthermore, the manner in which diagnostics are reimbursed in most developed countries may provide limited incentives for the development of new genomics-based tests. Both the proponents of genetic tests and the skeptics sometimes fail to appreciate this complexity. Proponents sometimes argue: “How could this not be a good thing?” We would be able to target drugs to the subgroups who respond most favorably and limit the delivery to those most likely to suffer side effects. Skeptics sometimes argue that no pharmaceutical company would have incentive to restrict its market to a smaller subgroup: “Why would they ever want to develop a test to do this?” Both sides have a valid point to some extent, but neither represents the full complexity of the incentives involved. Garrison and Austin analyzed the incentives in terms of “value creation and capture.” Innovative therapies can create aggregate economic value in at least five alternative ways that are not mutually exclusive. First, they may reduce mortality and morbidity compared to current treatments. Second, they may save on average on the costs of treating side effects or the underlying disease. Third, they may result in greater utilization by those afflicted. Fourth, they may encourage better compliance, providing individual patients with more net benefit. Fifth – and this is less recognized and is important for new diagnostic tests – they can make people better off by reducing uncertainty. The authors developed and analyzed five different scenarios of combining a new genomics-based diagnostic with a therapeutic treatment. They argued that who captures the value created by the combination depends on a number of factors, and that it is the potential for value capture that generates the incentives for genomic test development and marketing. In thinking about incentives in this marketplace, it is important to keep a few key features in mind. First, intellectual property rights are essential to provide a basic incentive for the invention and development for new diagnostic tests and new therapeutics, particularly pharmaceuticals. It is widely known and appreciated that new innovative drugs receive temporary (20-year) patents that limit the ability of competitors to enter since they can’t just copy the innovative compound and would have to develop and fully test a new molecule. It is less appreciated that effective patent life is 8–12 years, as it takes years to perform the trials that regulators review for licensing. A second key feature is that once a product is on the market and has a price, it can be difficult to change that price. Indeed, in some countries, “administered pricing” is used to set the price based on reference products or prices in other countries. The price granted in those situations may only relate in a limited way to the value created, and they may not be allowed to change over time. This is particularly true in Europe, whereas there is some ability to change pharmaceutical prices over time in the United States.
■
429
Diagnostics differ from drugs in that their prices are even more heavily controlled in most markets, as initial price is often pegged to perceived costs of provision rather than additional value created. This can be called “cost-based” pricing and reimbursement as opposed to “value-based” pricing and reimbursement. In effect, suppliers of new genomics-based diagnostics are likely to capture less of the incremental value they create. Given these marketplace realities and constraints, Garrison and Austin argue that the incentives for value creation through a linked genomics-based diagnostic and therapeutic will depend on these factors as well as whether the therapy is already on the market. For their hypothetical exercise, they posited that a new diagnostic is developed in a market where only 20% of patients truly benefit from a treatment, but heretofore they could not be identified. Eliminating the 80% of users who don’t benefit would be a great benefit to payers but would be damaging to the pharmaceutical company if they could not raise price for the 20% of responders. On the other hand, it should clear that if the test were available before the drug was launched, the company might have been able to charge a price nearly five times as high and still capture the same amount of value. In principle, if the innovative diagnostic enters the market after that drug has a price, then the diagnostic manufacturer might be able to capture this value, creating a powerful incentive for developing the diagnostic. However, in a regulated, costbased pricing and reimbursement environment for diagnostics, there will not be this strong incentive. Hence, if the drug manufacturer can’t easily increase price for the 20% who respond, and if the potential diagnostic manufacturer can’t capture the cost savings because of price controls, there will be a limited incentive to develop tests for already marketed products. The authors point out that if the diagnostic and therapeutic could be developed in tandem, even more total economic value would be created due to the additional reduction in uncertainty. The drug manufacturer would be in a better position to capture this value, given that they operate in a relatively value-based pricing environment. However, drug development is a challenging and uncertain scientific enterprise in and of itself, and adding the complexity of companion diagnostic may not be that appealing unless the scientific rationale is already well developed. In addition to the scientific and economic barriers highlighted above, two other factors deserve mention that might be inhibiting the development of pharmacogenetics-based test-drug combinations. One factor is the high cost of the basic research that is needed to validate genetic markers. Financing this remains a question: What should fall to the public sector, how much should the private sector contribute, and what is best done in partnership are under active debate. Second, approval and reimbursement for drugs in the United States and EU require greater levels of evidence than is customary for new diagnostic tests. Some would argue that the lower evidentiary requirements for regulatory approval of tests have discouraged the development of better clinical data and that payers practice “cost-based” reimbursement in part because of this lack of evidence on clinical and economic value. In contrast, both the AMCP Format for
430
CHAPTER 36
■
Economic Issues and Genomic Medicine
Formulary Submissions guidelines in the United States and the requirements for submissions to the National Center for Clinical and Health Excellence (NICE) in the United Kingdom request economic models that synthesize the clinical and cost evidence to assess value added of new drugs. Similar standards and mechanisms do not exist for diagnostics, although the discussion is beginning. But asking for more evidence means raising the costs of developing – and particularly validating – new diagnostics. What are the incentives for diagnostic manufacturers if no additional rewards are forthcoming through the reimbursement system? Below we discuss a potential approach for incorporating evidence assessment into reimbursement decisions for genomic tests.
ESTABLISHING VALUE-BASED REIMBURSEMENT FOR GENOMIC TECHNOLOGIES Rationale Laboratory tests are an important yet often overlooked segment of health care and the health economy. For example, it has been estimated that as much as $56 billion was spent on laboratory diagnostic services in 2005 (Blue Cross Blue Shield Association, 2005). More importantly, laboratory tests initiate a cascade of decisions regarding further testing, prevention or treatment-decisions that ultimately determine health outcomes and costs of care. A recent report estimated that although diagnostics account for 1.6% of the Medicare total costs, they influence 60–70% of downstream treatment-decisions (The Lewin Group, 2005). Genomic testing for mutations, polymorphisms, and haplotypes expands the laboratory diagnostics market in many areas, including pregnancy and neonatal testing, predictive testing for disease susceptibility, and pharmacogenomic testing. As noted above, gene-expression profiling of tumors is a burgeoning field that includes both predictive and pharmacogenomic characteristics. Historically, the evidence supporting the validity and clinical utility of laboratory tests has been limited (Feinstein 2002; Reid et al., 1995; Weinstein et al., 2005). The pathway to the marketing of tests does not require the same data as therapeutics, and technology assessments of laboratory tests have not undergone the revolutionary changes in evidence review that has occurred for drugs (Neumann, 2004). Instead, payment for tests is linked to an antiquated coding system, as noted above (Raab and Logue, 2001). Recently, we presented a rationale and outline for re-structuring the way laboratory tests are evaluated and reimbursed (Ramsey et al., 2006). We summarize those concepts below. Process We propose that coverage and reimbursement for laboratory tests, including genomic tests, should move toward an evidenceand value-based approach, using some of the tools that have been adopted for pharmaceutical assessment by many US health care payers.
Developing a Language for Describing Benefit Test manufacturers, laboratory service providers, and health insurance plans will benefit from standardizing the way evidence supporting new laboratory tests is presented. Methodological standards for the evaluation of diagnostic tests have been published (Reid et al., 1995), and a useful framework has been proposed to evaluate diagnostic technologies (Table 36.3) (Fryback and Thornbury, 1991). While many agree about the value of using these domains to evaluate tests, including genomic tests, there is less agreement on how much evidence is necessary for an insurance coverage decision. Considering the domains from Table 36.1 when applied to genomic tests, it is important to evaluate the incremental impact of a test; that is, the improvement that the new test provides overcurrent (non-genomic) strategies (Garrison and Austin, 2006). Since prospective trials directly comparing new laboratory tests with established diagnostic strategies are uncommon (particularly those evaluating their impact on patient outcomes), decisionanalytic modeling techniques can aid evaluations (Buxton et al., 1997). Models help frame questions, provide transparent mechanisms for stating hypotheses about cause and effect, highlight deficiencies in clinical data, and force decision-makers to make explicit judgments about values for data that are used to inform the model. Published standards for models are readily accessible to decision-makers who wish to assess their quality (Weinstein et al., 2003).
TABLE 36.3
Hierarchy of diagnostic evaluation
Level
Characteristic
Description
1
Technical feasibility and optimization
Ability to produce consistent results
2
Diagnostic accuracy
Sensitivity, specificity, positive predictive value, negative predictive value
3
Diagnostic thinking impact
Percentage of times physicians’ estimated probability of a diagnosis changes following the test result
4
Therapeutic choice impact
Proportion of times planned therapeutic strategy changes following test results
5
Patient outcome impact
Percentage of patients who improve with the test versus those who improve without the test
6
Societal impact
Cost-effectiveness
Adapted from Fryback and Thornbury (1991).
Conclusions
A Definition of Value for Genomic Tests We define value for genomic tests as it is defined for other health technologies: the intervention provides an overall benefit to the patient at an acceptable cost; or in other words is costeffective. There are four well-recognized criteria for identifying an intervention cost-effective: (1) Less costly and at least as effective; (2) More effective and more costly, with the added benefit worth the added cost; (3) Less effective and less costly, with the added benefit of the alternative not worth the added cost; and (4) Cost saving with an outcome equal to or better than that of the alternative. Assessing value for tests can be difficult because tests are intermediate steps in the treatment pathway. The advantage of a cost-effectiveness framework is that it allows flexibility, because value for genomic tests can be defined narrowly (e.g., least expensive way to make a diagnosis) or broadly (improves survival at an acceptable added cost). A Format for Dialogue Among Genomic Test Manufacturers, Providers, and Payers When health insurers and test representatives come together to discuss genomic tests, the process should be transparent, and adopting a standard format would help clarify expectations and improve the decision-making process. A useful template for tests has been developed by the AMCP for the evidence-based evaluation of drugs. More than 50 public and private health insurers covering over 100 million lives have adopted the AMCP format (Fry et al., 2003; Neumann, 2004). A template for manufacturer’s reporting of clinical and economic information regarding laboratory tests is shown in Figure 36.1. Since the clinical utility of genomic tests is most often similar to other laboratory tests, the template accommodates differences in evidence that are typically available for new genomic tests. Implementation Some health plans might have a designated organizational unit that evaluates genomic tests, providing a structure for soliciting
1. Product Information
■
431
and reviewing manufacturer’s products. In the case of pharmaceutical products, that unit is the Pharmacy and Therapeutics (P&T) committee. Health plan staff working with P&T committees usually are receptive to receiving information from manufacturers or service laboratories about new products. Indeed, the “unsolicited request” process pioneered by the AMCP was designed to create a structure for dialogue between manufacturers and payers. A similar scheme would improve transparency and the flow of information for new genomic tests. P&T committee members are often not health plan employees. Having such a quasi-independent group evaluate novel laboratory tests may improve the credibility of the decision process in the eyes of manufacturers, clinical laboratories, physicians, and patients. Still, maintaining committees is costly, and may not be justified given the relatively low volume of genomic tests that are introduced annually. One option is to fold the test evaluation process into the existing P&T structure. Alternatively, payers could hire consultants to evaluate select genomic tests and make recommendations regarding coverage and reimbursement. Payers should be timely, in both coverage decisions and setting reimbursement levels, and when decisions are made, they should be supported with a rationale. In cases where requests for coverage are denied, such information allows manufacturers to design studies or collect other data that address concerns regarding the quality or content of the information supporting the product. Genomic tests are developed under a different regulatory structure than pharmaceuticals and have a unique and complex function in medical care. Novel genomic tests often come to market with little information supporting their role in clinical decision-making or evidence regarding their impact on patient outcomes. We posit that patient care and outcomes will improve when there is a structured process for evidence generation and discussion between payers, manufacturers, and marketers. While it is unlikely and perhaps not necessary that the evaluation process for genomic tests will equal what is required for drugs, we can move much further toward a system that supports better gathering and sharing of high-quality evidence. Test manufacturers, clinical laboratories, payers, and clinicians all must play a role in this process.
1.1. Product description. 1.1.1. Place of the product in therapy. 1.1.2. Disease description. The disease description should include the disease and characteristics of the patients in the target population.
2. Supporting Clinical and Economic Information 2.1. Evidence-table spreadsheets of all published and unpublished clinical studies. 2.2. Outcomes studies and economic evaluation supporting data 2.2.1. Evidence-table spreadsheets of all published and unpublished outcomes studies.
3. Cost-effectiveness Modeling Report 3.1. Model overview. We recommend that producers and users of modeling studies subscribe to the sound guidance provided by the ISPOR Good Practice Modeling Principles.
4. Supporting Information 4.1. 4.2.
References contained in dossiers. References for economic models.
Figure 36.1 Evidence and transparency standard to support coverage and reimbursement for diagnostic, therapeutic, and genetic testing (from Ramsey et al., 2006).
CONCLUSIONS The primary economic challenges facing genomic medicine can be surmised as the following: (1) providing evidence of the value of genomics and (2) providing economic rewards/incentives commensurate with the value added. More specifically, as outlined in a recent report developed on the implications of pharmacogenomics for the pharmaceutical and biotechnology industries, there are several key challenges for genomic medicine (Garrison et al., 2007): ●
Regulatory pathways have not yet been optimized to encourage the co-development of diagnostics and therapeutics.
432
●
●
●
●
CHAPTER 36
■
Economic Issues and Genomic Medicine
Current economic incentives – as reflected in our intellectual property and reimbursement systems for diagnostics and drugs – are generally not structured to reward appropriately and consistently innovative value creation for drugs, diagnostics, and pharmacogenomics-based targeted regimens. The integration of pharmacogenomic diagnostic development with pharmaceutical development is difficult because of differences in the underlying business and translational science models of the two sectors. Genomics technologies are perceived to raise ethical, legal, and social issues to such a degree that a special National Institutes of Health program was established to address them; although specific genomic applications may not always involve such issues, this suggests the broad range of stakeholders that will be involved in the public debates. Stakeholder literacy about genomics is limited, and positions on public policy issues are not yet clearly defined.
What can be done to help achieve these goals? We suggest several policy areas for focus (Garrison et al., 2007): ●
●
●
●
●
Pharmaceutical and biotechnology companies will need to add a systematic evaluation of potential pharmacogenomic and other biomarkers as part of their due diligence research and development processes. Substantial federal government support for basic research generally will be critical before translational private research activities are viable. The pharmaceutical and biotechnology industries will be major beneficiaries of this basic research, and should participate actively in the public discussion of priorities. Companies will need to provide a rationale to regulators why they have or have not included genomics or other biomarkers in their clinical trial development programs. The pharmaceutical, biotechnology, and diagnostic industries have not taken a unified or proactive position on
●
appropriate regulatory processes and initiatives, and it may not be possible to reach a consensus. Still, it may be a good time to begin a policy review and discussion. For pharmacogenomics-targeted pharmaceuticals to have greater commercial viability, the pharmaceutical, biotechnology, and diagnostic industries must engage in the public policy debate on national coverage and reimbursement issues for such drugs and tests.
Will cost-effectiveness analysis by itself speed the development of genomic medicine? This is not inherently likely. The fundamental challenges facing genomic medicine at this point are (1) the identification and validation of associations between genetic variation and clinically meaningful outcomes and (2) evaluation of interventions based on genomic information in prospective, comparative trials. However, cost-effectiveness analysis can help guide this process in several important ways: ●
●
●
By identifying genomic technologies that have potential value in early stages of drug and test development, thus aiding research portfolio optimization. By providing a framework for value-based reimbursement of genomic technologies, thus conferring appropriate incentives to the market. By identifying areas where future research would provide significant value to the health care system even once a technology has been developed.
Thus, quantitative economic analyses in combination with scientific research and policy analysis can help optimize the benefits of genomic medicine to society. In summary, genomic medicine offers significant hope for fundamental improvements in health care outcomes. These gains can likely be achieved in a cost-effective manner, but the challenges of providing evidence of and reward for value will prove to be critical in the coming decade.
REFERENCES Bast, R.C., Jr., Ravdin, P. et al. (2001). 2000 update of recommendations for the use of tumor markers in breast and colorectal cancer: Clinical practice guidelines of the American Society of Clinical Oncology. J Clin Oncol 19, 1865–1878. Blue Cross Blue Shield Association. (2005). Medical technology as a driver of healthcare costs: Diagnostic imaging. Retrieved August, 2005, from http://www.bcbs.com/coststudies/reports/Medical_ Tech_Drivr_Rept_10.pdf. Buxton, M.J., Drummond, M.F. et al. (1997). Modelling in economic evaluation: An unavoidable fact of life. Health Econ 6(3), 217–227. CADTH (2007). Canadian Agency for Drugs and Technologies in Health. Califf, R.M. (2004). Defining the balance of risk and benefit in the era of genomics and proteomics. Health Aff (Millwood) 23(1), 77–87.
Detsky, A.S., Naglie, G. et al. (1997). Primer on medical decision analysis: Part 2—Building a tree. Med Decis Making 17(2), 126–135. Drummond, M., Dubois, D. et al. (1999). Current trends in the use of pharmacoeconomics and outcomes research in Europe. Value Health 2(5), 323–332. Eifel, P., Axelson, J.A. et al. (2001). National Institutes of Health Consensus Development Conference Statement: Adjuvant therapy for breast cancer, November 1–3, 2000. J Natl Cancer Inst 93, 979–989. Feinstein, A.R. (2002). Misguided efforts and future challenges for research on diagnostic tests. J Epidemiol Community Health 56(5), 330–332. Flowers, C.R. and Veenstra, D. (2004). The role of cost-effectiveness analysis in the era of pharmacogenomics. Pharmacoeconomics 22(8), 481–493.
References
Fry, R., Avey, S. et al. (2003). The Academy of Managed Care Pharmacy Format for Formulary Submissions: An evolving standard–a Foundation for Managed Care Pharmacy Task Force report. Value Health 6(5), 505–521. Fryback, D.G. and Thornbury, J.R. (1991). The efficacy of diagnostic imaging. Med Decis Making 11(2), 88–94. Garber, A.M. and Phelps, C.E. (1997). Economic foundations of costeffectiveness analysis. J Health Econ 16(1), 1–31. Garrison, L.P., Jr. and Austin, M.J.F. (2006). Linking pharmacogeneticsbased diagnostics and drugs for personalized medicine: Scientific and economic challenges. Health Aff (Millwood) 25(5), 1281–1290. Garrison, L.P. and Austin, M.J.F. (2007). The economics of personalized medicine: A model of incentives for value creation and capture. Drug Information journal 41, 501–509. Garrison, L.P., Veenstra, D.L. et al. (2007). Backgrounder on Pharmacogenomics for the Pharmaceutical and Biotechnology Industries: Basic Science, Future Scenarios, Policy Directions, University of Washington, Pharmaceutical Outcomes Research and Policy Program. Henrikson, N.B., Burke, W. et al. (2007). Ancillary risk information and pharmacogenetic tests: Social and policy implications. Pharmacogenomics J 8, 85–89. Hornberger, J., Cosler, L.E. et al. (2005). Economic analysis of targeting chemotherapy using a 21-gene RT-PCR assay in lymph-nodenegative, estrogen-receptor-positive, early-stage breast cancer. Am J Manag Care 11(5), 313–324. King, H.C. and Sinha, A.A. (2001). Gene expression profile analysis by DNA microarrays: Promise and pitfalls. JAMA 286, 2280–2288. Lo, S., Norton, J. et al. (2007). Prospective multi-center study of the impact of the 21-gene Recurrence Score (RS) assay on medical oncologist (MO) and patient (pt) adjuvant breast cancer (BC) treatment selection. Am Soc Clin Oncol. Comment: Chicago NCCN (1999). Update: NCCN practice guidelines for the treatment of breast cancer. National Comprehensive Cancer Network. Oncology (Huntington) 13, 41–66. Neumann, P.J. (2004). Evidence-based and value-based formulary guidelines. Health Aff (Millwood) 23(1), 124–134. Neumann, P. (2007). The Cost-Effectiveness Analysis Registry [Internet], Tufts-New England Medical Center, ICRHPS. NICE (2007). National Institute for Health and Clinical Excellence. Oestreicher, N., Ramsey, S.D. et al. (2005). Gene expression profiling and breast cancer care: What are the potential benefits and policy implications?. Genet Med 7(6), 380–389. Paik, S., Shak, S. et al. (2004). A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med 351(27), 2817–2826. Paik, S., Tang, G. et al. (2006). Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer. J Clin Oncol 24(23), 3726–3734.
■
433
PBAC. (2007). Pharmaceutical Benefits Advisory Committee. Retrieved 7/10/07, from http://www.health.gov.au/internet/wcms/ publishing.nsf/Content/health-pbs-general-listing-committee3. htm#esc. Raab, G.G. and Logue, L.J. (2001). Medicare coverage of new clinical diagnostic laboratory tests: The need for coding and payment reforms. Clin Leadersh Manag Rev 15(6), 376–387. Ramsey, S.D., Clarke, L. et al. (2001). Cost-effectiveness of microsatellite instability screening as a method for detecting hereditary nonpolyposis colorectal cancer. Ann Intern Med 135(8 Pt 1), 577–588. Ramsey, S.D., Veenstra, D.L. et al. (2006). Toward evidence-based assessment for coverage and reimbursement of laboratory-based diagnostic and genetic tests. Am J Manag Care 12(4), 197–202. Reid, M.C., Lachs, M.S. et al. (1995). Use of methodological standards in diagnostic test research. Getting better but still not good. JAMA 274(8), 645–651. Sauter, G. and Simon, R. (2002). Predictive molecular pathology. N Engl J Med 347, 1995–1996. Shapiro, C.L. and Recht, A. (1994). Late effects of adjuvant therapy for breast cancer. J Natl Cancer Inst Monogr, 101–112. Shapiro, C.L. and Recht, A. (2001). Side effects of adjuvant treatment of breast cancer. N Engl J Med 344, 1997–2008. Spooner, J.J., Gandhi, P.K. et al. (2007). AMCP Format dossier requests: Manufacturer response and formulary implications for one large health plan. J Manag Care Pharm 13(1), 37–43. The Lewin Group. (2005).The Value of Diagnostics Innovation, Adoption and Diffusion into Health Care. Retrieved August, 2005, from http://www.advamed.org/publicdocs/thevalueofdiagnostics.pdf. The Royal Society (2005). Personalised Medicine: Hopes and Realities.The Royal Society, London. van de Vijver, M.J., He, Y.D. et al. (2002). A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347, 1999–2009. Watkins, J., Bresnahan, B. et al. (2007). Determining the Value of Diagnostic/ Genetic Testing to Inform Health Plan Decision Making: Oncotype DX as a Case Study. International Society of Pharmacoeconomics and Outcomes Research,Washington DC. Weinstein, M.C., O’Brien, B. et al. (2003). Principles of good practice for decision analytic modeling in health-care evaluation: Report of the ISPOR Task Force on Good Research Practices—Modeling Studies. Value Health 6(1), 9–17. Weinstein, M.C., Siegel, J.E. et al. (1996). Recommendations of the Panel on Cost-effectiveness in Health and Medicine. JAMA 276(15), 1253–1258. Weinstein, M.C. and Stason, W.B. (1977). Foundations of costeffectiveness analysis for health and medical practices. N Engl J Med 296(13), 716–721. Weinstein, S., Obuchowski, N.A. et al. (2005). Clinical evaluation of diagnostic tests. Am J Roentgenol 184(1), 14–19.
CHAPTER
37 Public–Private Interactions in Genomic Medicine: Research and Development Subhashini Chandrasekharan, Noah C. Perin, Ilse R. Wiechers, and Robert Cook-Deegan
INTRODUCTION Genomics grew out of the publicly funded Human Genome Project (HGP), which began to take shape in 1985 from programs initiated by the US Department of Energy (DOE), National Institutes of Health (NIH), and other non-profit and government partners around the world (Cook-Deegan, 1994a; National Research Council, 1988; US Congress Office of Technology Assessment, 1988). As originally conceived, the HGP was a public works project intended to create informational tools to enhance and expedite science. These tools would reside in the public domain available through scientific publications in open literature and public databases. One goal of the HGP was to create a reference sequence of the human genome, a “Rosetta Stone from which the complexities of gene expression in development can be translated and the genetic mechanisms of disease interpreted” (McKusick and Ruddle, 1987). The HGP also depended on automated DNA sequencing and sample-handling robotics, products of private sector Research and Development (R&D). DNA sequencing instruments made by Applied Biosystems and LKB-Pharmacia, LI-COR, and MJ played a role in creating the human reference sequence. In the late 1980s, instrumentation for DNA sequencing, mapping, and polymerase chain reaction became a growth sector. Applied
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 434
Biosystems (ABI), which introduced instruments to sequence and synthesize both DNA and proteins when it was an independent startup, subsequently became a major profit center within its parent company (initially Perkin-Elmer Cetus, then PE, and later Applera). The 1987 debut of ABI’s DNA “sequenator”, which pioneered four-color fluorescence in each lane of slab-gel electrophoresis, was a major technological landmark. In the early 1990s, more than a half decade after the government and non–profit-funded genome projects began, the first dedicated genomic startup firms began to appear. Most of the private genomics R&D investment that began in 1992 and 1993 was initially in US firms or foreign forms investing in US genomics R&D (Cook-Deegan, 1994b). In 1992 and 1993, new genomics firms such as Human Genome Sciences (HGS), Hyseq, Mercator Genetics, Millennium Pharmaceuticals, Myriad Genetics, and Sequenom were created. Incyte shifted from doing contract research for Genentech and turned its attention to genomics; Collaborative Research changed its name to Genome Therapeutics to reflect its emphasis on genomics. These first stirrings later proliferated to the more than 300 firms reported here, which have a significant aspect of their business plan based on genomics. A feature shared by the many companies that have become known as “genomics” firms is that all or a substantial fraction of
Copyright © 2009, Elsevier Inc. All rights reserved.
Landscape of Private Sector Genomics
their business plans hinge on creating or using data about DNA: for example, using large datasets containing information about DNA sequence, or interpreting such data for practical use (e.g., as a research service, in diagnostics, or as a component in drug discovery). These firms have drawn heavily upon the scientific commons – three decades of molecular biology and human genetics research – to create the subsector of biotechnology that is commercial genomics. Some of the firms have been acquired or merged; others have died; some continue as independent firms. Large and established firms such as Perkin-Elmer, HewlettPackard (through Agilent), Beckman, and others have developed technologies for both public and private genomics R&D laboratories. While US federal health research funding grew faster than funding in other areas of basic science for most of the period 1985–2004 (National Science Foundation, 2005), privately funded health R&D grew even faster, and surpassed public and non-profit funding by the late 1980s (Moses, 2005). This general pattern of private health R&D exceeding government and nonprofit funding was even more sudden and extreme in genomics. A 2000 world survey of genomics research funding found that private firms funded significantly more genomics R&D than the public sector (government and non-profit) (Cook-Deegan et al., 2001). Another distinctive historical feature of genomics is its growth at a time when intellectual property norms shifted to much wider use of patenting in the academic sector. The HGP was conceived in 1985, just four years after the Bayh-Dole and Stephenson-Wydler Acts went into effect in the United States (Rudolph, 1995; US Congress Office of Technology Assessment, 1995). Those laws codified and strengthened incentives for patenting inventions arising from federal funding (Mowery et al., 2004). Many other countries also developed incentives to stimulate technology transfer, and the nascent field of commercial genomics was greatly affected by these policies. Genomic technologies, some of them developed in private R&D laboratories, become applied in “downstream” biotechnology, namely, pharmaceutical, agricultural, or other R&D, either immediately or after a period of gestation in academic R&D. One interesting aspect of genomics has been the degree to which small firms have developed technologies far “upstream,” as tools for science, including academic science. Most genomic technologies are in fact research tools, and genomics as a field quickly grew large enough to sustain a market for DNA synthesis and sequencing machines, sample-handling robotics, microarrays, and other technologies. Academic science is a first home and the “opening market” for these technologies that may later develop into much broader application markets such as diagnostics, agriculture, environmental remediation, or energy production. For example firms such as Lynx, Illumina, Solexa, 454, Affymetrix, and many others have been started to develop technologies being used extensively in academic genomics laboratories. The market for genomic technologies is thus a complex hybrid of public and private R&D laboratories, and many technologies start in private R&D or are moved into private R&D
■
435
very early in their history, with an eye toward development for academic research, such that academic research laboratories, as users, are actually downstream from private R&D. This history gives genomics firms an unusual business model that involves both direct and indirect funding from government and non-profit sources. Federal grants and contracts are an important source of the money that drives this sector, occasionally through direct grants to the firms, but much more consistently and to a larger extent by sale of products and services to federally funded or non-profit laboratories. This model belies the simple pipeline model that generally sees federal and nonprofit basic science producing knowledge that is developed into end products and services through industrial R&D and then sold in private markets.
LANDSCAPE OF PRIVATE SECTOR GENOMICS In this section, we present financial and intellectual property data for private sector genomics firms through 2004. These figures provide a snapshot of privately funded genomics R&D and of some underlying trends in the financial inputs and scientific outputs of this sector. Defining the Genomics Sector What Is Genomics? The word “genome” is an irregular hybrid of the words “gene” and “chromosome,” a term coined in 1920 by Hans Winker decades before DNA was understood to be the repository of Mendel’s “particles of inheritance,” and well before the concept and application of genetic mapping and nucleic acid sequencing had become fully intertwined with efforts to understand development and disease (Lederberg and McCray, 2001). The origin of the term “genomics” can be traced to Tom Roderick, as first cited by McKusick and Ruddle in 1987, in the inaugural editorial for their new journal, Genomics. Genomics is a relative term referring to the study of genomes, coined to contrast with traditional genetics, which studied genes for traits and disorders, often focusing on Mendelian characters. At that time, genomics distinguished large-scale mapping and sequencing efforts from molecular studies of one or a few genes (Kuska, 1998). The term genomics gained popularity and came to describe a rapidly growing field of molecular biology, applying to large-scale, rapid DNA analysis, and intensive use of instruments and new technologies. Lederberg and McCray (2001) noted that, by 2001, “genomics” had acquired a broader meaning, referring to any study that involved the analysis of DNA sequence and even to the study of how genes affected biological mechanisms and phenotypes. This included the original meaning of genomics, but went well beyond it to almost any analysis that incorporated DNA technologies on a grand scale. What Is a Genomics Firm? The definition of a genomics firm is somewhat arbitrary. “Genomics” applies more to an approach than to an industry.
436
CHAPTER 37
■
Private Firm Genomics R&D in Genomic Medicine
The “biotechnology industry” is often, if somewhat misleadingly, used to refer to a subset of mainly pharmaceutical and agricultural firms that employ molecular biological methods as their R&D base. In commercial biotechnology, genomics came to refer to biotechnology firms that employed techniques that generated massive amounts of data about DNA of various kinds (e.g., DNA sequence information or chromosomal position data) in all or a substantial part of their work. Genomics was adopted as a “buzz word” to attract private capital, particularly from 1998 through mid-2001. After the financial market bubble burst in 2001, genomics went from being a desirable term to one used with some caution when trying to attract new capital. As a financial term, after 2001 it carried undesirable associations with the bubble it helped to create. Capital markets moved on to new terms associated with newer technologies such as synthetic biology, nanotechnology, interfering RNA, and other “hot” subfields of the
TABLE 37.1
day. It has become increasingly difficult to determine exactly what portion of a firm’s business is related to genomics. R&D allocations by firms on our list range from complete dedication to genomics to only a small, but important and crucial, fraction of R&D funds attributed to genomics. With this in mind, our dataset of genomics firms is a best effort estimation of private R&D in genomics, but should not be viewed as a precise valuation of how much R&D is truly “genomics”. A “genomics” taxonomy emerged from reviewing descriptions of R&D carried out by private firms. The firms were described by themselves, on websites or in annual reports, or by others in the trade press and news websites as “genomics firms.” The categories in the taxonomy are derived from these descriptions (Table 37.1). The categories are not mutually exclusive; each firm can be classified under multiple headings in the taxonomy.
Genomics firms taxonomy
AGRIVET
Agriculture and veterinary genomics
DATABSE
Database creation, subscription, or licensing
DNASEQN
DNA sequencing
DNATEST
DNA testing service, clinical or diagnostic screening service, test kit manufacturing
DRUGDEV
Drug, biologic, and vaccine development
GENEFNL
Gene function and functional genomics; characterization of genes and their products
GENEMAP
Gene mapping; linkage, association studies; SNP discovery, use and analysis
GENEPOP
Genetic epidemiology; population studies
GENETFR
Gene transfer and gene therapy; vectors for gene therapy
GENEXPR
Gene expression analysis; microarray analysis; analysis of siRNAs and other regulation element
IDNTFCN
DNA forensics, DNA identification service
INFRMTX
Bioinformatics for DNA analysis; data mining
INSTRMT
Instruments for DNA analysis
LEGLSVC
Legal services; privacy protection
PHRMGEN
Pharmacogenetics or pharmacogenomics
STANDRD
Setting standards, testing service benchmarks
SUPPLYR
Genomics reagents supplier; microarray manufacturer; service provider
TRSTFND
Trust fund or genomics capital source
SYNBIOL
Synthetic biology
Criteria for deciding if a firm is a “genomics” firm: 1. Analysis of DNA analysis or synthesis a core business (types of analyses classified above). 2. “Genomics” listed on website, annual report, or in news stories as part of business plan. 3. Listed as “genomics” by stock analysts or trade press (subject to correction if determined not to meet criteria 1 or 2 above). Firms may be assigned to one or more categories, and the categories are not mutually exclusive.
Landscape of Private Sector Genomics
Snapshot of the Genomics Sector – Financial and Market Factors The survey database spanned 26 years and included 470 genomics firms from 25 countries. Figure 37.1 shows the firms by designation in 2004 – private (211), public (88), non-profit (2), acquired (90), subsidiary (27), dissolved (23), and other (29). An industry snapshot for 2004 found the total global market capitalization for the 88 publicly traded firms to be around $41 billion, a 53% drop from the 2000 peak value of $88 billion (Figure 37.2). The genomics subsector employed over 25,000 people, invested $3.2 billion in R&D, and generated $6.3 billion in revenue during 2004. The top 15 firms by market value represented almost 70% of the total genomics sector’s value, approximately $28 billion. These top 15 firms spent a combined $2 billion in R&D, generated $4.3 billion in annual revenues, and employed over 15,000 people in 2004. Analysis of the top 15 firms demonstrated broad growth trends, indicating that R&D, capital improvement, and hiring continued to increase both before and
■
437
after the 1998–2001 bubble (Figure 37.3), despite the 2001 peak and post-2001 decline in market capitalization. These trends in the top 15 firms paralleled the consistent growth in total revenues for all publicly traded firms, despite significant fluctuations in market valuation (see later). The number of publicly traded firms grew to 88, reaching this peak in 2002 and subsequently stabilizing in 2003 and 2004. Steady state occurred after 2001, as some firms disappeared as they went out of business, merged, were acquired, while other firms were created (but far fewer per year than 1998–2001) (Figure 37.4). Business Plans and Applications in the Genomics Sector The most common categories of activity and representative examples from both public and private firms for each category $5.0 $4.5
Subsidiary 27 (6%) Acquired 90 (19%) Public 88 (19%)
USD in billions
$4.0 $3.5 $3.0
$2.0 $1.5 $1.0
Dissolved 23 (5%)
$0.5
Non-profit 2 (0.4%)
$0.0
Figure 37.1 Distribution of genomics firms by type in 2004. Genomics firms were classified into the types indicated above based on authors’ analysis of study data.
Total Revenue Total Exp PPE
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004
Other 29 (6%)
Private 211 (45%)
R&D Exp
$2.5
Years
Figure 37.3 Financial trends of top 15 genomics firms. Top 15 genomics firms by market capitalization based on authors’ analysis of study data are: Applera, Millennium Pharmaceuticals, Invitrogen, OSI Pharmaceuticals, Gen-Probe, Affymetrix, Protein Design Labs, Human Genome Sciences, ZymoGenetics, Abgenix, Incyte, Digene, Exelixis Pharmaceuticals, Lexicon Genetics, Rigel Pharmaceuticals. Total Exp, total expenditures; R&D Exp, research and development expenditures; PPE, plant, property, and equipment.
$100
100
$90 90 $80 80 70
$60
Number of firms
USD in billions
$70
$50 $40 $30
60 50 40
$20
30
$10
20
$0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Years
Figure 37.2 Aggregate market capitalization of all genomics firms. The aggregate market capitalization of all publicly traded genomics firms for the years 1990–2004 in US dollars based on authors’ analysis of study data.
10 0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Years
Figure 37.4 Aggregate number of public genomics firms. The aggregate number of publicly traded genomic firms for the years 1990–2004 based on authors’ analysis of study data.
438
CHAPTER 37
■
Private Firm Genomics R&D in Genomic Medicine
are illustrated in Table 37.2. One important point to glean from the taxonomy is that the genomics sector comprises firms with varying business plans, resulting in strategic heterogeneity despite similar core technologies. This fact is often glossed over when analysts describe “genomics” as a category, usually referring to early drug discovery using platform technologies, DNA sequencing, or just one or a few categories within the taxonomy. The most common category for both public and private firms was “drug, biologic and vaccine development” (55% of public firms; 33% of privately held firms). As illustrated in Table 37.2, nearly 25% of public firms are in the business of providing DNA analysis instruments (e.g., Affymetrix or Applied Biosystems). Another 20% of public firms supply genomics reagents including microarrays (e.g., Invitrogen) while almost a quarter of public firms provide services including DNA testing, clinical or diagnostic screening and test kit manufacturing (e.g., Gen-Probe). Almost 30% of private firms are involved in “bioinformatics for DNA analysis; data mining” (e.g., DNAStar) and over 20% conduct “gene expression analysis; microarray analysis; or analysis of siRNAs; and other regulation elements” (e.g., Ipsogen). A significant fraction of genomics companies (15%) focus on diagnostic applications and DNA testing services. Even among the early leaders in genomic drug development, there were many business plans and strategies for commercializing genomic discoveries. For example, Millennium sought
TABLE 37.2 firms
Top five categories in the taxonomy of
Public firms
Category DRUGDEV
Percenta 55
Representative firms Millennium, Incyte
INSTRMT
25
Gen-Probe, Affymetrix
SUPPLYR
23
Invitrogen, Affymetrix
DNATEST
22
Gen-Probe, Digene
GENEXPR
15
Exelixis, Diversa
Category
Percenta
Representative firms
DRUGDEV
33
AGY Therapeutics, Xenon
Private firms
INFRMTX
29
Genomatix, DNAStar
GENEXPR
21
Ipsogen, Ambion
DNATEST
15
Gentris, HandyLab
GENEFNL
14
Agilix, Xantos
Five most common categories of public and private firms by number of firms in each category. Firms can be classified by multiple taxonomies based on business function. a Percent of firms conducting research or business under a given category in the taxonomy.
a competitive advantage through genetic linkage analyses, largescale sequencing, and characterization of genes to link genomic information with disease biology in order to develop a targeted, novel pipeline of drug candidates (Millennium Pharmaceuticals, 2003). HGS and Incyte started by sequencing and characterizing full-length human genes, focusing especially on those with characteristics likely to make them promising drug targets (e.g., with sequence motifs predicting cellular transport, membrane or DNA binding potential). Both HGS (Human Genome Sciences, 2007) and Incyte (Incyte) have since augmented this approach by focusing on validation of drug targets, developing diagnostics, licensing-in promising drugs and biologics, and focusing on drug discovery. Celera Genomics initially focused on bioinformatics and producing reference genomic sequences (e.g., fruit fly, mouse, and human), under a business plan premised on selling subscriptions to the Celera database and informational tools. The business plan shifted to include drug discovery and a separate unit, Celera Diagnostics, aiming to develop diagnostics that identify and link genetic variation with disease was formed (Celera Genomics, 2007). Not all genomics companies are startups. Some firms moved into genomics from other lines of business. For instance, in 1993, the general science instrumentation firm Perkin-Elmer acquired the startup firm Applied Biosystems, whose primary business was instrumentation for analysis of DNA and proteins (for both synthesis and sequencing of nucleic acids and peptides). This line of business became the most profitable and fastest-growing component of Perkin-Elmer through the 1990s. After a series of name changes and re-organizations, what emerged was Applera, the parent firm of Applied Biosystems, Celera Genomics, and Celera Diagnostics, becoming one of the largest genomics firms. Output and Intellectual Property The majority of genomics firms were not profitable by the end of 2004; even those considered successful and ranking among the top 15 by market capitalization had an aggregate net income in 2004 of negative $1.2 billion. However, beginning in 2003 net income for the sector began an upward trend (i.e., an aggregate reduction in losses). Although many firms were not yet profitable at the end of 2004, total revenues for the genomics sector continued to climb, and in 2004 it generated approximately $6.3 billion in revenues, with a combined net income of negative $2.5 billion (Figure 37.5). Another measure of output for genomics firms is the number of patents issued. While patent counts are only rough indicators of productivity, and can even be considered crude measures, they are better than the tools available to measure output from other areas of research-intensive innovation, such as software, telecommunications, or computing. The patents associated with genomics and genetics can be identified more readily than for many other fields, providing an analytical tool to study genomic innovation. The many distinctive terms for DNA (and RNA) permit DNA patents to be identified with a relatively high degree of specificity and sensitivity. The claims of US patents that contain one or more terms specific to DNA
Landscape of Private Sector Genomics
(or RNA) can be used to capture intellectual property corresponding roughly to genetics and genomics, permitting analysis of trends. The intellectual property owned by currently active and independent genomics firms was 5859 total US patents and 3682 DNA-based US patents. Table 37.3 illustrates the top firms by total US and DNA patents owned. In 2004, the top 10 firms held nearly 60% of the total US patents and 64% of DNA patents. Among the top 10 DNA patent-holding firms, the percentage of their intellectual property portfolio attributed to DNA patents ranged from 44% to 93%. This ratio of DNA-based patents to total patents was a rough indicator of “genomics” intensity. Global Scope of the Genomics Private Sector The majority of publicly traded and privately held genomics firms in our database are in the United States; 75% (public) and 62% (private), respectively. Canada, Germany, France, United Kingdom, and Japan round out the top six countries, as shown in Table 37.4. These numbers should be interpreted with some caution because of the bias in favor of North America and Western Europe in our data sources. The database missed some firms even in the United States and is most likely missing even more from other countries, particularly outside North America and Europe. The heavy concentration of firm formation in North America and Europe, particularly the United States, reflects the origins of private genomics, which developed concurrently with the publicly funded HGP, based primarily in the regions where the publicly funded science was conducted. These trends also indicate the supportive venture capital environment in the United States and Europe. In recent years, leading private genomics firms from the United States and Europe have increasingly partnered in collaborative international genome sequencing projects focused on both agriculture and human disease, specifically, infectious diseases in the developing world. For example, primary sequencing $8
USD in billions
$6
Total Revenue
$4
$2
$0
Net Income
$2
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Years
Figure 37.5 Aggregate net income and total revenue for all genomics firms. Aggregate net income and total revenue of public and private genomics firms for the years 1990–2004 in US dollars based on authors’ analysis of study data.
439
for the Anopheles gambiae genome sequencing project (Holt et al., 2002) was performed by Celera Genomics (Acharya et al., 2004; Hoffman et al., 2002;WHO, 2002). The geographic base of genomics will continue to broaden as firms mature and as more TABLE 37.3 Intellectual property ownership of genomics firmsa Top 10 firms by total US patent countb Name
US patents
Incyte
779
Human Genome Sciences
461
Millennium Pharmaceuticals
457
Applera
380
ZymoGenetics
345
Affymetrix
289
Caliper Life Sciences
206
Gen-Probe
205
Invitrogen
195
Sirna Therapeutics
130
Top 10 firms by DNA patent countc Name
DNA patents
Genomics patent intensity (%)d
Incyte
561
72
Human Genome Sciences
390
85
Millennium Pharmaceuticals
302
66
Affymetrix
247
85
Applera
182
48
Gen-Probe
180
88
ZymoGenetics
151
44
Invitrogen
139
71
Sirna Therapeutics
121
93
94
86
Stratagene a
$4
■
Data on active firms only; excluding genomics firms designated as acquired, subsidiary, dissolved, or other. b Total US patent data obtained through searches for firm name as assignee using Delphion patent database as of February 7, 2006. c DNA patent data obtained through searches for firm name as assignee in DNA Patent Database (http://dnapatents.georgetown. edu) on January 11, 2006. d Percentage of DNA patents as a fraction of total US patents for each firm.
440
CHAPTER 37
■
Private Firm Genomics R&D in Genomic Medicine
countries develop the requisite R&D infrastructure and expertise that will stimulate investment (Reineke and Cook-Deegan – in preparation). Several countries, such as Singapore (Genome Institute of Singapore, 2007), Estonia (Estonian Genome Project, 2007), Taiwan (Harris, 2002; National Research Program for Genomic Medicine, 2007), China (Hui, 2000) South Korea, Mexico (Jimenez-Sanchez, 2003; Mothelet and Herrera, 2005; Rabinowicz, 2001), and India (Mudur, 2001), have explicit government programs to support genomics R&D. Iceland and Brazil have seen private institutions invest substantial resources in genomics. In Iceland, much of the investment came from for-profit ventures through deCODE Genetics Inc. (2007) and a few other companies (Hakonarson et al., 2003; Specter, 1999). In Brazil, genomics largely grew from a set of non-profit research projects conducted at public universities and at privately owned Ludwig Institute for Cancer Research in the State of Sao Paolo (Camargo and Simpson, 2003; Simpson, 2001; Simpson and Perez, 1998). A committed investment in genomics R&D has produced strong scientific output from nations such as Brazil (Chandrasekharan and Fric-Shamji – in preparation), Iceland (Grant et al., 2006; Helgadottir et al., 2007), and Singapore (Ruan et al., 2003). India (Jayaraman and Louet, 2004) and China (Qiang, 2004) have been intensively building capacity for genomics R&D and bioinformatics through publicly funded initiatives, both as a tool for economic development, by cultivating a talent pool and attracting foreign investment, and for improving national health. Much like the trend observed in North America and Western Europe, a commercial genomics sector is growing up alongside publicly funded genomics science programs, with the founding of startup companies, some as spin-offs of publicly funded efforts. In contrast to private venture capital in the United States, however, much of the venture capital in India comes from national or state government (Sunder Rajan, 2006). This indicates increasing intellectual property consciousness in these countries. Similar to the history of genomics in the United States, larger biotech/pharmaceutical firms (e.g., firms involved in manufacture of generic drugs and vaccines) are also increasingly adopting or integrating genomic technologies in their R&D. Commercial genomics was an offshoot of the publicly funded HGP and has coalesced into a set of technologies used TABLE 37.4
Top countries with genomics firms
Country
Public
Private
Non-profit
Total
65
130
1
196
Canada
6
17
1
24
Germany
6
15
21
France
0
10
10
United Kingdom
3
6
9
Japan
0
5
5
United States
for diverse purposes in biomedical research and other fields. Genomics has evolved from purely non-profit and government research into a hybrid public–private R&D enterprise. While private sector genomics was practically non-existent in 1990, by 2000, private genomics R&D expenditure exceeded public and non-profit funding. Revenues of genomics firms rose to over $6.3 billion in 2004, $3.2 billion of which was expended on R&D. As indicated by our taxonomy of firm business strategies, most of this investment was for medical applications such as drug, biologics, and vaccine development, as opposed to agriculture and veterinary genomics or other areas. Genomic medicine was thus clearly the dominant end-target for genomics firms. The genomics subsector of biotechnology is not profitable, although the size of net losses diminished from 2002 to 2004. A long period of aggregate net loss has a precedent in biotechnology more generally. It took the better part of two decades for Amgen, Genentech, Biogen, Genzyme, and other “first generation” biotechnology firms to become established as a profitable subsector of the pharmaceutical industry. If this pattern holds true for genomics firms, some may become profitable over the next decade. In fact, several firms are already showing positive net incomes – 15 public firms reported a positive net income in 2004: Affymetrix, Applera, Digene, Discovery Partners International, Eurogentec, Genetix, Gen-Probe, Harvard Bioscience, Insightful, Invitrogen, Maxygen, Molecular Devices, Morphosys, Stratagene, and Tripos (Chandrasekharan et al. – in preparation). Many of the “platform technology” companies, which sell tools used in conducting R&D, can be expected to be profitable before other companies with longer product cycles, such as development of therapeutics, biologics, or vaccines. Private sector genomics R&D interacts with publicly funded genomics research in highly complex ways that do not entirely comport with traditional frameworks of R&D that assume federal and non-profit funding for “basic” research will produce knowledge that is applied downstream by private firms. Academic genomic research now entails the use of “research tools” purchased from private genomics firms, the majority of which were created after the HGP began. This is also reflected by increasing private partnership in large-scale public genomic projects such as the Genetic Association Identification Network (GAIN, 2007), a partnership between government funders, academic centers, and commercial firms including Pfizer, Affymetrix, and Perlegen Sciences. The public–private Encyclopedia of DNA Elements Project is another prominent example (ENCODE Project Consortium, 2004, 2007). Private sector firms are significant contributors to innovation in genomics. One such area of commercial R&D is the development of new technologies for cheaper, faster DNA sequencing, commonly referred to as “Next-Gen Sequencing.” Sequencing technology innovation is being driven by a complex and often collaborative relationship between private and public research efforts. These technologies are likely to have a significant impact on genomic medicine by paving the way for and possibly making whole-genome sequencing of individuals a routine medical procedure.
Future Trends
A number of firms dedicated to Next-Gen sequencing technology have emerged in the past few years and have received significant funding from federal sources including the NHGRI (National Human Genome Research Institute [2007], Genome Technology Program, Prasad S. and Chandrasekharan S. – unpublished data). These companies have received much industry interest and media attention, particularly in the context of the “race for the $1000 genome” (Service, 2006). Companies with more mature sequencing technologies, such as 454 Life Sciences (recently acquired by Roche Life Sciences) and Solexa (acquired by Illumina), have increasingly partnered with academic institutions to support genome sequencing projects for several organisms and, in return, have academic laboratories test and validate their technologies. For example, 454 Life Sciences has entered collaborative research agreements with the Broad Institute of MIT and Harvard for “Ultra-deep Sequencing” of specific genes of interest involved in complex diseases, such as cancer, diabetes, and heart disease. Similarly 454 has collaborated with the Max Planck Institute (Neanderthal Genome Sequencing), Yale University (HIV variants sequencing), and with the Malaysia Genome Resource Center (454 Life Sciences, 2007). Advances in conventional and Next-Gen sequencing have dramatically reduced the cost and time of sequencing and moved the goal of personal genome sequencing within reach, illustrated by the recent sequencing of the diploid genomes of geneticists, James Watson, by a 454 and Baylor College of Medicine partnership (454 Life Sciences, 2007; Wade, 2007) and J. Craig Venter’s sequence compiled by the J Craig Venter Institute (Levy et al., 2007). Individual whole-genome sequencing projects such as The Personal Genome Project (Church, 2005) headed by George Church at Harvard and the ClinSeq project at NHGRI, that aims to sequence 1000 individual genomes (ClinSeq, 2007) are also enabled by Next-Gen sequencing technologies developed in the private sector. These projects presage much more widespread use of these technologies. The ultimate markets may be clinical genetic testing, or industrial applications in pharmaceutical R&D and other areas, but academic research constitutes a market critical to the early history of these new technologies. Similarly, high-throughput genotyping is progressing rapidly, along with bioinformatics databases, analytical algorithms, analytical software, and other tools for creating and analyzing genomic data. Platform technologies for high-throughput SNP genotyping developed by companies such as Illumina and Affymetrix were integral to the International HapMap project and genome-wide association studies. Furthermore, products of such platform technologies, enabling rapid and extensive genotyping with several thousands of SNPs, have greatly expedited research aimed at identifying susceptibility loci for diseases such as macular degeneration (Hirschhorn and Daly, 2005) and several common diseases such as diabetes Type I and II, hypertension, coronary artery disease, bipolar disease rheumatoid arthritis, using genome-wide association approaches, as reported by the Wellcome Trust Case Control Consortium (2007). Foreseeable applications of such research include products that can facilitate early diagnosis, preventive interventions, and more individualized
■
441
therapies. As detailed in other chapters in this book, genomic technologies are expected to make a large impact on pharmacogenomics and pharmacogenetics, where greater understanding of polymorphisms in genes responsible for drug metabolism may be translated into medical practice by using diagnostics to identify these genetic variations and help tailor pharmacotherapy in terms of drug selection and dosage (Need et al., 2005). Large-scale expression profiling, particularly of human tumors, has relied heavily on microarray platform technologies (primarily from Affymetrix and Agilent). This research has yielded a wealth of information, enhancing our understanding of the molecular underpinnings of different cancers. More importantly, the molecular signatures of different tumor types revealed by expression profiling have also allowed for reliable predictions about individuals’ response to chemotherapeutic agents or their risk of tumor recurrence (Potti et al., 2006), suggesting that genomic profiling will be incorporated into clinical decision trees to guide cancer diagnosis and treatment. The impact of genomic technologies in the emergence of actual products that can be integrated into medical practice is illustrated by the introduction of two in vitro diagnostics, Amplichip CYP450 from Roche Molecular Diagnostics and Mammaprint™ developed by Agendia. AmpliChip CYP450 (AmpliChip® CYP450 Test, 2007), which is based on Affymetrix microarray technology, detects polymorphisms and mutations in two CYP450 genes, CYP2D6 and CYP2C19, that are involved in the metabolism of a variety of prescription drugs. Introduced into the US market in 2003 (and 2004 in the EU), AmpliChip was the first FDA-approved pharmacogenetic test, which by allowing the physician to predict the drug metabolism phenotype of the patient, promised to improve patient outcome by increasing drug efficacy and reducing adverse reactions (AmpliChip®). Mammaprint™ was the first commercially available microarray-based cancer in vitro prognostic test. Based on research from the Netherlands Cancer Institute and the Antoni Van Leeuwenhoek Hospital in Amsterdam, the test determines the mRNA expression profile of 70 genes in breast tumor tissue and predicts the risk for breast tumor metastasis and recurrence within 10 years for the patient (Glas et al., 2006). The test, which uses custom microarrays manufactured by Agilent, was commercialized by Netherlands-based Agendia and received FDA approval in February 2007 (Food and Drug Administration, 2007; Mammaprint™, 2007). This was the first FDA-approved “in vitro diagnostic multivariate index assay” (IVDMIA) device and was seen as the first of several expression profiling-based diagnostic and prognostic tests that will enter the market.
FUTURE TRENDS The commercial genomics sector is a key stakeholder in genomic and personalized medicine. It accounts for a significant fraction of genomics research investment, as supported by our data showing high R&D expenditure by genomics firms.
442
CHAPTER 37
■
Private Firm Genomics R&D in Genomic Medicine
It promises to be a major player in clinical applications, as illustrated by the strong representation of private sector firms with a primary or significant business interest in genomics in both the Personalized Medicine Coalition (see http://www.personalizedmedicinecoalition.org/) and in the Coalition for 21st Century Medicine in the United States (Coalition for 21st Century Medicine, 2007). While the commercial development of genomic diagnostics appears poised for growth, new drugs, biologics, and vaccines developed from genomic information obtained by whole-genome sequencing and functional genomics approaches are still scarce. The identification of genetic variants associated with common diseases such as hypertension and diabetes may fuel further growth in custom diagnostic tests and testing services. Similarly, data from pharmacogenetic/pharmacogenomic and gene-expression profiling studies should help inform and design “individualized” drug regimens. These new application areas are likely to spur a further increase in the number of companies developing and providing custom pharmacogenetic and prognostic tests. Given the long products cycle for drugs and biologics, the coming decade may also see the first of personalized therapies for common diseases, particularly as advances are made in understanding the functions of disease susceptibility genes identified from genome-wide association studies and the molecular pathways they illuminate. “Personal Genomics” companies that use a direct to consumer model, such as Navigenics and 23andMe Inc., are also likely to proliferate. The market here is individuals who seek information about their own DNA, either through extensive genotyping or whole-genome sequencing of their DNA. This business model avails itself not only of genomic technologies, but depends crucially on the Internet, with its capacious data-transmission and individualized communication features.
Genome sequence analyses of various pathogens, combined with identification of human (and veterinary) host genetic variation that alters susceptibility, will likely result in the development of novel high-throughput tests for diagnosis and management of infectious diseases. Such tests will potentially be used in combination with specific therapeutics (drugs or biologics) or prevention strategies (e.g., vaccines) that are themselves products of functional genomics-based target discovery and drug development. The systematic integration of genomics with other “omics” and computational tools for the analysis and interpretation of information will be an essential next step in realizing the promise of genomic medicine and its implementation in health care. This is therefore likely to translate into a growth of companies offering solutions for genomic information storage, processing and integration for specific applications, and meeting the needs of individuals. Private sector genomics in both the developed and developing world is likely to evolve in unexpected ways to the meet the growing need for high-throughput and automated technologies for basic research and remain an integral player in the emergence of genomic medicine for addressing global health needs in the coming decades.
ACKNOWLEDGEMENTS The authors would like to thank the following students and research assistants at the Center for Genome Ethics, Law & Policy for invaluable help with data gathering and entry: Phebe Ko, Suparna Salil, Whitney Laemmli, Joe Fore, Alessandra Colaianni, Daidree Tofano, Molly Nicholson, Stacy Lavin, Britt Rusert, Marjorie Gurganus, Anupama Kotha, Cindy Wang, and Nancy Wang.
REFERENCES 23andMe Inc. (2007). http://23andme.com, accessed October 29, 2007. 454 Life Sciences (2007). Press release http://www.454.com/newsevents/press-releases.asp, accessed October 29, 2007. Acharya, T. and Daar, A.S. et al. (2004). Strengthening the role of genomics in global health. PLoS Med 1(3), e40. AmpliChip® CYP450 Test. (2007). http://www.amplichip.us/accessed/ September 2007. Camargo, A.A. and Simpson, A.J. (2003). Collaborative research networks work. J Clin Invest 112, 468–471. Celera Genomics (2007). Celera Genomics: Targeted Medicine http:// www.celera.com/celera/targeted_medicine, accessed September 2007. Chandrasekharan, S. and Fric-Shamji, E. Metrics of Genomics and Bioinformatics R&D activity – A case study of Brazil, India and Singapore – in preparation. Chandrasekharan, S., Perin, N.C., Ilse, R., Weichers, I.R. and Cook– Deegan, R.M. The landscape of commercial genomics – in preparation.
Church, G.M. (2005). The personal genome project. Mol Syst Biol 1, 0030. ClinSeq (2007). A Large-Scale Medical Sequencing Clinical Research Pilot Study. http://www.genome.gov/20519355, accessed October 29, 2007. Coalition for 21st Century Medicine (2007). About us. http:// www.twentyfirstcenturymedicine.org/about.html, accessed 10 September 2007. Cook-Deegan, R. (1994a). The Gene Wars: Science, Politics, and The Human Genome.WW Norton, New York. Cook-Deegan, R. (1994b). Survey of Genome Science Corporations, Contract report for the Office of Technology Assessment, US Congress; cited in Tables 1–5, The Human Genome Project and Patenting DNA Sequences, unpublished draft report approved by the Technology Assessment Board http://www.kie.georgetown.edu/ nrcbl/documents/dnapatents/OTAdraft.pdf, accessed 13 September 2007. Cook-Deegan, R., Chan, C. and Johnson, A. (2001). World Survey of Funding for Genomics Research: Final Report to the Global
References
Forum for Health Research and the World Health Organization. Journal of Biolaw and Business Special Supplement, 22–36. DeCODE Genetics Diagnostics Inc. (2007). http://www.decode.com, accessed October 29, 2007. ENCODE Project Consortium (2004). The ENCODE (ENCyclopedia Of DNA Elements) Project. Science 306, 636–640. http://www. genome.gov/10005107, accessed October 29, 2007. ENCODE Project Consortium (2007). Identification and Analysis of Functional Elements in 1% of the Human Genome by the ENCODE Pilot Project. Nature 447, 799–816. Estonian Genome Project (2007). http://www.geenivaramu.ee/index. php?lang eng&sub 58, accessed October 29, 2007. Food and Drug Administration. (2007). FDA Clears Breast Cancer Specific Molecular Prognostic Test, February 6, 2007, http://www. fda.gov/bbs/topics/NEWS/2007/NEW01555.html, accessed September 2007. GAIN (2007). The Genetic Association Information Network http:// www.fnih.org/GAIN2/gain_press_release.shtml, accessed October 29, 2007. Glas, A.M. et al. (2006). Converting a breast cancer microarray signature into a high-throughput diagnostic test. BMC Genomics 7, 278. Genome Institute of Singapore. (2007). http://www.gis.a-star.edu.sg/ internet/site/about_us.php?fvision_mission, accessed October 29, 2007. Grant, S.F. et al. (2006). Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes. Nat Genet 8(3), 320–323. Harris, S. (2002). Biotechnology: The next engine of growth for Taiwan’s economy? EMBO Reports 3(7), 598–600. Hakonarson, H., Gulcher, J.R. and Stefansson, K. (2003). DeCODE Genetics, Inc. Pharmacogenomics 4(2), 209–215. Helgadottir, A. et al. (2007). A common variant on chromosome 9p21 affects the risk of myocardial infarction. Science 316(5830), 1491–1493. Hirschhorn, J.N. and Daly, M.J. (2005). Genome wide association studies for common diseases and complex traits. Nat Rev Genet 6, 95–108. Hoffman, S.L. et al. (2002). Plasmodium, human and Anopheles genomics and malaria. Nature 415, 702–709. Holt, R.A. et al. (2002). The genome sequence of the malaria mosquito Anopheles gambiae. Science 298(5591), 129–149. Hui, L. (2000). Genomics. Money and machines fuel China’s push in sequencing. Science 288(5467), 795–798. Human Genome Sciences (2007). Technology. www.hgsi.com/ technology/index.html, accessed September 13, 2007. Jayaraman, K.S. and Louet, S. (2004). India aims to become the main bioinformatics hub. Nat Biotech 22(8), 933. Jimenez-Sanchez, G. (2003). Developing a platform for genomic medicine in Mexico. Science 300, 295–296. Kuska, B. (1998). Beer, Bethesda, and Biology: How “Genomics” Came Into Being. J Natl Cancer Inst 90(2), 93. Lederberg, J. and McCray, A.T. (2001). Ome Sweet Omics – A Genalogical Treasury of Words. Scientist 15(7), 8. Levy, S., Sutton, G., Ng, P.C., Feuk, L. and Halpern, A.L. et al. (2007). The diploid genome sequence of an individual human. PLoS Biology 5(10), e254. Mammaprint™ (2007). http://www.agendia.com/en/Professional/ About-MammaPrint/About-MammaPrint, accessed September 2007.
■
443
McKusick, V.A. and Ruddle, F.H. (1987). A New Discipline, A New Name, A New Journal. Genomics 1, 1–2. Millennium Pharmaceuticals (2003). Millennium: News Release, 27 October 2003, http://investor.millennium.com/phoenix. zhtml?c80159&pirol-newsArticle_Print&ID462997&highli ght accessed Sep 2007. Moses, H. et al. (2005). Financial Anatomy of Biomedical Research. JAMA 294(11), 1333–1342. Mothelet, V.G. and Herrera, S. (2005). Mexico launches bold genome project. Nat Biotechnol 9, 1030. Mowery, D., Nelson, R., Sampat, B. and Ziedonis, A. (2004). Ivory Tower and Industrial Innovation: University-Industry Technology Transfer Before and After the Bayh-Dole Act. Stanford Business Press, Palo Alto, CA. Mudur, G. (2001). India invests heavily in Genomics Research. BMJ 322(7286), 576. National Human Genome Research Institute. (2007). Genome Technology Program http://www.genome.gov/10000368, accessed October 29, 2007. National Research Council (1988). Mapping and Sequencing the Human Genome. National Academy Press,Washington, DC. National Research Program for Genomic Medicine, Taiwan (2007). http://nrpgm.sinica.edu.tw/en/content.php?catagtc, accessed October 29, 2007. National Science Foundation (2005). National Patterns of Research and Development Resources: 2003, NSF 05–308, February 2005, Division of Science Resources Statistics, National Science Foundation, Arlington, VA http://www.nsf.gov/statistics/ nsf05308/pdf/nsf05308.pdf, accessed October 29, 2007. Need, A.C., Motulsky, A.G. and Goldstein, D.B. (2005). Priorities and standards in pharmacogenetic research. Nat Genet 37(7), 671–681. Potti, A. et al. (2006). Genomic signatures to guide the use of chemotherapeutics. Nat Med 12(11), 1294–1300. Qiang, B. (2004). Human Genome Research in China. J Mol Med 82(4), 214–222. Rabinowicz, P.D. (2001). Genomics in Latin America: Reaching the frontiers. Genome Res 3, 319–322. Reineke-Pohlhaus, J. and Cook-Deegan, R.M. Genomics Research: A World Survey of Public Funding – in preparation. Ruan, Y.J. et al. (2003). Comparative full-length genome sequence analysis of 14 SARS coronavirus isolates and common mutations associated with putative origins of infection. Lancet 361, 1779–1785. Rudolph, L. (1995). Overview of Federal Technology Transfer. Risk 5, 133–142. Service, R.F. (2006). The race for the $1000 genome. Science 311, 1544–1546. Simpson, A.J. (2001). Genome sequencing networks. Nat Rev Genet 2, 979–983. Simpson, A.J. and Perez, J.F. (1998). ONSA, the Sao Paulo Virtual Genomics Institute. Organization for nucleotide sequencing and analysis. Nat Biotechnol 16(9), 795–796. Specter, M. (1999). Decoding Iceland: The next big medical breakthroughs may result from one scientist’s battle to map the Viking gene pool. New Yorker 74(42), 40. 42–46, 48–51. Sunder Rajan, K. (2006). Biocapital: The Constitution of Postgenomic Life. Duke University Press, Durham, NC. US Congress Office of Technology Assessment (1988). Mapping Our Genes – The Genome Projects: How Big? How Fast?. Government Printing Office.,Washington, DC.
444
CHAPTER 37
■
Private Firm Genomics R&D in Genomic Medicine
US Congress Office of Technology Assessment. (1995). Federal Technology Transfer and the Human Genome Project (OTA-BP-EHR162, Government Printing Office stock # 052–003-01432-7; appendix A) http://www.wws.princeton.edu/ota/ disk1/1995/ 9526/952608.PDF. Wade, N. (2007). Genome of DNA Discoverer Is Deciphered. New York Times Published online on June 1, 2007 at http://www.nytimes. com/2007/06/01/science/01gene.html, accessed October 29, 2007.
Wellcome Trust Case Control Consortium (2007). Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 447, 661–678. World Health Organization (2002). Genomics and World Health. World Health Organization, Geneva. http://whqlibdoc.who.int/ hq/2002/a74580.pdf, accessed October 29, 2007.
Section
Genomic Medicine and Public Health
6
38. What Is Public Health Genomics? 39. Why Do We Need Public Health in the Era of Genomic Medicine? 40. Principles of Human Genome Epidemiology 41. Genomics and Population Screening: Example of Newborn Screening 42. Family History: A Bridge Between Genomic Medicine and Disease Prevention
CHAPTER
38 What Is Public Health Genomics? Alison Stewart and Ron Zimmern
INTRODUCTION In the mid-1990s, as the Human Genome Project gathered pace, some within the profession of public health began to realize that, in time, new knowledge and technologies stemming from this endeavor would have profound implications for the health of populations and the organization and delivery of health care services. These developments would mean that public health could no longer focus exclusively on social and environmental determinants of health, but would need to incorporate relevant knowledge from the molecular and cellular life sciences in the development of public health policies and programs. Public health genomics is a new discipline that brings together genetic and genomic science, genetic epidemiology, and a recognition that scientific and technological advance must go hand in hand with an understanding of its ethical, legal, and social dimensions. It is distinct from “genomic medicine” in that its focus is primarily on populations, health services, and public health programs rather than on individual clinical care. This chapter outlines the development of public health genomics over the last decade, sets out its key concepts, and describes examples of public health genomics programs and activities. New national, regional, and international networks are being established to further the goals of this new field.
THE EMERGENCE OF PUBLIC HEALTH GENOMICS Public health genomics, which began life as “public health genetics,” has its origins in the public health communities of the Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 446
United Kingdom and the United States. In the United States, publication of a seminal paper by Khoury (1996) on the application of genetics in disease prevention was followed by the establishment of a Task Force on this topic by the Centers for Disease Control (CDC) and Prevention in Atlanta. The Task Force’s report (Centers for Disease Control and Prevention, 1997) led to the setting up of the Office of Genetics and Disease Prevention under Khoury’s leadership in 1997. At around the same time, Schools of Public Health at some US universities were beginning to recognize the need to ensure that public health professionals understood the implications of genetics for public health. Multi-disciplinary post-graduate courses, involving input from disciplines including law, medicine, arts and science, pharmacy, nursing and public policy, were developed at the Universities of Michigan and Washington. In the United Kingdom, an expert advisory group reporting to the Department of Health in 1995 made the point that the country’s National Health Service could not ignore the potential implications of the “new genetics” for health services (NHS Central Research and Development Committee, 1995). The first move toward bringing genetics into the realm of public health came in 1997, when Ron Zimmern set up the Public Health Genetics Unit in Cambridge. During the last few years of the 20th century, the principles of public health genetics became established on both sides of the Atlantic. A growing body of literature setting out the intellectual underpinnings of the field was accompanied by the development of programs and activities in areas such as human genome epidemiology, service development, policy analysis, and education and training of health professionals. Copyright © 2009, Elsevier Inc. All rights reserved.
Key Concepts in Public Health Genomics
There was also recognition that the scope of the field was broader than that implied by the word “genetics.” As well as heritable disease, it included knowledge about the structure and composition of the human genome, the properties of genes and their products, and how genes function together during the development and life of the organisms – in other words, all of modern molecular and cell biology. For this reason, there has been a trend toward using the term “genomics” rather than “genetics” (Khoury, 2003). CDC’s Office of Genetics and Disease Prevention became first the Office of Genomics and Disease Prevention and now the National Office of Public Health Genomics, while in 2007 Cambridge’s Public Health Genetics Unit became the Foundation for Genomics and Population Health. Interest in public health genomics has also grown in other countries of Europe, North America, Asia and Australasia. In 2005, at an international workshop held in Bellagio, Italy, agreement was reached to set up an international network that brings together all those involved in furthering the aims of public health genomics (Burke et al., 2006; Stewart, 2006). The Genomebased Research and Population Health International Network (GRaPH Int) was launched in June 2006, with an administrative hub hosted by the Public Health Agency of Canada.
THE DEFINITION OF PUBLIC HEALTH GENOMICS The early practitioners of public health genetics suggested a variety of definitions for their discipline. At the University of Washington Institute of Public Health Genetics it was defined as “The application of advances in human genetics and molecular biotechnology to improve public health and prevent disease,” while in the United Kingdom a modified form of the Acheson definition of public health was adopted, describing public health genetics as “The application of genetics on the art and science of promoting health and preventing disease through the organized efforts of society.” All of the early definitions emphasized that the subject matter of public health genetics was how the health of populations and the practice of public health and clinical medicine were affected by genetic science and technology. They also highlighted the importance of harnessing genetic knowledge for disease prevention; the word “prevention” in this context included not just measures to prevent or delay disease onset (primary prevention), but also, if disease was already present, interventions that enabled early detection and treatment, reduced disability, and delayed progression (secondary and tertiary prevention). The UK definition also made it clear that practitioners of public health genetics needed, as well as a working knowledge of genetic science, an understanding of the diverse views within society about genetics and its consequences, and an ability to work across a range of disciplines and cultures. As the field broadened its outlook beyond the traditional connotations of “genetics” and began to think in terms of genomics, a new definition became necessary. As a result of the 2005 Bellagio workshop, public health genomics is now defined
■
447
as: “The responsible and effective translation of genome-based knowledge and technologies for the benefit of population health” (Burke et al., 2006; Stewart, 2006). The Bellagio definition makes a number of important points. By using the term “genome-based” it indicates that the scope of the knowledge base includes not only genes but also their protein products, the metabolites synthesized by those proteins, and the complex molecular and cellular interactions that make up a biological system. It also includes bioinformatics: the management and analysis of large amounts of biological information using advanced computing techniques. “Technologies” are mentioned explicitly in the definition because biotechnological developments such as drug delivery platforms, highthroughput sequencing technologies, and microarray applications are vital to the clinical exploitation of genomic science. The adjectives “responsible” and “effective” convey two concepts: firstly, that any clinical tests or interventions resulting from genomic research must be fully validated before they are implemented and, second, that this evidence base must include due consideration of their ethical, legal, and social implications (often shortened to the acronym ELSI) (Clayton, 2003; Ojha and Thertullien, 2005). Finally, the Bellagio definition emphasizes that public health genomics focuses on the health and health care of populations. This may at first sight seem to be at odds with the individual or familial nature of genomic information and the prospects for a future era of “personalized medicine.” However, public health genomics recognizes that populations are not genetically homogeneous and that programs and policies incorporating differences in individual susceptibility to disease and response to treatment offer new opportunities that are complementary to the traditional “one size fits all” approach of public health.
KEY CONCEPTS IN PUBLIC HEALTH GENOMICS Genes and Environmental Factors as Determinants of Health Implicit in the various definitions of public health genomics is the recognition that all human characteristics – including susceptibility to disease – result from the combined effects of genes and environment. Figure 38.1 is a conceptual representation of determinants of health, showing the complex array of possible interactions between genomic determinants and the components of the physical and social environment. The relative contributions of genes and environment to disease risk can vary widely. In the case of highly penetrant heritable conditions such as cystic fibrosis or Huntington’s disease, a single genetic mutation may be sufficient to cause disease, but even for these diseases the range and severity of symptoms may vary widely among individuals, partly as a result of environmental factors. In the case of common chronic diseases, where a variety of low-penetrance genomic variants are implicated, the effect of environment is generally much more evident.
448
CHAPTER 38
■
What Is Public Health Genomics?
GENETIC
Genetic endowment
PHYSICAL
BIOLOGICAL
POPULATIONS
INDIVIDUALS SOCIAL
POLITICAL
Individual behavior
Natural environment
Structural environment
BEHAVIORAL
Figure 38.1 Determinants of health. The figure makes the point that external factors such as natural environmental and structural determinants are modifiable by interventions at the population level, whereas genetic factors and behavior are essentially individual (although they are, of course, influenced by external factors).
It is a mistake, however, to think of any disease as resulting from a fixed and quantifiable combination of genomic and environmental determinants. For example, phenylketonuria, a condition inherited in a Mendelian fashion, is classified as a genetic (heritable) disease because the environmental determinant (dietary phenylalanine) is ubiquitous and the mutation rare and highly penetrant. However, the disease could also be thought of as 100% environmental, because it is only manifest when phenylalanine is present. We might also characterize the situation by stating that the genomic defect is a necessary factor in the pathogenesis of phenylketonuria but not in itself sufficient. From the representation in Figure 38.1 it is clear that the health status of an individual may be influenced by altering either genomic or environmental factors, or both. In the context of disease prevention, Juengst (1995) distinguished these two types of intervention as “genotypic” and “phenotypic.” Genotypic intervention may be appropriate in some circumstances, such as the use of genetic testing to avoid the birth of a child affected by a highly penetrant genetic disease. However, in the context of common chronic disease, which does not usually develop until some time during adult life, only phenotypic intervention will generally be either feasible or ethically acceptable. Public health genomics aims to prepare the ground for a future era when we have a much more complete understanding of the full range of genomic and environmental determinants of health and – crucially – how they work together. The hope is that this fuller understanding of disease etiology and of the mechanisms of disease at the molecular and cellular levels will enable development of new diagnostic tools, new therapies, and perhaps new preventive options (Guttmacher and Collins, 2005), and that this new era of genomic medicine will offer opportunities to achieve benefits for population health. Enthusiasm about the potential benefits from genomic research must be balanced by realism about the likely time scale of these developments (Davey Smith et al., 2005; Haga et al., 2003). Unraveling the genomic contribution to disease susceptibility is an immensely complex task, and many applications in mainstream
health care are unlikely to materialize for some decades to come. In the meantime, it is a key function of public health genomics to protect patients and health services from the damaging effects of interventions that are premature, not supported by robust evidence or do not fulfill the criterion of public acceptability. Avoiding Genetic Exceptionalism It is important to avoid “hype” about advances in genomics and it is equally important to dispel the notion that genetic or genomic factors have a significance and predictive power beyond that of other determinants of health, a view that has been dubbed “genetic exceptionalism” (Murray, 1997). A related pitfall is genetic reductionism, the tendency to over-simplify the relationship between a genetic factor and a disease or other phenotypic trait (Sankar, 2003). Genetic exceptionalism and reductionism demand special protection for genomic information on the grounds of its predictive power and implications for other family members but this thinking derives from a misguided attempt to extrapolate from the highly penetrant mutations associated with Mendelian disease to the much more weakly predictive variants associated with common disease. Public health genomics recognizes that genes must be included among the determinants of health but their significance should be kept in proportion: they should neither be over-emphasized nor demonized. Genetic exceptionalism can lead to policy responses that are both disproportionate and irrational, in that they place restrictions on uses of DNA test information but leave other physiological or biomarker tests – which may be equally or even more highly predictive – unregulated.
THE “ENTERPRISE” OF PUBLIC HEALTH GENOMICS The Bellagio workshop participants developed a visual representation of the “enterprise” of public health genomics, which is represented by the shaded regions of Figure 38.2; the aim of
Core Activities in Public Health Genomics
■
449
Society Communication and Stakeholder Engagement Knowledge generation Population Sciences
Genome-based Science and Technology
Informing Public Policy Knowledge integration
Strategy
Analysis
within and across disciplines
Action
Improvement in Population Health
Evaluation
Developing and Evaluating Health Services Humanities and Social Sciences Education and Training
Applied and Translational Research
Figure 38.2 Strategy for the effective translation of genome-based knowledge and technologies for the benefit of population health. The shaded areas represent the “enterprise” of public health genomics.
this representation is to convey a consensus understanding of what sorts of activities come under the banner of public health genomics, and how its work is conducted. The starting point for the “enterprise” (but not part of the enterprise itself) is genomics research, shown at the left of Figure 38.2. Basic research is also needed in the population sciences (including, e.g., epidemiology and biostatistics) and in the humanities and social sciences (including law, philosophy, social anthropology, theology, and so on). Basic research can be thought of as the phase of knowledge generation. Public health genomics comes into play with the phase of knowledge integration, both within and across disciplines. The raw material for this process is the output from the research phase. Knowledge integration is defined as the process of selecting, storing, collating, analyzing, integrating, and disseminating information. This includes computer-based approaches to classifying, linking, and retrieving information (essentially, what an informatician would understand by the term “knowledge integration”) but it is also a broader and more qualitative process encompassing selection, critical analysis, and synthesis of concepts and information from many different fields in the arts, humanities, biological, and social sciences. As the means by which information is transformed into knowledge, it is an essential step in making that information usable in the practical world of clinical medicine and public health practice. The knowledge base generated by the knowledge integration function of public health genomics is used to underpin four core sets of activities: communication and stakeholder engagement; informing public policy; developing and evaluating health services; and education and training. Examples of some of these activities will be discussed later in this chapter.
Knowledge integration and the four core sets of activities set out what public health genomics does. How it carries out its work may be described by the cycle of public health action developed by the Institute of Medicine Committee for the Study of the Future of Public Health (1998) and others. This cycle consists of an initial phase of analysis followed by development of a strategy, implementation of that strategy, and evaluation of its outcomes. This is not a once-and-for-all process: successive rounds of the cycle lead to continued refinement of the strategy, both before and after implementation. Research does have a place in the enterprise of public health genomics, as shown at the bottom of the diagram. Distinct from basic academic research, this is applied and translational research with direct applicability to health service implementation; such research can also identify gaps in the knowledge base that need to be filled by further basic research. Finally, Figure 38.2 emphasizes that the whole enterprise of public health genomics – and the phase of basic research – is embedded within a societal context. The views and priorities of society play a part in determining what research is carried out and how it is used and, in turn, the perceived results of those activities shape societal attitudes.
CORE ACTIVITIES IN PUBLIC HEALTH GENOMICS Knowledge Integration Knowledge integration can be thought of as the driving force of public health genomics. A number of initiatives to support knowledge integration are underway at various centers. These activities
450
CHAPTER 38
■
What Is Public Health Genomics?
aim to bring knowledge together from different sources across the sciences, arts, and humanities so that a multi-disciplinary resource is available for application to projects in public health genomics. For example, the HumGen database at the Centre de recherche en droit public (CRDP) at the University of Montreal, Canada, is a searchable international database of laws and policies relevant to the application of human genetics. The PolicyDB database curated by the Public Health Genetics Unit in Cambridge also deals with the policy field but differs from HumGen in that it concentrates mainly on the United Kingdom and the EU, and coverage includes policy for scientific research, public health, and health services as well as legal and bioethical documents. A highly developed example of knowledge integration in public health genomics is the Human Genome Epidemiology Network (HuGE Net) initiative (Ioannidis, 2005). HuGE Net was set up by Muin Khoury in 1999 with the aim of assembling and critically analyzing population data on genotype prevalences and gene–disease associations emerging from research in genetic epidemiology. This task is an essential prerequisite to the development of any clinical or public health applications of this information. HuGE Net, discussed in more detail in another chapter of this book, has grown into an international multi-disciplinary initiative comprising around 800 collaborators in 43 countries. HuGE Net’s core activities are information exchange through its website and newsletter, training and technical assistance in systematic review and meta-analysis, methodological development, knowledge base development, and information dissemination. Another international initiative, spearheaded by the CRDP in Montreal, is the Public Population Project in Genomics (P3G) Consortium. The aim of P3G is “to provide the international population genomics community with the resources, tools, and know-how to facilitate data management for improved methods of knowledge transfer and sharing.” P3G members include representatives from the major biobank projects underway across the world. A key feature is the P3G Observatory, a knowledge transfer platform that includes information on the design and status of the various projects and their governance arrangements, to enable comparisons and sharing of best practice.
Service Development and Evaluation This activity includes development of policies, programs, and services – both clinical and preventive – in the health sector; strategic planning; service organization; manpower planning and capacity building; service review and evaluation; and guideline development. The evaluation of population genetic screening programs, either proposed or current, is an example of work in this strand of public health genomics. In the United States, for example, public health leadership was instrumental in securing a rational approach to the proposed introduction of population screening for hereditary hemochromatosis. A multi-agency process produced a consensus statement that the current evidence base did not justify screening and made recommendations for further research (Burke et al., 1997). Further population-based research was funded to establish the penetrance of hemochromatosis-associated mutations and the prevalence of symptomatic hemochromatosis, and a web page is available on the website of the National Office of Public Health Genomics providing education about hereditary hemochromatosis for health care providers and the general public. The public health genomics community has also taken the lead on the evaluation of DNA tests for clinical use, developing both the theoretical foundations for test evaluation and protocols for implementation in practice. For example, CDC’s National Office of Public Health Genomics has convened a project known as EGAPP (Evaluation of Genomic Applications in Practice and Prevention) that aims “to develop a coordinated process for evaluating genetic tests and other genomic applications that are in transition from research to clinical practice.”The EGAPP project draws on experience gained from application of an earlier evaluation protocol, the ACCE framework (standing for analytical validity, clinical validity, clinical utility, and ethical, legal, and social implications; Haddow and Palomaki, 2003). The UK’s Public Health Genomics Foundation also has a strong involvement in genetic test evaluation, working with the UK Genetic Testing Network to develop and apply a protocol for the evaluation of tests funded by the National Health Service (Sanderson et al., 2005).
Communication and Stakeholder Engagement Productive dialogue with the public and other stakeholders, including patient groups, government, and industry, is important to ensure that scientific and clinical developments proceed in harmony with the attitudes and expectations of other sectors of society. Among examples of work in this area, the Public Health Genetics Unit (PHGU) in Cambridge established a public panel to provide an independent perspective on its work and actively involved patients and carers in its service development projects. For example, a project on learning disability employed a mixture of online discussion, telephone interviews, and focus groups to develop a document, Parents as Partners, which provides detailed evidence of the value that parents place on a genetic diagnosis in learning disability, and sets out recommendations for the health service (Gogarty, 2006).
Informing Public Policy The term “policy” refers to a broad range of public policies and programs that have a direct or indirect impact on the application of genomics in health care. Activities may encompass legal, philosophical, and social analysis at an applied level; development of regulatory frameworks; engagement in the policy making process; promoting relevant research; and seeking international comparisons. Current policy issues include legal and regulatory frameworks for genetic testing; the funding of science and the prioritization of relevant research; consent, confidentiality, data protection, and the use of human tissue; attitudes to and relationships with the pharmaceutical and biotechnology industries; and the patenting of genes and genomic technologies. There is considerable academic research activity on these topics. For example, the US National Human Genome Research
Moving Public Health Genomics Forward: Leadership and Networks
Institute “ELSI” program has an annual budget of around $18 million to support research on topics including privacy and fair use of genomic information, DNA banking, ethical conduct of genomic research, genetic discrimination, and the psychosocial impact of genomic testing and technologies. In the United Kingdom, the Economic and Social Research Council has also invested heavily in this field, supporting social science research in genomics through six Genomics Centers that together form the ESRC Genomics Network. Despite the volume of theoretical work that has been done on these issues, however, there has been less attention to finding practical policy solutions; an aim of public health genomics is to fill this gap. The PHGU’s policy team, for example, carried out detailed scrutiny of draft UK legislation governing the storage and use of human tissue samples for purposes including DNA analysis (Liddell and Hall, 2005). Flaws in the draft legislation were identified that could have impeded both research and clinical practice. Working with partner organizations in clinical genetics and biomedical research, PHGU was able to help influence the political process to achieve amendments that at least partly resolved these problems, and was involved in initiatives to produce practical guidance for health professionals needing to comply with the new legislation. PHGU’s mode of working on the Human Tissue Bill highlights another key aspect of public health genomics: the importance of forging working alliances with relevant stakeholder groups from research, clinical medicine, industry, patients, and the general public to achieve a consensus approach that is vital for securing “ownership” of a project and its conclusions. Education and Training As genomics and genomic technologies begin to have an impact on clinical practice, all health professionals will need to become “literate” in the general principles of genomics and its specific applications in their own specialist field. Public health professionals and health service managers will also need a working knowledge of genomics in order to make sound judgments about the planning and evaluation of health services (Burke, 2005). Public health genomics contributes to these goals by promoting programs of genomic literacy for health professionals (as well as generally within society); specific training for public health genomics specialists; and development of educational materials, courses, workshops, and seminars. For example, a set of competencies in genomics for the public health workforce has been developed by the National Office of Public Health Genomics in Atlanta. Competencies are documented for the public health workforce as a whole and for specific groups including leaders/administrators, clinicians, epidemiologists, health educationalists, laboratory staff, and environmental health workers. For public health genomics specialists, the multidisciplinary Masters and PhD programs developed at the Universities of Michigan and Washington offer a solid grounding in the field: the University of Washington Masters course, for example, includes core modules in genetic epidemiology; ethical
■
451
and social issues; pharmacogenetics and toxicogenomics; genetics and the law; public perspectives on ethical issues; introduction to genetic services and bioinformatics; and economic and policy issues for genetic technologies and services. In the United Kingdom, the PHGU led a project to develop a national strategy for genetics education for health professionals; the project’s report (Burton, 2003) was one of the influences that led to the establishment of a government-funded NHS National Genetics Education and Development Centre in Birmingham. A multi-disciplinary post-graduate course in public health genomics is under development at Cranfield University, based on a textbook authored by members of the Public Health Genetics Unit (Stewart et al., 2007).
MOVING PUBLIC HEALTH GENOMICS FORWARD: LEADERSHIP AND NETWORKS Public health genomics is still in an early phase of development. It has yet to reach “critical mass” in any individual country and in many parts of the world does not yet exist. The pioneer groups and organizations are taking the lead and working together to share resources, provide credibility for those wishing to develop public health genomics in their own countries, and establish collaborations in key areas of work. GRaPH Int, the international public health genomics network, aims to fulfill a coordinating role and to facilitate collaboration and sharing of information (Burke et al., 2006). The GRaPH Int website provides a portal to the organizations and resources that are available worldwide to support public health genomics. Multi-disciplinary GRaPH Int working groups are identifying priorities and setting up collaborations in key areas including education and achieving more effective integration between the ELSI field and other parts of the enterprise. Regional and national networks also have a role to play. In Europe, the Public Health Genomics European Network (PHGEN), funded by the EU, supports the development of public health genomics in the countries of the EU, the EU-applicant countries, and the members of the European Free Trade Area. Its activities include developing an inventory of key issues for public health genomics in Europe, identifying areas of similarity and diversity in legal and regulatory regimes for genomic applications, stimulating exchange of information, and identifying key experts in relevant fields. In the United Kingdom, a national society for public health genomics, the Society for Genomics, Policy and Population Health has been set up as a special interest subgroup of the British Society for Human Genetics. Partnerships and collaborations with institutions and organizations that have related goals will also be an important part of the development of public health genomics. Such partnerships will be particularly important in extending the reach of public health genetics into the developing world.
452
CHAPTER 38
■
What Is Public Health Genomics?
CONCLUSION The new field of public health genomics faces many challenges, both intellectual and practical. Chief among the intellectual challenges is the need to inspire a change in the traditional mindset within the profession of public health. Public health professionals can no longer ignore the role of genes as important determinants of health, or the recognition that disease susceptibility is a consequence of gene–gene and gene–environment interaction. They must begin to master a broad knowledge base in genetic and genomic science, the population sciences, and the humanities and social sciences (Table 38.1; also see Fineman, 1999). Practical challenges for public health genomics include finding ways of facilitating the integration of genomic science into mainstream medicine; devising and implementing sound methodologies for evaluation of genomic tests and complex molecular biomarkers; achieving a rational organization of molecular and cytogenetic testing services; evaluating current and proposed genetic screening programs; and working with primary care practitioners to clarify the role of genomics in the primary care setting. A further challenge will be to ensure that the developing world does not miss out on the potential benefits of advances in genomics. Heritable diseases such as sickle cell disease and the thalassemias are a major burden in many developing countries but screening, diagnostic, and health care services for these conditions are often rudimentary at best (Weatherall, 2005). Developing countries also stand to derive huge benefit from new vaccines and diagnostics for infectious diseases and from biotechnological approaches to fight malnutrition and environmental degradation (Daar et al., 2002). The approach of public health genomics, emphasizing knowledge integration and a multidisciplinary approach, proceeding from a sound evidence base, promoting coherent public policy development, and ensuring that the necessary infrastructure is in place before interventions
TABLE 38.1 genomics
The knowledge base for public health
Genomic science
Population sciences
Humanities
Basic concepts of molecular, cell, and developmental biology
Epidemiology
Sociology
Biostatistics
Anthropology
Environmental health sciences
Law
Infectious diseases
Ethics
Structure and composition of the human genome Roles of genes in health and disease Gene–gene and gene–environment interactions Mendelian genetics, family histories, and pedigrees Principles of genome epidemiology Genetic testing and screening
Social and behavioral sciences Health Economics
Economics Metaphysics and epistemology Theology Political philosophy
Health services research Management science Information science
Pathogen genomes Genomic technologies Bioinformatics
are implemented, is as relevant to ensuring population benefit in this setting as it is in the developed world. In rising to all these challenges, public health genomics is poised to make an important contribution to the goal of translating the fruits of genomic research into health benefits for all.
REFERENCES Burke, W. (2005). Contributions of public health to genetics education for health professionals. Health Educ Behav 32, 668–675. Burke, W. et al. (1997). Hereditary haemochromatosis: Gene discovery and its implications for population-based screening. JAMA 280, 172–178. Burke, W., Khoury, M.J., Stewart, A. and Zimmern, R.L. (2006). The path from genome-based research to population health: Development of an international collaborative public health genomics initiative. Genet Med 8, 451–458. Burton, H. (2003). Addressing Genetics, Delivering Health. Public Health Genetics Unit, Cambridge. Available at http://www.phgfoundation.org. Centers for Disease Control and Prevention (1997). Translating advances in human genetics into public health action: A strategic plan. Clayton, E.W. (2003). Genomic medicine: Ethical, legal, and social implications of genomic medicine. N Engl J Med 349, 562–569. Daar, A., Thorsteinsdottir, H., Martin, D.K., Smith, A.C., Nast, S. and Singer, P.A. (2002). Top ten biotechnologies for improving health in developing countries. Nat Genet 32, 229–232.
Davey Smith, G.D., Ebrahim, S., Lewis, S., Hansell, A.L., Palmer, L.J. and Burton, P.R. (2005). Genetic epidemiology and public health: Hope, hype, and future prospects. Lancet 366, 1484–1498. EGAPP project. National Office of Public Health Genomics, Centers for Disease Control and Prevention. http://www.cdc.gov/genomics/ gtesting/egapp.htm. Fineman, R. (1999). Qualifications of public health geneticists? Community Genet 2, 113–114. Gogarty, B. (2006). Parents as Partners. Public Health Genetics Unit, Cambridge. Available at http://www.phgfoundation.org. GRaPH Int. http://www.graphint.org. Guttmacher, A.E. and Collins, F.S. (2005). Realizing the promise of genomics in biomedical research. N Engl J Med 347, 1512–1520. Haddow, J.E. and Palomaki, G.E. (2003). ACCE: A model process for evaluating data on emerging genetic tests. In Human Genome Epidemiology: A Scientific Foundation for Using Genetic Information
Recommended Resources
to Improve Health and Prevent Disease (M.J. Khoury, J. Little and W. Burke, eds), Oxford University Press, New York, pp. 217–233. Haga, S.B., Khoury, M.J. and Burke, W. (2003). Genomic profiling to promote a healthy lifestyle: Not ready for prime time? Nat Genet 34, 347–350. Human Genome Epidemiology Network (HuGE Net). http://www. cdc.gov/genomics/hugenet/default.htm. HumGen database. http://www.humgen.umontreal.ca/int/. Ioannidis, J.P. (2005). A road map for efficient and reliable human genome epidemiology. Nat Genet 38, 3–5. Institute of Medicine Committee for the Study of the Future of Public Health (1998). The Future of Public Health. National Academies Press,Washington DC. Juengst, E.T. (1995). ‘Prevention’ and the goals of genetic medicine. Hum Gene Ther 6, 1595–1605. Khoury, M.J. (1996). From genes to public health: Applications of genetics in disease prevention. Am J Public Health 86, 1717–1722. Khoury, M.J. (2003). Genetics and genomics in practice. The continuum from genetic disease to genomic information in health and disease. Genet Med 5, 261–268. Liddell, K. and Hall, A. (2005). Beyond Bristol and Alder Hey: The future regulation of human tissue. Med Law Rev 13, 170–223. Murray, T. (1997). Genetic exceptionalism and ‘future diaries’: Is genetic information different from other medical information? In Genetic Secrets: Protecting Privacy and Confidentiality in the Genetic Era (M.A. Rothstein, eds.),Yale University Press, New Haven, pp. 60–73. National Office of Public Health Genomics, Centers for Disease Control and Prevention. Genomic Competencies for the Public Health Workforce. http://www.cdc.gov/genomics/training/competencies/default.htm. National Office of Public Health Genomics, Centers for Disease Control and Prevention. Hereditary Haemochromatosis: A Public
■
453
Health Perspective. http://www.cdc.gov/genomics/training/perspectives/hemo.htm. NHS Central Research and Development Committee (1995). Genetics and Common Disease. Department of Health, London. NHS National Genetics Education and Development Centre. http:// www.geneticseducation.nhs.uk/. Ojha, R.P. and Thertullien, R. (2005). Health care policy issues as a result of the genomic revolution: Implications for public health. Am J Public Health 95, 385–388. PHG Foundation Policy DB Database. http://www.phgfoundation. org/policydb. Public Health Genomics European Network. http://www.phgen. nrw.de/. Public Population Project in Genomics (P3G). http://www.p3gconsortium.org/. Sanderson, S., Zimmern, R., Kroese, M., Higgins, J., Patch, C. and Emery, J. (2005). How can the evaluation of genetic tests be enhanced? Lessons learned from the ACCE framework and evaluating genetic tests in the United Kingdom. Genet Med 7, 495–500. Sankar, P. (2003). Genetic privacy. Annu Rev Med 54, 393–407. Stewart, A. (2006). Genome-based Research and Population Health. Report of an expert workshop held at the Rockefeller Foundation Study and Conference Centre, Bellagio, Italy, 14–20 April, 2005. Available at http://www.phgfoundation.org. University of Michigan School of Public Health Interdepartmental Concentration in Public Health Genetics. http://www.sph.umich. edu/genetics/. University of Washington Institute of Public Health Genetics, Degree Programs in Public Health Genetics. http://depts.washington. edu/phgen/degreeprograms/DegreeProgs_OV.shtml. Weatherall, D.J. (2005). The global problem of genetic disease. Ann Hum Biol 32, 117–122.
RECOMMENDED RESOURCES Gwinn, M. and Khoury, M.J. (2006). Genomics and public health in the United States: Signposts on the translation highway. Community Genet 9, 21–26. (A recent review of progress in public health genomics in the United States.) Halliday, J.L., Collins, V.R., Aitken, M.A., Richards, M.P. and Olsson, C.A. (2004). Genetics and public health – evolution, or revolution? J Epidemiol Community Health 58, 894–899. (A discussion of the prospects and challenges raised by a convergence between public health and genetics.) Institute of Medicine Committee on Genomics and the Public’s Health in the 21st Century (2005). In Implications of Genomics for Public Health (Hernandez, L., ed.), National Academies Press, Washington DC. (This report summarizes the “state of the art” for public health genomics in the United States, and prioritizes the issues that need to be addressed.) Khoury, M.J., Burke, W. and Thomson, E., eds. (2000). Genetics and Public Health in the 21st Century. Oxford University Press, New York. (An excellent overview of the conceptual framework of public health genetics and its early development during the 1990s.) National Office of Public Health Genomics website. http://www.cdc. gov/genomics. Provides information on NOPHG projects, training
resources, and the GDPInfo database of published literature relating genes, diseases, and environmental exposures. Hosts the HuGE Net website. Users can sign up for a weekly e-mail update of news, event, and publications in public health genomics. PHG Foundation website. http://www.phgfoundation.org. Features information on the PHGU work program, and an online newsletter of genomics policy news, linked to related websites and records in the PolicyDB database. Users can request a monthly e-mail compilation and commentary of genomics policy news. Stewart, A., Brice, P., Burton, H., Pharoah, P., Sanderson, S. and Zimmern, R. (2007). Genetics, Health Care and Public Policy. Cambridge University Press, Cambridge UK. (This textbook sets out the knowledge base for public health genetics and the policy and service issues raised by genomic advances, from a UK perspective.) Zimmern, R. and Cook, C. (2000). Genetics and health. Policy issues for genetic science and their implications for health services. The Nuffield Trust, London. (A useful and particularly broad-ranging discussion of service and policy issues raised by genomics.)
CHAPTER
39 Why Do We Need Public Health in the Era of Genomic Medicine? Muin J. Khoury and Marta Gwinn
INTRODUCTION We live in the “omics” era. New fields with “omics” attached to existing disciplines seem to emerge on a regular basis. For example, we now have pharmacogenomics, nutrigenomics, metabonomics, transcriptomics, proteomics, toxicogenomics, and many others (Khoury, 2003). Although there are currently a few clinical applications that we can ascribe to the genomic technologies, more applications are anticipated in the next decades. In a book devoted to the applications of genomic sciences in clinical practice, one might wonder about the role of public health in the practice of genomic medicine in the 21st century. This section explores the growing intersection of public health with genomic medicine. Chapter 38 outlines the need and emergence of “public health genomics” as multidisciplinary field of science and practice. This chapter describes the growing role of epidemiologic methods and approaches in collecting the type of information needed for genomic medicine to become a reality. Chapter 42 explores the growing importance of using family history in practice in relation to genetics and genomics. Finally, Chapter 41 explores the evolution of newborn screening, a traditional public health function from traditional genetic diseases to the new ear of genomic medicine. In this introductory chapter, we describe the importance of public health-medicine collaboration in the genomics era, namely in four areas for a joint vision: the focus on prevention,
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 454
the importance of population sciences, the importance of evidence-based knowledge synthesis, and the need for health services research and population health outcome monitoring. This collaboration is discussed in detail elsewhere (Khoury et al., submitted). First, we explore the continuum from genetic to genomics in practice and discuss the implications to how public health and medicine can collaborate in the era of genomics.
THE CONTINUUM FROM GENETICS TO GENOMICS IN HEALTH PRACTICE Guttmacher and Collins (2002) viewed “genetics as the study of single genes and their effects” while genomics refers “to the study not just of single genes, but of the functions and interactions of all the genes in the genome.” This definition implies a quantitative difference between the two fields (the study of multiple genes versus one gene) and makes genetics part of genomics. In addition to this quantitative difference, genomicsbased sciences lead to a fundamental qualitative difference in the application of genetics in medicine and public health. Although set as a dichotomy, Table 39.1 shows how genetics and genomics can be viewed on a continuum ranging from the two extremes of a genetic disorder in genetics to the concept of genetic information in genomics. The current practice of medical genetics has focused on rare clinically recognized disorders associated with
The Role of Public Health in the Translation of Human Genome Discoveries into Health Applications
TABLE 39.1
■
455
The continuuma from genetics to genomics in health practice Genetics
Genomics
Focus
Genetic disorders
Genetic information
Types of conditions
Single gene disorders Chromosomal anomalies Diseases with genetic component
All human diseases
Genetic involvement
Mutations with high or incomplete penetrance
Genetic variation at single and multiple loci
Environmental involvement
Variable; many disorders have no known environmental determinants
Strong interaction with chemical, infectious, physical, pharmacologic agents
Medical practice
Genetic services (Genetic counseling, testing, management)
●
Public health practice
Assuring delivery of genetic services to individuals, families and society Population screening (e.g., newborn screening)
Health care Risk stratification using genetic variation/family history ● Pharmacogenomics ● Diagnostic tests ● Prognostic tests Developing health policy and assuring appropriate use of genetic information in population health Population screening
a
From Khoury M.J. (2003). Genetics and genomics could be viewed on a continuum from clinically recognized disorders associated with single gene mutations for which the traditional genetic services model applies, to genetic information expressed at multiple loci which can be used in prediction, diagnosis or treatment of a disease not classified as genetic. Along this continuum could be clinically recognized genetic disorders with low penetrance resulting from interaction with other genes and exposures.
single gene mutations, for which the traditional genetic services model applies with its accompanying medical processes (genetic counseling/testing/management) and public health processes (delivery of genetic services and newborn screening). Until recently, the focus of genetics has been on those conditions that are known to be due to mutations in single genes (e.g., Huntington disease), whole chromosomes (e.g., Down syndrome), or diseases with strong genetic components (e.g., birth defects and developmental disabilities). On the other hand, the practice of genomics in medicine will be centered on the concept of “genetic information” due to variation at one or multiple loci and strong interactions with environmental factors (broadly defined to include diet, drugs, infectious agents, chemicals, physical agents, and behavioral factors). Such information can then be used in diagnosis, treatment and prevention of all diseases, not only genetic disorders.
THE ROLE OF PUBLIC HEALTH IN THE TRANSLATION OF HUMAN GENOME DISCOVERIES INTO HEALTH APPLICATIONS Public health has been traditionally identified with state, federal and local public health agencies; however, recent reports by the Institute of Medicine in the United States and others have adopted a more inclusive view, defining public health professionals
as those working on improving health from a population perspective (IOM, 2003). “Public health professionals” are those employed not only in government but also in health care delivery, academia, community organizations, and the private sector; together, they are actors in the “public health system” (IOM, 2002), which is working to assure the conditions under which a population can be healthy. If we adopt the expanded view of public health as population health, it becomes much easier to consider genomics at the interface of the translational research agenda between medicine and public health. Another reason to adopt an expanded view of population health is that future applications of genomics will occur primarily in the health care delivery system and not in the context of the national or state-mandated screening programs (traditional public health). Unfortunately, the sad history of eugenics will serve as constant reminder of horrifically failed applications for the perceived benefits of population health. Furthermore, the only current model of intersection of genetics and public health, namely newborn screening, should not necessarily serve as a model of future applications and joint partnership between medicine and public health. Undoubtedly, while newborn screening programs make important contributions to health, they continue to raise important ethical questions about informed consent, especially in the light of proposed expanded programs where health benefits are absent or less clear cut than medical emergencies provided by phenylketonuria (PKU) and congenital hypothyroidism. Thus, under the rubric of population health, collaboration between public health and medicine in genomics will emphasize the
456
CHAPTER 39
TABLE 39.2
■
Why Do We Need Public Health in the Era of Genomic Medicine?
The continuum of translation research in human genomics: Types of research and BRCA example
Translation research phase
Notation
Types of research
Examples
TI
Discovery to candidate health application
Phases I and II clinical trials Observational studies
Is there an association between BRCA mutations and breast cancer?
T2
Health application to evidence-based practice guidelines
Phase III clinical trials Observational studies Evidence synthesis and guidelines development
What is the positive predictive value of BRCA mutations in at-risk women?
T3
Practice guidelines to health practice
Dissemination research Implementation research Diffusion research Phase IV clinical trials
What proportion of women who meet the family history criteria are tested for BRCA and what are the barriers to testing?
T4
Practice to population health impact
Outcomes research (includes many disciplines) Population monitoring of morbidity, mortality, benefits, and risks
Does BRCA testing in asymptomatic women reduce breast cancer incidence/improve outcomes?
Adapted from Khoury et al. (2007b). See also Chapter 22.
importance of an individual-oriented perspective but also the need for a population-based evaluation strategy and evidencebased health care with access of validated information to all segments of the population. Advances in genomics have also been accompanied by an emerging emphasis on translational research, a movement fueled by the NIH road map initiative (Zerhouni, 2003, 2005) to accelerate the translation from basic science to clinical applications. The translation framework is based primarily on the discovery of new drugs and their accelerated use in human clinical trials, with little or no emphasis on prevention. However, the “bench to bedside” paradigm covers only part of the distance from discovering new knowledge to delivering health benefits at the population level (Green and Seifert, 2005; Horig and Pullman, 2004). In 2003, Claude Lenfant (2003), the retiring director of the National Heart, Lung and Blood Institute, presented the Shattuck lecture titled “Clinical research to clinical practice – lost in translation?” He described many discoveries of curative or preventive interventions that nevertheless do not reach the end of the translation highway and asked, “If we can’t do it with aspirin, how will we do it with DNA?” The “lost in translation” problem is complicated by the increasing costs of health care delivery and persistent inequities in access. Some have called the next stage in translation “translation type II” (Rohrbach et al., 2006), which requires more applied research on the best ways to deliver or disseminate interventions that work in real life settings (delivery research) and to evaluate health outcomes and population impact (outcomes research). In 2007, Westfall et al. (2007) proposed that the evaluation of evidence-based interventions in practice can be called “type 3 translation research.” As shown in Table 39.2 we have also extended (Khoury et al., 2007b) the translation pathway
Focus on prevention
Public health sciences
Discovery
T1/T2/T3/T4
Delivery
Assessment policy assurance evaluation
Knowledge integration
Figure 39.1 A framework for medicine-public health partnership in the genomics era (Adapted from Khoury et al., 2007a, b).
to type 4 translation (T4) research, which seeks to evaluate the “real world” health outcomes of a genomic application in practice. We have reported that most human genomics research between 2001 and 2006, inclusive, was not translational and have estimated that no more than 3% of published research focuses on T2 and beyond. Indeed, evidence-based guidelines and T3 and T4 research currently are rare (Khoury et al., 2007b). As discussed below, an enhanced focus on translation and translation research from T1 through T4 can foster a true partnership between medicine and public health in the genomics era in four major areas (see Figure 39.1).
The Population Perspective: Crucial Role of Public Health Sciences
THE FOCUS ON DISEASE PREVENTION AND HEALTH PROMOTION Advances in genomics will provide new opportunities for disease prevention and health promotion – the main focus of public health, regardless of whether it is delivered at the individual level or through population-wide interventions. Understanding genetic effects and gene–environment interactions in disease processes could lead to the recommendation that certain subgroups avoid defined exposures or receive targeted interventions. Stratification by genotype or family history already provides a means for tailoring screening tests for early disease detection (e.g., colorectal cancer screening in genetically susceptible persons), a paradigm likely to be extended to early detection of other conditions. A review of the public health implications of genomic research related to asthma (UWCGPH, 2004) illustrates the potential opportunities and challenges for translating new knowledge into improved prevention and treatment of a common disease. Strong evidence supports a causal role for both genetic and environmental factors. Genomics research has identified numerous gene loci associated with asthma, and further studies of biological pathways associated with asthma, are likely to yield new approaches to prevention and therapy. The earliest clinical applications will be in pharmacogenomics, using genetic information to optimize therapy and prevent adverse events. The path for translating results of genomic research to population-level interventions will not always lead through genetic testing and knowledge of individual genotypes. For example, a study in Mexico of children with asthma found that supplementation with the antioxidant vitamins C and E improved lung function in children with a common polymorphism of glutathione S-transferase M1 (GSTM1) who are exposed to ozone (Romieu et al., 2004). These findings might suggest a simple intervention – antioxidant vitamin supplementation – for children with asthma who are exposed to ozone. Without genotype-specific analysis, a potentially important population-level intervention could have been overlooked. New gene discoveries are reported daily and have initiated dialog on the value of genetic information with respect to prevention. Consider, for example, the discovery in 2006 that a variant of the TCFT7L2 gene is associated with increased risk of type 2 diabetes (Grant et al., 2006). This finding was noteworthy for several reasons. First, type 2 diabetes is a serious disease and a major public health problem. Family history is an important risk factor but until now, few genetic associations have been identified. The investigators replicated the association with TCF7L2 in three independent populations and presented molecular evidence that the gene product is a high mobility group boxcontaining transcription factor related to blood glucose homeostasis. The gene product may act through regulation of proglucagon gene expression via the Wnt signaling pathway (Grant et al., 2006). The senior investigator commented to the New York Times that a practical consequence of the discovery could be a diagnostic test to identify people who are at increased risk of type 2 diabetes. He speculated that “these people, knowing their
■
457
risk, would be motivated to exercise more and adopt a healthier diet” (Jannssens et al., 2006). However, a relatively simple analysis is sufficient to show that a test for TCF7L2 variants by itself would have very poor predictive value, adding little to the risk information already provided by family history, and would have limited utility for prevention (Jannssens et al., 2006). In particular, the hypothesis that genetic risk information of this kind will motivate behavioral change remains to be tested. Once a genetic variant is discovered, producing a molecular test to detect it is theoretically straightforward. However, the clinical validity of such a test is highly dependent on population characteristics; these include not only prevalence of the genetic variant and the strength of its association with disease, but prevalence of the disease and interactions between this genetic variant and many other risk factors. Clinical utility is even more difficult to establish because it depends on the availability of a specific, effective intervention that adds value to existing practice. Although most gene discoveries for common diseases are not ready for prevention applications, they have, nevertheless, initiated a common interest in medical and public health to develop a common framework for how such discoveries can be evaluated for their potential applications for disease prevention.
THE POPULATION PERSPECTIVE: CRUCIAL ROLE OF PUBLIC HEALTH SCIENCES Gene discovery and characterization is a basic science enterprise. Gene discovery typically occurs in highly selected patient groups and the associations observed in such groups may not be generalizable to the population as a whole. This is where the population perspective is crucial. Indeed, Omenn (2002) described a “golden age” for the public health sciences which include epidemiology, biostatistics, environmental health, health education, behavioral and social sciences, and many other fields. As predicted by Kerr White at the start of the Human Genome Project, “molecular biology, especially once the human genome is mapped, must surely turn increasingly to the study of populations” (White, 1991), seeking the partnership of epidemiologists. In fact, we are seeing a convergence of genetic and epidemiologic methods beginning to take the field of genetic epidemiology beyond gene discovery. As discussed in Chapter 40 human genome epidemiology is a multidisciplinary field that uses systematic application of epidemiologic methods in studies of human genetic variation in association with health and disease in populations. Using the previous example of the TCF7L2 gene and type 2 diabetes gene discovery above, a population perspective would require an evaluation of whether or not a specific intervention or prevention strategy can be developed beyond the universal importance of healthy lifestyle in the prevention of type 2 diabetes. Even if this is the case, additional research will be needed to assess the prevalence of high-risk variants in the populations where testing is considered, a population-based estimate of risk, information about modification of risk by other genetic and environmental factors and behavioral
458
CHAPTER 39
■
Why Do We Need Public Health in the Era of Genomic Medicine?
and social science research to determine how genetic information can be used, if at all, to promote behavior change (Ames and McBride, 2006). Because of the well known difficulties of implementing behavior change for the prevention of common diseases, it is not clear whether or not adding genetic information such as the diabetes genetic risk would be helpful or detrimental.
THE ROLE OF KNOWLEDGE INTEGRATION ACROSS DISCIPLINES As discussed in Chapter 38 knowledge integration is at the heart of public health genomics. The translation of genomic discoveries from the bench to the bedside is a long and arduous process that requires accumulation and synthesis of knowledge in many fields, including observational epidemiologic studies on gene-disease associations, gene–environment interactions, and clinical trials of efficacy of general and genotype-specific interventions. Because of the proliferation of information on genomics and health (e.g., more than 25,000 articles on gene-disease associations have been published in the past 6 years, Lin et al., 2006), evidence must be integrated systematically before “discovery” can lead to “delivery.” False leads and blind alleys have to be eliminated through an iterative process of evidence-based information synthesis, such as efforts of the Human Genome Epidemiology Network, which promotes systematic reviews of gene-disease associations (CDC, 2007a). An enhanced partnership between medicine and public health will be essential in the knowledge integration process given the transdisciplinary nature of research that leads to genetic discoveries and their potential applications. By its nature, it will require a close collaboration of basic sciences, clinical sciences, public health and social sciences. This process will lead to evidence-based guidelines on appropriate use of genetic information in health practice, such as ones provided by the Cochrane collaboration (Grimshaw et al., 2006), US Preventive Services Task Force (Harris et al., 2001); and the Guide to Community Preventive Services (CDC Community Guide, 2007).These efforts examine on a regular basis what works and what does not work in health services to achieve health impact for individuals, families, and populations. An extension to these efforts in the United States is the CDC-led Evaluation of Genomic Applications in Practice and Prevention (EGAPP) initiative, which focuses exclusively on the utility of genomic applications and family history in clinical practice and disease prevention (CDC, 2007b).
THE ROLE OF HEALTH SERVICES RESEARCH AND POPULATION HEALTH ASSESSMENT, ASSURANCE, AND EVALUATION Like other challenges in translation, genomics can hit a roadblock if no investments are made in evaluating the best ways to assure
delivery and monitor effectiveness and safety of gene-based interventions, whether they are population screening programs like newborn screening, or early case detection and interventions delivered by clinicians. This step requires a crucial partnership between medicine and public health. Beyond basic research, efficacy research, and development of practice guidelines, ensuring access depends on adoption by health care providers of practice guidelines, and acceptance by consumers. All of these factors contribute to the effectiveness of interventions in the real world. Many of the challenges have been articulated before, including public participation, information systems, workforce training, and funding (Crowley et al., 2004). Models of research-practice partnerships have been proposed to overcome some of the current difficulties, such as the national public–private partnership research enterprise (Crowley et al., 2004). As Zerhouni and others are predicting, the advent of genomics-based healthcare delivery will require research on the best ways to integrate new knowledge into practice, achieving translation for best results from the population perspective (Lenfant, 2003). As a major new research area with real promise (as well as commercial potential), genomic medicine may offer a new opportunity to shine the spotlight on translation because it is grabbing attention, represents a major new research area, has commercial potential, and offers genuine promise – but its applications are readily generalizable to other areas of health. For example, HMO-based research (Mouchawar et al., 2005) and general provider surveys (Myers et al., 2006) were collaboratively conducted to assess the impact of a direct-to-consumer campaign for BRCA1 testing on providers’ attitude, knowledge, behaviors, and practices. Another example is the collaboration between providers, consumers, professional organization and various government agencies to assess and increase the public’s awareness and use of family history as an additional tool for disease prevention and public health (CDC, 2007c; DHHS, 2007). Clearly, such efforts need to be sustained and extended into emerging areas like pharmacogenomics and genetic tests in general (Davis and Khoury, 2006). Among others, these applications should include economic evaluations of gene-based tests and interventions which are increasingly important in the face of ever escalating health care expenditures. Invariably, translation efforts will uncover gaps in our knowledge base on genes and health, and in the processes of health care delivery and population health outcomes. With the current expanding use of genetic information in practice without any real oversight, a medicine-public health partnership will be essential for identifying these gaps, further shaping the research agenda, and keeping practitioners, consumers and policy makers abreast of the best applications of scientific discoveries for population health impact.
CONCLUSION Advances in genomics – especially in relation to common diseases – are leading to increasing interaction and interdependence
References
between the traditional health care delivery system, which focuses on treatment of individuals, and the public health system that focuses on prevention and control in populations. This closer interaction is creating a shared “population health” focus on using genomic advances appropriately and effectively to promote health and prevent disease. Because the field of genomics is still in its infancy, this is a crucial time for medicine and public health to get together to develop a strong partnership with a joint focus on prevention, population sciences, knowledge translation and health
■
459
services research and outcome monitoring. Emerging genomic information is presenting policy makers, practitioners, and researchers with a new opportunity and new urgency to bridge the divide for the benefit of population health. Dr James Marks from the Robert Wood Johnson Foundation commented that “no important health problem will be solved by clinical care alone, or research alone, or by public health alone but rather by all public and private sectors working together” (Marks, 2005). We hope the chapters in this section can begin to explore this enhanced collaboration in the era of genomic medicine.
REFERENCES Ames, S.L. and McBride, C. (2006). Translating genetics, cognitive science, and other basic science research findings into applications for prevention. Eval Health Prof 29, 277–301. Centers for Disease Control and Prevention Community Guide (2007). Evidence-based recommendations for programs and policies to promote population health. Accessed online on March 18, 2007 at: http://www.thecommunityguide.org/ Centers for Disease Control and Prevention (2007a). The Human Genome Epidemiology Network. Accessed online March 19, 2007 at: http://www.cdc.gov/genomics/hugenet/default.htm Centers for Disease Control and Prevention (2007b). Evaluation of Genomic Applications in Practice and Prevention. Accessed online March 19, 2007 at: http://www.cdc.gov/genomics/gtesting/egapp. htm#topics Centers for Disease Control and Prevention (2007c). Family history public health initiative. Accessed online March 19, 2007 at: http:// www.cdc.gov/genomics/activities/famhx.htm Crowley, W.F., Sherwood, L., Salber, P. et al. (2004). Clinical research in the United States at a crossroads: Proposal for a novel publicprivate partnership to establish a national clinical research enterprise. JAMA 291, 1120–1126. Davis, R.L. and Khoury, M.J. (2006). A public health approach to pharmacogenomics and gene-based tests. Am J Pharmacogenomics 7, 331–337. Department of Health and Human Services (2007). US Surgeon General’s family history initiative. Accessed online March 19, 2007 at: http:// www.hhs.gov/familyhistory/ Grant, S.F., Thorleifsson, G., Reynisdottir, I. et al. (2006). Variant of transcription factor 7-like 2 (TCF7L2) gene confers risk of type 2 diabetes. Nat Genet 38, 320–323. Green, L.A. and Seifert, C.M. (2005). Translation of research into practice: Why we can’t just do it. J Am Board Fam Pract 18, 541–545. Grimshaw, J.M., Santesso, N., Cumpston, M., Mathew, A. and McGowan, J. (2006). Knowledge for knowledge translation: The role of the Cochrane Collaboration. J Contin Educ Health Prof 26, 55–62. Guttmacher, A.E. and Collins, F.S. (2002). Genomic medicine – a primer. N Engl J Med 347, 1512–1520. Harris, R.P., Helfand, M., Woolf , S.H. et al. (2001). Current methods of the US Preventive Services Task Force: A review of the process. Am J Prev Med 3(Suppl), 21–35. Horig, H. and Pullman, W. (2004). From bench to clinic and back: Perspectives on the 1st IQPC translational research conference. J Trans Med 2, 44.
Institute of Medicine (2003). Who Will Keep the Public Healthy? Educating Public Health Professionals for the 21st Century. National Academies Press,Washington, DC. Institute of Medicine (2002). The Future of the Public’s Health in the 21st Century. National Academies Press,Washington, DC. Jannssens, A.C.W., Gwinn, M.L., Valdez, R. et al. (2006). Predictive genetic testing for type 2 diabetes may raise unrealistic expectations. BMJ 333, 509–510. Khoury, M.J. (2003). From genetics to genomics in practice: The continuum from genetic disease to genetic information in health and disease. Genet Med 5, 261–268. Khoury, M.J., Gwinn, M., Burke, W., Bowen, M.S. and Zimmern, R. (2007a). Will genomics widen or heal the schism between medicine and public health. Am J Prev Med 33, 310–317. Khoury, M.J., Gwinn, M., Yoon, P.W., Dowling, N., Moore, C.A. and Bradley, C.A. (2007b). The continuum of translation research in genomic medicine how can we accelerate the appropriate integration of human genome discoveries into health care and disease prevention?. Genet Med 9, 665–674. Lenfant, C. (2003). Shattuck lecture: Clinical research to clinical practice – lost in translation?. N Engl J Med 349, 868–874. Lin, B., Clyne, M., Walsh, M., Gomez, O., Yu, W., Gwinn, M. and Khoury, M.J. (2006). Tracking the epidemiology of human genes in the literature: The HuGE published literature database. Am J Epidemiol 164, 1–4. Marks, J.S. (2005). Preventive care – the first step. Manag Care 14(Suppl), 10–12. Mouchawar, J., Hensley-Alford, S., Laurion, S. et al. (2005). Impact of direct-to-consumer advertising for hereditary breast cancer testing on genetic services at a managed care organization: A naturallyoccurring experiment. Genet Med 7, 191. Myers, M.M., Chang, M.H., Jorgensen, C. et al. (2006). Genetic testing for susceptibility to breast and ovarian cancer: Evaluating the impact of a direct-to-consumer marketing campaign on physicians’ knowledge and practices. Genet Med 8, 361–370. Omenn, G.S. (2002). The crucial role of the public health sciences in the postgenomic era. Genet Med 4(suppl), 21S–26S. Rohrbach, L.A., Grana, R., Sussman, S. and Valente, T.W. (2006). Type II translation: Transporting prevention interventions from research to real world settings. Eval Health Prof 29, 302–333. Romieu, I., Sienra-Monge, J.J., Ramirez-Aguilar, M. et al. (2004). Genetic polymorphism of GSTM1 and antioxidant supplementation influence lung function in relation to ozone exposure in asthmatic children in Mexico City. Thorax 59, 8–10.
460
CHAPTER 39
■
Why Do We Need Public Health in the Era of Genomic Medicine?
University of Washington Center for Genomics and Public Health (UWCPGH) (2004). Asthma genomics: Implications for public health. 2004. Accessed online on March 15, 2007 at: http://depts. washington.edu/cgph/workinggroups/Asthma%20Genomics_Im plications%20for%20Public%20Health.pdf Westfall, J.M., Mold, J. and Faqnan, L. (2007). Practice-based research – blue highways on the NIH road map. JAMA 297, 403–406.
White, K.L. (1991). Healing the schism: Epidemiology, medicine, and the public’s health. Springer-Verlag, New York. Zerhouni, E.A. (2003). The NIH Roadmap. Science 302, 63–72. Zerhouni, E.A. (2005). Translational and clinical science: Time for a new vision. New Engl J Med 353, 1621–1623.
CHAPTER
40 Principles of Human Genome Epidemiology Marta Gwinn and Muin J. Khoury
INTRODUCTION Early forecasts for genomic medicine underestimated the challenge of translating advances in basic science into new applications for improving health (Collins, 1999; Hoffman, 1994). The genetic architecture of human health and disease arises from interactions among multiple genes and environmental factors over the course of a lifetime and thus is far more complex than the genome itself. In this context of complex causation, the effects of single factors – including genetic variants – are difficult to detect and interpret (Ioannidis et al., 2006b). Current research approaches are often inconsistent, hindering attempts to synthesize the available evidence. Successful translation now hinges on establishing new norms for integrating the results of biologic and epidemiologic research and for bridging individual and population-level interpretations. In this chapter, we present the perspective of human genome epidemiology, which promotes the use of epidemiologic methods to study the distribution of genetic variants in populations, along with their associations with disease, their interactions with other factors, and their value for informing interventions. Human genome epidemiology is concerned with genetic variation – whether measured at the level of DNA sequence, gene expression (transcriptomics), or protein analysis (proteomics) – and more broadly with “omic” biomarkers based on modern quantitative biology (Thomas, 2006). The principles
of human genome epidemiology include population-based sampling, sound epidemiologic study design and analysis, emphasis on gene–environment interactions, careful attention to potential errors and biases, and a commitment to systematic review and integration of biological information (Little et al., 2003). Human genome epidemiology aims to provide a scientific basis for using genetic information to prevent, diagnose, and treat disease. This chapter provides an overview of human genome epidemiology and its role in translating genomic research results into an evidence base for medical and public health practice.
HUMAN GENOME EPIDEMIOLOGY Genetics and epidemiology share a common approach, using quantitative methods to examine the variation of normal and pathological traits in human populations. Until recently, however, the intersection of genetics and epidemiology was narrowly defined and was focused almost exclusively on discovering genetic “causes” of traits and diseases by studying familial aggregation. Developments in genetic epidemiology and statistical genetics have mostly centered on methods for discovering disease susceptibility genes in families with a clear pattern of inheritance (Burton et al., 2005) (described in other chapters). This approach has identified mutations in more than 10,000 genes associated with Mendelian disorders (e.g., cystic fibrosis), as well as high-penetrance
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 461
462
CHAPTER 40
■
Principles of Human Genome Epidemiology
genetic variants associated with certain common diseases (e.g., breast cancer) in families with multiple affected members (National Library of Medicine [OMIM]). Typically, the variants identified in such families explain only a small proportion of all cases; for example, BRCA1 and BRCA2 mutations account for only an estimated 1–2% of all breast cancer cases in the general population (McClain et al., 2005). During the past 10 years, the search for genetic factors in common diseases has relied increasingly on genetic association studies. Association studies compare the distribution of one or more genetic variants in persons with or without a disease, either related (e.g., affected child and unaffected parents) or unrelated (Hattersley and McCarthy, 2005). Family-based study designs and methods are reviewed in other chapters as well as in several articles in a valuable series published in The Lancet (Cordell and Clayton, 2005; Palmer and Cardon, 2005;Teare and Barrett, 2005). Association studies that compare the frequency of genetic variants in unrelated persons with or without disease are fundamentally identical to traditional epidemiologic studies. They can be analyzed by using many of the same methods and are subject to the same types of limitations, errors, and biases (Cordell and Clayton, 2005; Hattersley and McCarthy, 2005). The value of genetic association studies for discerning useful gene–disease relationships is still a matter of debate (Davey Smith et al., 2006). Although many such studies aim to detect a novel gene–disease association, many more attempt to replicate previously reported findings. Much of the frustration with association studies stems from their poor track record when it comes to replication. Reasons for this phenomenon include not only the complexity of gene–disease relationships but also methodologic shortcomings in the conduct, analysis, and reporting of the studies themselves (Davey Smith et al., 2006; Ioannidis et al., 2001). In addition to improved research approaches, the field can benefit from more rigorous application of established methods. The essential characteristics of a good association study were recently reviewed by Hattersley and McCarthy (2005). Beyond gene discovery, which is the traditional domain of genetic epidemiology, human genome epidemiology is primarily concerned with characterizing the effects of genetic factors in populations (Khoury et al., 2004a). The main objectives of human genome epidemiology are to measure the prevalence of genetic polymorphisms in well-defined populations, characterize genotype–phenotype associations, investigate gene– environment interactions, and evaluate the clinical validity and utility of genetic tests.
EPIDEMIOLOGIC STUDY DESIGNS Population-based study design is a benchmark in epidemiologic research, but the term is often misapplied. Strictly speaking, a study is population-based only if the subjects are representative of a population defined by geopolitical boundaries and demographic characteristics. In general, the results of a well-conducted
study provide a basis for inference only to the population from which the study sample was drawn. Nevertheless, studies based on a well-defined population subgroup (e.g., defined by demographic, personal, or clinical criteria) may offer useful insights into larger groups that share similar characteristics. Most epidemiologic studies are based on cross-sectional, cohort, or case–control designs. Each design is useful for addressing particular aspects of genetic variation in relation to disease outcomes. Cross-sectional studies ascertain risk factors and disease status at the same time. Cross-sectional studies are relatively quick and inexpensive, and they are often used to suggest avenues for further epidemiologic studies. Such studies can be used to estimate the prevalence of genetic variants (although variants associated with decreased survival may be underrepresented). Crosssectional studies may be unable to distinguish between the effects of genetic factors on disease occurrence or on sequelae and survival. However, because a person’s genotype remains the same throughout life, a cross-sectional study shares characteristics with a retrospective cohort study when genotypes are the only factors analyzed. Cohort studies follow two or more groups of people with different risk factors over time (either prospectively or retrospectively) looking for the development of disease. Although they are useful for studying rare exposures and describing natural history, cohort studies are usually lengthy, expensive, and not well suited to examining rare outcomes. Cohort studies yield a direct estimate of disease incidence or cumulative incidence, which can be interpreted in terms of relative and absolute risk. “Experimental” cohort studies (randomized, controlled trials) are ideal for evaluating the effects of gene–environment interactions or specific interventions, but they are often not feasible; “Mendelian randomization” has been proposed as an alternative approach (see “Gene–Environment Interaction”) (Davey Smith and Ebrahim, 2003). Retrospective cohort studies are attractive because genetic information is invariant and can be measured long after the study has ended; however, they are subject to the usual biases introduced by non-random selection of participants, exposures, or interventions. Although cohort studies are expensive and time-consuming, there is increasing global interest in conducting them under the umbrella of population biobanks (see Chapter 24) to study the effects of genes, environments, and their interactions on various disease outcomes throughout the lifespan (Austin et al., 2003; Collins, 2004). Case–control studies compare groups of persons with disease (cases) and without disease (controls) on the basis of risk factors that are ascertained retrospectively. Case–control studies are useful for studying multiple risk factors and rare outcomes, but not for studying rare risk factors. Case–control studies can generally measure gene–disease associations more quickly, efficiently, and at lower cost than cohort studies. Case–control or case–cohort studies can also be “nested” within existing cohort studies or population-based biobanks, an approach that combines advantages of both case–control (efficiency, cost) and cohort (causal inference, risk estimation) study designs.
Epidemiologic Measures of Disease Frequency, Association and Risk
TABLE 40.1
Characteristics
Measures of disease frequency
Cross-sectional
Persons assessed simultaneously for disease and risk factors
●
Persons with and without risk factors are followed up for disease
●
Case–Control
Persons with and without disease are studied for risk factors
●
●
Epidemiologic methods are based on comparing disease frequency in different groups. Measures of disease frequency in epidemiologic studies include the following:
●
●
Prevalence
Measures of association ● ●
EPIDEMIOLOGIC MEASURES OF DISEASE FREQUENCY, ASSOCIATION, AND RISK
●
463
Characteristics, measures of disease frequency, and measures of association in epidemiologic studies
Study design
Cohort
■
Incidence rate: The number of new (incident) cases of disease occurring during a defined time period, divided by the sum of all person-time at-risk during that period (expressed as cases per unit person-time). Cumulative incidence: The number of new cases of disease occurring during a defined time period, divided by the number of people under observation at the beginning of the time period; this measure corresponds to the risk or probability of developing disease, to the attack rate in traditional epidemiologic studies of infectious diseases, and to penetrance in genetics. Prevalence: The number of persons with disease at a specific point in time, which is a function of incidence and disease duration.
Measures of disease frequency and measures of association for cross-sectional, cohort, and case–control study designs are summarized in Table 40.1. The most basic analysis of crosssectional, cohort, and case–control studies can be represented as a 2 2 table, in which a given risk factor and disease are either present or absent (Table 40.2). Relationships among the elements of this table are the basis for estimating key parameters, including measures of risk (rates and ratios). Human genome epidemiology aims to estimate the contribution of genetic variants to the health of individuals and populations by estimating the associated absolute, relative, and attributable risks. Absolute risk is the probability that persons with a particular characteristic (e.g., a specific genotype) will develop disease. Absolute risk is estimated from cumulative incidence of disease in a well-defined population with the relevant characteristic. Only cohort studies can provide a direct estimate of absolute risk related to environmental exposures (or gene–environment
Incidence Cumulative incidence None
TABLE 40.2
Prevalence difference Relative prevalence (rate ratio)
●
Risk difference Relative risk (risk ratio) Odds ratio
●
Odds ratio
● ●
Measures of association in a 2 2 table Disease
Risk factor
Present
Absent
Total
Present
A
B
AB
Absent
C
D
CD
Total
AC
BD
A BC D
E proportion of persons exposed to risk factor who have disease A/(A B). U proportion of persons unexposed to risk factor who have disease C/(C D). Difference measure: E U A/(A B) C/(C D) Ratio measures: (Rate ratio, risk ratio, or prevalence ratio): E/U [A/(A B)]/[C/(C D)] Odds ratio: E/(1 E)/U/(1 U) AD/BC
interactions). Even though genotype is invariant, analyzing cross-sectional data by genotype as a retrospective cohort is problematic; in particular, the age distribution of the population must be taken into account. Measures of relative risk include the rate ratio, risk ratio, and prevalence ratio (Table 40.2). These ratios are calculated in the same way but inference depends on the study design. Cohort studies support direct estimation of relative risk. In case–control studies, the odds of exposure (or genotype) among cases, divided by the odds of exposure (or genotype) among controls, is identical to the odds of disease in cohort studies. Relative risk can be estimated from case–control studies by the odds ratio, which approximates the relative risk as long as the disease is rare or controls are sampled randomly (independent of disease status or genotype) from the source population (Rothman and Greenland, 1998). Two reasons why the odds ratio has become increasingly popular in epidemiologic studies are that not only can it be obtained from cross-sectional, cohort, and case–control studies but also it is amenable to common statistical techniques for multivariate analysis (e.g., logistic regression). Population attributable fraction (a measure of attributable risk) is the overall contribution of a particular risk factor (e.g., genotype)
464
CHAPTER 40
■
Principles of Human Genome Epidemiology
to the occurrence of disease in a given population. The term refers to the fraction of cases that would not have occurred within a certain time period had the risk factor been absent. Several formulas for calculating the attributable fraction have been proposed; one common formula, proposed by Miettinen, is: population attributable fraction f c (R 1)/R where fc is the fraction of cases with the risk factor, and R is the measure of relative risk (Miettinen, 1974). Attributable fraction is not a straightforward concept in the context of complex interactions among genetic and environmental factors that are poorly understood (Vineis and Kriebel, 2006). The epidemiologic concept of attributable fraction needs to be differentiated from the genetic concept of heritability. Heritability, a measure derived from analysis of variance or correlation, is sometimes used to describe the extent to which genotypic variation appears to explain phenotypic variation in a particular population. Heritability estimates are often based on studies of closely related persons (e.g., twins or other siblings), and they are highly dependent on context. They are not useful for determining the genetic or environmental contributions to disease or for estimating attributable fraction (Burton et al., 2005).
MEASUREMENT AND BIAS A well-conducted epidemiologic study first defines the sampling frame or target population from which participants will be selected (see “Epidemiologic Study Designs”). Population-based sampling is relatively straightforward in those few countries providing their people with universal health care and unique personal identifiers. In other areas, geographically defined disease registries offer a starting point (Goodman et al., 2005). The defining characteristic of a population-based case–control study is a set of a priori criteria – applied to the selection of controls as well as cases – that specifies the study’s source population (e.g., by geographic area and ethnicity). Systematic differences between cases and controls on such characteristics can introduce confounding and produce spurious results (Colhoun et al., 2003). Confounding introduced by selecting cases and controls with different genetic backgrounds (e.g., ancestry, race, or ethnicity) is termed population stratification. The potential effects of population stratification have been widely discussed, and several different statistical approaches have been proposed to test or control for it (Marchini et al., 2004). Studies comparing the genotypes of patients in a clinical case series with those of an undefined convenience sample of control subjects are numerous in the published literature, but they provide little basis for inference. An epidemiologic case definition is analogous to the genetic concept of phenotype. The case definition uses explicit, observable criteria to classify cases and non-cases. The case definition may include a time dimension, such as disease stage or diagnosis by a given age. Imprecise case criteria that disregard potentially important phenotypic differences, such as age at onset or key
clinical manifestations, can lead to case heterogeneity and misclassification. On the other hand, very narrow selection criteria limit the sample size available for study. Thus, the investigator faces a trade-off between specificity and power (Hattersley and McCarthy, 2005). If sufficiently detailed clinical data are collected systematically and the sample size is large enough, cases can be further stratified for analysis on the basis of phenotypic subgroup. Candidate genes are often selected for analysis on the basis of information from prior studies of gene function (in humans or animals) or association, although other data may also be available (e.g., from pharmacologic studies) (Hattersley and McCarthy, 2005). Because the molecular pathways underlying human health and disease are not well understood, candidate gene selection is as much art as science. Even within a candidate gene, particular variants (most often single nucleotide polymorphisms, or SNPs) must be selected for typing because determining the entire sequence in large numbers of study participants is still too expensive. Ideally, variants are selected because they are known to alter the structure and function of gene products; however, the functional effects of sequence variation are largely unexplored, especially in non-coding regions (e.g., regulatory elements, splice sites) (Todd, 2006). As increasingly powerful, fast, and inexpensive methods for genotyping become available, researchers are able to scan the entire genome for associations, raising the stakes for strategic SNP selection (Thomas et al., 2005). The International HapMap Project has approached this problem by using linkage disequilibrium to identify a reduced set of “tag” SNPs for capturing variation throughout the genome (International HapMap Consortium, 2005). Several platforms for whole-genome association studies have been evaluated by comparison with data from HapMap (Di et al., 2005; Gunderson et al., 2006). Genome-wide association studies are now demonstrating the potential to identify novel genetic variants associated with common, complex diseases (Manolio et al., 2008). New gene-chip technologies based on HapMap tag SNPs allow cost-efficient scanning of thousands of genetic variants in large numbers of people. For example, the Wellcome Trust Case Control Consortium has analyzed 500,000 genetic variants in approximately 2000 people affected with each of seven common diseases and in 3000 healthy people who served as control subjects (Wellcome Trust Case Control Consortium, 2007). The novel associations that they uncovered had modest effect sizes (i.e., per-allele odds ratios of 1.2–1.5); the sample sizes needed to measure such small effects are in the thousands (Khoury et al., 2007). Published reports of association studies often fail to address the potential effects of genotyping error, which can never be entirely eliminated even in the best laboratories. When random, the resulting misclassification reduces study power; however, systematic genotyping error can also introduce bias. Systematic errors may occur when cases and controls are genotyped in different laboratories or even in different batches in the same laboratory (Davey Smith et al., 2006). In studies of complex diseases, even true associations are likely to be small, and genotyping error can easily produce spurious associations of
Gene–Environment Interaction
similar magnitude (Hattersley and McCarthy, 2005; Ioannidis et al., 2006b). Until more journals require authors of association studies to discuss the performance characteristics of their genotyping assays and quality control measures, genotyping error must be considered as a potential explanation for virtually any reported association (Bogardus et al., 1999; Little et al., 2002). In the absence of a strong rationale for selecting or prioritizing candidate genes and variants, the association study that examines multiple genotypes for association with a disease is similar to an epidemiologic “fishing expedition.” In the setting of multiple comparisons, statistically significant associations are likely due to chance alone, resulting in type 1 error. No clear consensus exists on how to handle this problem. The goal of this type of study is to find true associations, all with a low (usually unknown) prior probability. A proposed approach that aims to optimize the balance between true-positive and false-positive findings is closer in outlook to screening than hypothesis testing (Storey and Tibshirani, 2003). In genetic association studies – whether they are candidate gene analyses or genome-wide association studies – small effects are to be expected (Ioannidis et al., 2006b). With rare exceptions, associations that are consistently replicated have relative risks on the order of 1.1–2.0 (Hattersley and McCarthy, 2005;Todd, 2006). Many association studies are conducted without sufficient power to detect effects of this size. Meta-analysis of published data is one approach that has proved useful, especially when several well-conducted studies have used the same case definition and genotyping methods to study an association in similar populations (GarciaClosas et al., 2005; Ioannidis et al., 2001). Meta-analysis of individual participant data (pooled analysis) requires a much greater commitment of time and resources than traditional meta-analysis of published data; however, it also has substantial advantages, including the ability to standardize definitions of cases and phenotypic subgroups and to achieve better control of confounding (Ioannidis et al., 2002). Joint efforts on the pooled analysis may create a focus for prospective collaboration to update the meta-analysis and conduct other collective analysis projects (see “Building the Evidence Base”) (Ioannidis et al., 2006a). Improving power and allowing examination of between-study heterogeneity are potential benefits of meta-analysis even in the setting of genome-wide association studies (Evangelou et al., 2007).
GENE–ENVIRONMENT INTERACTION As the basic science of public health, epidemiology aims to identify environmental causes of disease that are amenable to intervention. Much research in epidemiologic methods has focused on problems of measurement, analysis, and bias in evaluating the relationships of environmental factors with health outcomes. Nevertheless, because the physiologic and behavioral changes they bring about are dynamic, environmental exposures present far more difficult measurement and analysis challenges than simple genotypes. Case–control studies are particularly problematic
■
465
because exposures are assessed only after disease has developed in cases, often many years after the exposure actually occurred. So far, only a small fraction of published research in human genome epidemiology includes analysis of gene–environment interactions (Lin et al., 2006). Often, the results are presented in formats that do not permit evaluation of the marginal and joint effects of genetic and environmental factors. In the simplest case, this evaluation can be accomplished by using a “two-by-four table.” A simple gene–environment interaction model in the context of a cohort study is described in Table 40.3. Stratified analysis to estimate the risk of disease in persons with only environmental risk factor, with only the genetic factor, or with both can clarify whether gene–environment interactions exist; however, sample size often becomes the limiting factor in such analyses. For example, to detect with 80% power a strong interaction (twofold risk) between a genotype and exposure, each with 10% prevalence in the population, would require 1500 cases and the same number of controls (Hoover, 2007). For less common atrisk genotypes – especially those composed of variants at multiple loci, such as those proposed for “genomic profiling” – the necessary sample size is much larger. The case-only study has been proposed as an exploratory method for identifying interaction in a case series (Gatto et al., 2004; Khoury and Flanders, 1996). Under a multiplicative model of interaction where the genotype and exposure are independent, a departure of the case-only odds ratio from 1.0 indicates the presence of gene–environment interaction (Table 40.3). The case-only study is limited, however, because it cannot evaluate effects of genotype or exposure alone. Interactions observed at the population level do not translate easily into models of pathogenesis within individuals (Vineis
TABLE 40.3 A simple gene–environment interaction model in a cohort study Genotype
Environmental exposure
Disease risk
Relative risk
Absent
Absent
I
1
Absent
Present
IRe
Re
Present
Absent
IRg
Rg
Present
Present
IRge
Rge
I background risk of disease in persons without genotype or environmental exposure. Rerelative risk in persons with environmental exposure but not genotype. Rgrelative risk in persons with genotype but not environmental exposure. Rgerelative risk in persons with both genotype and environmental exposure. Re, Rg and Rge can also be estimated as odds ratios in a case–control study, where the reference group has neither the genotype nor the exposure. Two common statistical models of interaction are: 1. Additive model, where Rge Rg Re 1 and 2. Multiplicative model, where Rge Rg Re In a case-only study: ORcase-only ORge/(ORe ORg) 1 when there is no multiplicative gene–environment interaction.
466
CHAPTER 40
■
Principles of Human Genome Epidemiology
and Kriebel, 2006). However, Davey Smith and others have proposed the concept of “Mendelian randomization” as a way to use biological information to strengthen the evidence for the causal role of environmental factors from observational studies (Clayton and McKeigue, 2001; Davey Smith and Ebrahim, 2003). The basic idea is that if a genetic variant has a well-characterized biological function that imitates or reverses the effect of an environmental risk factor, then that variant should also have a predictable association with disease risk. The assumption that genotype is “randomized” at birth (i.e., independent of environmental exposure) provides a less-biased framework for interpreting observational data on gene–environment interactions. This approach could have important implications for traditional risk-factor epidemiology, in which the seemingly robust findings of case–control studies all too often prove illusory when tested in clinical trials (Davey Smith et al., 2006). At present, the further investigation and application of Mendelian randomization as an epidemiological tool is limited by incomplete understanding of human pathobiology, as well as by inadequacy of the knowledge base on gene–disease association (Davey Smith et al., 2006; Little et al., 2003). Advances in both areas will lead to additional opportunities to test the utility of this approach.
PROBABILITY AND PERSONALIZED MEDICINE Genomics research enjoys considerable public support because of its promise to deliver health benefits, including new therapies tailored to individual genotypes (Personalized Medicine Coalition, 2007). The concept of “personalized medicine” both inspires and frustrates researchers although, ironically, the most extreme optimistic and pessimistic viewpoints tend to stem from the same misconception: that “genomic tests” should be as predictive as the conventional genetic tests used to diagnose hereditary disorders resulting from single, high-penetrance gene variants (Collins and McKusick, 2001; Holtzman and Marteau, 2000). Indeed, this paradigm is out of date even for classic “single gene disorders” such as cystic fibrosis, in which DNA sequencing is revealing increasingly diverse genotype–phenotype relationships (Cutting, 2005). As observed by critics of individualized general preventive medicine, few risk factors for common chronic diseases have sufficient predictive ability to serve as screening tools (Wald et al., 1999); in this respect, common polymorphisms associated with disease susceptibility are unlikely to be different. Most risk estimates useful to individuals will be obtained only by considering the joint effects of many factors, including health history, personal characteristics, environmental exposures, and behaviors. This revelation surely comes as no surprise either to most medical practitioners, who are accustomed to integrating imperfect data from multiple sources to arrive at a clinical judgment, or to health care managers,
who strive to base decisions on evidence-based, predictive models (Bianchi and Alexander, 2006; Sipkoff, 2005). Whether genetic information adds value to prediction models based on clinical characteristics is an empirical question; one approach to evaluating it is by analyzing receiver operating characteristics (ROC) curves (Janssens et al., 2006). One proposed approach to predictive genomic testing is based on assessing the joint effects of multiple genes by “genomic profiling.” Although theoretically feasible, the evidence base for such applications is almost completely lacking (Haga et al., 2003; Khoury et al., 2004b). “Proteomic profiles” based on mass spectrometry have also been proposed as predictive tests, although their validity is even more difficult to establish (Myers et al., 2006). As large-scale genotyping and other sophisticated laboratory methods become increasingly available, technology is no longer the greatest challenge; instead, it is epidemiology (Khoury et al., 2004a). Functional information on gene products and biological pathways cannot substitute for the knowledge gained from genetic association studies (Todd, 2006). Only epidemiology has the capacity to integrate population-level information on intrinsic characteristics, exposures, and their interactions, while acknowledging the essential role of chance. Existing epidemiologic methods are challenged to collect, assimilate, analyze, synthesize, and interpret data on complex relationships among all of these factors in large numbers of people across the lifespan (De Stavola et al., 2006). The results – like those of all epidemiologic studies – will be expressed in terms of probabilities because they are observations of a stochastic process. Researchers, health care providers, and the media will be challenged to explain the results in terms of probabilities, which are often not intuitive (Bianchi and Alexander, 2006). Thus, paradoxical as it may seem, the key to the realization of the vision of personalized medicine is not contained within each person’s genome but derived from a synthesis of population-level research. The Human Genome Project was able to capture most of the information in the genome with a single sequence because all human genomes are 99.9% identical, but developing health applications will depend on understanding variation in the remaining 0.1%. Even this small fraction of the genome contains millions of SNPs, too many to study systematically for association with clinical phenotypes. The International HapMap Project has made a start by studying SNPs throughout the genomes of 269 persons from four ethnic groups (McVean et al., 2005). Clearly, expanding the vision of human genomics research to encompass individual variation and its role in health and disease is a challenge that dwarfs the Human Genome Project in complexity, scale, and scope.
BUILDING THE EVIDENCE BASE The Human Genome Project demonstrated the value of collaboration, standards, and synthesis in a large scientific enterprise. Now these principles can be a guide to building the evidence base on human genome epidemiology. Two major approaches
References
have been proposed: creation of “biobanks” and collaborative efforts to promote research synthesis. These approaches are not mutually exclusive. Prospective studies of large, representative populations have been proposed in many countries, including the United States (Austin et al., 2003; Collins, 2004). Of those currently underway, the largest is the UK Biobank, which aims to collect information on the health and lifestyle of 500,000 volunteers aged between 40 and 69; during 20 or more years of follow-up, this information – along with DNA samples – will be available for scientifically and ethically approved research (UK Biobank, 2006). The UK National Health Service, which provides health care to the entire population, provides the framework for enrolling and following up the biobank cohort. No comparable infrastructure exists in the United States, with its highly decentralized health system and mobile population. Population-based biobanks have the disadvantages of other prospective cohort studies, being slow, expensive, and inefficient for studying rare outcomes; thus, even where they are feasible, complementary approaches are desirable. New consortia, such as the Wellcome Trust Case Control Consortium and the Genetic Association Information Network (GAIN), have been organized to mine genetic association data within previously conducted, large-scale epidemiologic studies (Foundation for the National Institutes of Health, 2006;Wellcome Trust Case Control Consortium, 2006). A self-organized “network of networks” is extending the reach of formal consortia created by funding agencies to focus collaborative efforts on particular diseases, from construction of retrospective cohorts to research synthesis (Ioannidis et al., 2005). Such networks have a critical role in “harmonizing” research approaches, study methods, nomenclature, definitions of phenotype, and data presentation to promote the integration of research results, including meta-analysis and pooled analysis. By mobilizing the combined expertise and resources of researchers in a particular field, such efforts can reduce duplication of effort,
■
467
promote good research practices, and more quickly recognize promising findings and research gaps (Ioannidis et al., 2006a).
CONCLUSION In 1999, Francis Collins presented a memorable “hypothetical case study in 2010,” in which a young man undergoes a battery of DNA-based tests to predict his risk of chronic diseases (Collins, 1999). Since that scenario appeared in print, thousands of studies have examined genetic variants for association with chronic diseases; however, none of those on the hypothetical patient’s test report – or any others studied so far – has been found to predict lung cancer, colorectal cancer, or coronary heart disease well enough to be useful for screening healthy adults (Davey Smith et al., 2006). Nevertheless, several companies now offer “genetic profile” tests to practitioners and the public, along with personalized lifestyle recommendations, such as advice on dietary supplements. Although these tests are not currently regulated by the Food and Drug Administration, their potential to mislead consumers has been reviewed by Congress and further scrutiny is likely (Food and Drug Administration, 2006; United States Government Accountability Office, 2006). All too often, a new gene discovery is followed by a haphazard rush of scientific excitement, media coverage, and commercial interest in developing and marketing genetic tests, without the necessary intervening research (Janssens et al., 2006). Among the numerous reasons for this phenomenon is a profound misunderstanding of the relationship between observation and prediction, cause and effect at the individual and population levels (Vineis and Kriebel, 2006). By attempting to make these issues more explicit and to promote transparency, rigor, and synthesis in epidemiologic research that includes genetics, human genome epidemiology aims to improve the return on public investment in genomics.
REFERENCES Austin, M.A., Harding, S. and McElroy, C. (2003). Genebanks: A comparison of eight proposed international genetic databases. Community Genet 6, 37–45. Bianchi, M.T. and Alexander, B.M. (2006). Evidence based diagnosis: Does the language reflect the theory? BMJ 333, 442–445. Bogardus, S.T., Jr., Concato, J. and Feinstein, A.R. (1999). Clinical epidemiological quality in molecular genetic research: The need for methodological standards. JAMA 281, 1919–1926. Burton, P.R., Tobin, M.D. and Hopper, J.L. (2005). Key concepts in genetic epidemiology. Lancet 366, 941–951. Clayton, D. and McKeigue, P.M. (2001). Epidemiological methods for studying genes and environmental factors in complex diseases. Lancet 358, 1358–1360. Colhoun, H.M., McKeigue, P.M. and Davey Smith, G. (2003). Problems of reporting genetic associations with complex outcomes. Lancet 361, 865–872.
Collins, F.S. (1999). Shattuck lecture – medical and societal consequences of the Human Genome Project. N Engl J Med 341, 28–37. Collins, F.S. (2004). The case for a US prospective cohort study of genes and environment. Nature 429, 475–477. Collins, F.S. and McKusick, V.A. (2001). Implications of the Human Genome Project for medical science. JAMA 285, 540–544. Cordell, H.J. and Clayton, D.G. (2005). Genetic association studies. Lancet 366, 1121–1131. Cutting, G.R. (2005). Modifier genetics: Cystic fibrosis. Annu Rev Genomics Hum Genet 6, 237–260. Davey Smith, G. and Ebrahim, S. (2003). ‘Mendelian randomization’: Can genetic epidemiology contribute to understanding environmental determinants of disease?. Int J Epidemiol 32, 1–22. Davey Smith, G., Ebrahim, S., Lewis, S., Hansell, A.L., Palmer, L.J. and Burton, P.R. (2006). Genetic epidemiology and public health: Hope, hype, and future prospects. Lancet 366, 1484–1498.
468
CHAPTER 40
■
Principles of Human Genome Epidemiology
De Stavola, B.L., Nitsch, D., dos Santos Silva, I., McCormack, V., Hardy, R., Mann, V., Cole, T.J., Morton, S. and Leon, D.A. (2006). Statistical issues in life course epidemiology. Am J Epidemiol 163, 84–96. Di, X., Matsuzaki, H., Webster, T.A., Hubbell, E., Liu, G., Dong, S., Bartell, D., Huang, J., Chiles, R., Yang, G. et al. (2005). Dynamic model based algorithms for screening and genotyping over 100 K SNPs on oligonucleotide microarrays. Bioinformatics 21, 1958–1963. Evangelou, E., Maraganore, D.M. and Ioannidis, J.P. (2007). Meta-analysis in genome-wide association datasets: Strategies and application in Parkinson disease. PLoS ONE 7(2), e196. Food and Drug Administration. (2006). Draft Guidance for industry, clinical laboratories, and FDA staff on in vitro diagnostic multivariate index assays; availability [Docket No. 2006D–0347]. http:// www.fda.gov/OHRMS/DOCKETS/98fr/ch0641.pdf. Foundation for the National Institutes of Health (2006). Genetic Association Information Network (GAIN). http://www.fnih.org/ GAIN/GAIN_home.shtml. Garcia-Closas, M., Malats, N., Silverman, D., Dosemeci, M., Kogevinas, M., Hein, D.W., Tardon, A., Serra, C., Carrato, A., Garcia-Closas, R. et al. (2005). NAT2 slow acetylation, GSTM1 null genotype, and risk of bladder cancer: Results from the Spanish Bladder Cancer Study and meta-analyses. Lancet 366, 649–659. Gatto, N.M., Campbell, U.B., Rundle, A.G. and Ahsan, H. (2004). Further development of the case-only design for assessing geneenvironment interaction: Evaluation of and adjustment for bias. Int J Epidemiol 33, 1014–1024. Goodman, M.T., Hernandez, B.Y., Hewitt, S., Lynch, C.F., Cote, T.R., Frierson, H.F., Jr., Moskaluk, C.A., Killeen, J.L., Cozen, W., Key, C.R. et al. (2005). Tissues from population-based cancer registries: A novel approach to increasing research potential. Hum Pathol 36, 812–820. Gunderson, K.L., Kuhn, K.M., Steemers, F.J., Ng, P., Murray, S.S. and Shen, R. (2006). Whole-genome genotyping of haplotype tag single nucleotide polymorphisms. Pharmacogenomics 7, 641–648. Haga, S.B., Khoury, M.J. and Burke, W. (2003). Genomic profiling to promote a healthy lifestyle: Not ready for prime time. Nat Genet 34, 347–350. Hattersley, A.T. and McCarthy, M.I. (2005). What makes a good genetic association study?. Lancet 366, 1315–1323. Hoffman, E.P. (1994). The Evolving Genome Project: Current and future impact. Am J Hum Genet 54, 129–136. Holtzman, N.A. and Marteau, T.M. (2000). Will genetics revolutionize medicine?. N Engl J Med 343, 141–144. Hoover, R. (2007). The evolution of epidemiologic research: From cottage industry to “big science”. Epidemiology 18, 13–17. International HapMap Consortium (2005). A haplotype map of the human genome. Nature 437, 1299–1320. Ioannidis, J.P., Ntzani, E.E., Trikalinos, T.A. and Contopoulos-Ioannidis, D.G. (2001). Replication validity of association studies. Nat Genet 29, 306–309. Ioannidis, J.P., Rosenberg, P.S., Goedert, J.J. and O’Brien, T.R. (2002). International Meta-analysis of HIV Host Genetics. Commentary: Meta-analysis of individual participants’ data in genetic epidemiology. Am J Epidemiol 156, 204–210. Ioannidis, J.P., Bernstein, J., Boffetta, P., Danesh, J., Dolan, S., Hartge, P., Hunter, D., Inskip, P., Jarvelin, M.R., Little, J. et al. (2005). A network of investigator networks in human genome epidemiology. Am J Epidemiol 162, 302–304.
Ioannidis, J.P., Gwinn, M., Little, J., Higgins, J.P., Bernstein, J.L., Boffetta, P., Bondy, M., Bray, M.S., Brenchley, P.E., Buffler, P.A. et al. (2006a). A road map for efficient and reliable human genome epidemiology. Nat Genet 38, 3–5. Ioannidis, J.P., Trikalinos, T.A. and Khoury, M.J. (2006b). Implications of small effect sizes of individual genetic variants on the design and interpretation of genetic association studies of complex diseases. Am J Epidemiol 164, 609–614. Janssens, A.C., Gwinn, M., Valdez, R., Narayan, K.M. and Khoury, M.J. (2006). Predictive genetic testing for type 2 diabetes. BMJ 333, 509–510. Khoury, M.J. and Flanders, W.D. (1996). Nontraditional epidemiologic approaches in the analysis of gene-environment interaction: Casecontrol studies with no controls! Am J Epidemiol 144, 207–213. Khoury, M.J., Millikan, R., Little, J. and Gwinn, M. (2004a). The emergence of epidemiology in the genomics age. Int J Epidemiol 33, 936–944. Khoury, M.J., Yang, Q., Gwinn, M., Little, J. and Flanders, W.D. (2004b). An epidemiologic assessment of genomic profiling for measuring susceptibility to common diseases and targeting interventions. Genet Med 6, 38–47. Khoury, M.J., Little, J., Gwinn, M. and Ioannidis, J.P. (2007). On the synthesis and interpretation of consistent but weak gene-disease associations in the era of genome-wide association studies. Int J Epidemiol 36, 439–445. Lin, B.K., Clyne, M., Walsh, M., Gomez, O., Yu, W., Gwinn, M. and Khoury, M.J. (2006). Tracking the epidemiology of human genes in the literature: The HuGE Published Literature database. Am J Epidemiol 164, 1–4. Little, J., Bradley, L., Bray, M.S., Clyne, M., Dorman, J., Ellsworth, D.L., Hanson, J., Khoury, M.J., Lau, J., O’Brien, T.R. et al. (2002). Reporting, appraising, and integrating data on genotype prevalence and gene-disease associations. Am J Epidemiol 156, 300–310. Little, J., Khoury, M.J., Bradley, L., Clyne, M., Gwinn, M., Lin, B., Lindegren, M.L. and Yoon, P. (2003). The Human Genome Project is complete. How do we develop a handle for the pump? Am J Epidemiol 157, 667–673. Manolio, T.A., Brooks, L.D. and Collins, F.S. (2008). A HapMap harvest of insights into the genetics of common disease. J Clin Invest 118, 1590–1605. Marchini, J., Cardon, L.R., Phillips, M.S. and Donnelly, P. (2004). The effects of human population structure on large genetic association studies. Nat Genet 36, 512–517. McClain, M.R., Palomaki, G.E., Nathanson, K.L. and Haddow, J.E. (2005). Adjusting the estimated proportion of breast cancer cases associated with BRCA1 and BRCA2 mutations: Public health implications. Genet Med 7, 28–33. McVean, G., Spencer, C.C.A. and Chaix, R. (2005). Perspectives on human genetic variation from the HapMap Project. PLoS Genet 1, e54. DOI: 10.1371/journal.pgen.0010054. Miettinen, O.S. (1974). Proportion of disease caused or prevented by a given exposure, trait or intervention. Am J Epidemiol 99, 325–332. Myers, E.R., Havrilesky, L.J., Kulasingam, S.L., Sanders, G.D., Cline, K.E., Gray, R.N., Berchuck, A. and McCrory, D.C. (2006). Genomic tests for ovarian cancer detection and management. Evidence Report/Technology Assessment No. 145. (Prepared by the Duke University Evidence-based Practice Center under Contract No. 290-02-0025.) AHRQ Publication No. 07-E001. Rockville, MD: Agency for Healthcare Research and Quality. http://www.ahrq. gov/downloads/pub/evidence/pdf/genomicovc/genovc.pdf.
Recommended Resources
National Library of Medicine. Online Inheritance in Man (OMIM). http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?dbOMIM. Palmer, L.J. and Cardon, L.R. (2005). Shaking the tree: Mapping complex disease genes with linkage disequilibrium. Lancet 366, 1223–1234. Personalized Medicine Coalition (2007). http://www.personalizedmedicinecoalition.org/. Rothman, K.J. and Greenland, S. (1998). Case-control studies. In Modern Epidemiology 2nd edition (K.J. Rothman, ed.). Lippincott Williams & Wilkins, Philadelphia, pp. 93–114. Sipkoff , M. (2005). Predictive modeling and genomics: Marriage of promise and risk. Manag Care 14, 60 63–64, 66. Storey, J.D. and Tibshirani, R. (2003). Statistical significance for genomewide studies. Proc Natl Acad Sci USA 100, 9440–9445. Teare, M.D. and Barrett, J.H. (2005). Genetic linkage studies. Lancet 366, 1036–1044. Thomas, D.C. (2006). High-volume “-omics” technologies and the future of molecular epidemiology. Epidemiology 17, 490–491. Thomas, D.C., Haile, R.W. and Duggan, D. (2005). Recent developments in genomewide association scans: A workshop summary and review. Am J Hum Genet 77, 337–345.
■
469
Todd, J.A. (2006). Statistical false positive or true disease pathway? Nat Genet 38, 731–733. UK Biobank (2006). UK Biobank: Improving the health of future generations. http://www.ukbiobank.ac.uk/. United States Government Accountability Office. (2006). Testimony before the Special Committee on Aging, U.S. Senate. Nutrigenetic testing: Tests purchased from four web sites mislead consumers. http://www.gao.gov/new.items/d06977t.pdf. Vineis, P. and Kriebel, D. (2006). Causal models in epidemiology: Past inheritance and genetic future. Environ Health 5, 21. Wald, N.J., Hackshaw, A.K. and Frost, C.D. (1999). When can a risk factor be used as a worthwhile screening test? BMJ 319, 1562–1565. Wellcome Trust Case Control Consortium (2006). http://www.wtccc. org.uk/. Wellcome Trust Case Control Consortium (2007). Genome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 447, 661–678.
RECOMMENDED RESOURCES GeneTests. http://www.geneclinics.org/. A publicly funded medical genetics information resource developed for physicians, other health care providers, and researchers and available at no cost. Includes expert-authored disease reviews, laboratory directory, and clinic directory. Human Genome Epidemiology Network (HuGENet). http://www. cdc.gov/genomics/hugenet/default.htm. Website for a global collaboration of individuals and organizations committed to the assessment of human genome variation in population health. Includes a searchable database of publications in human genome epidemiology compiled weekly from PubMed. Links provide information on becoming a HuGENet collaborator and connect with other HuGENet hubs in Cambridge (UK), Ottawa (Canada), and Ioannina (Greece). HuGENet Network of Networks. http://www.hugenet.org.uk/ networks/list.html. Directory of participating networks, maintained by the HuGENet UK hub. International HapMap Project. http://www.hapmap.org/. Website for an international partnership of scientists and funding agencies to identify and catalog genetic similarities and differences in human beings, which affect health, disease, and individual responses to medications and environmental factors.
Khoury, M.J., Beaty, T.H. and Cohen, B.H. (1994). Fundamentals of Genetic Epidemiology. Oxford University Press, Oxford. Textbook of genetic epidemiology. Khoury, M.J., Little, J. and Burke, W. (2004). Human Genome Epidemiology. Oxford University Press, Oxford. Overview of human genome epidemiology in 29 chapters by experts, describing methods and case studies. National Institutes of Health. National Human Genome Research Institute. Human Genome Project. http://www.genome.gov/. Website for the international, publicly funded Human Genome Project. Online Mendelian Inheritance in Man (OMIM). http://www.ncbi. nlm.nih.gov/sites/entrez?dbOMIM. Searchable public online database of human genes and diseases. Although the original focus was on Mendelian disorders, it now includes information on genes associated with common, complex diseases. Public Population Project in Genomics (P3G). http://www.p3gconsortium.org/. Website for a not-for-profit, international consortium to promote collaboration among researchers and “harmonization” of data management in the field of population genomics with the aim of improving knowledge transfer and sharing.
CHAPTER
41 Genomics and Population Screening: Example of Newborn Screening John D. Thompson and Michael Glass
INTRODUCTION Newborn screening (NBS) is a population screening program that tests babies to identify those who have treatable congenital disorders and strives to link these babies into the specialty care systems that will provide treatment and long-term support for them and their families. The focus of this chapter will be on NBS using dried blood specimens, usually collected from the newborn’s heel, absorbed onto filter paper and dried before shipping to a screening laboratory. The argument for administering NBS as a population screening program (rather than a screening tool for at-risk populations) is so compelling that offering NBS is mandated by law in every state in the United States and in many other countries.1 Because the disorders are so rare, most affected babies have no
1
Newborn screening was first introduced in the United States in the 1960s. The subsequent widespread implementation of newborn screening in all developed countries and in most developing nations is a testament to the critical role this public health program plays in the lives of newborns. The authors both work for the Washington State Newborn Screening Program and are therefore most familiar with dried blood spot newborn screening practices in the United States. We believe, however, that the issues discussed in this chapter are relevant to programs throughout the world.
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 470
family history which would indicate a need for testing. Severe neurological and physical problems and death may result if the conditions are not treated soon after birth, so NBS conditions are considered to be public health emergencies requiring immediate attention (Grosse et al., 2006). Although a key component, NBS is a small piece of the system in place for newborn health. In the hours after birth, babies are checked for vital signs, measured, weighed, administered Hepatitis B vaccine and Vitamin K shots, given Erythromycin ointment in both eyes and screened for early hearing loss and elevated bilirubin. Dried blood spot NBS, or “the PKU test” as it is often misleadingly called, is just one of many events for infants that occur prior to leaving the hospital or birthing center. As technologies rapidly advance, the number of conditions that could be included in NBS panels continues to grow. Traditionally, disorders in NBS programs meet the following inclusion criteria: 1. Identifying the condition early provides a clear health benefit to the newborn 2. Irreversible harm is done before the condition is identified clinically 3. A sensitive and specific test is available, adapted for highthroughput screening
Screening Technology: Simple Ideas, Complex Realities
4. Treatments and systems of care are available for babies with the condition 5. The benefits justify the costs of screening2 Because the conditions are relatively rare, NBS is often lost in the bustle of everything else going on during post-partum care (Clayton, 2005). For the vast majority of babies, NBS is a poke in the foot and nothing more. But for the few affected individuals, it makes a world of difference.
COMPONENTS OF THE NBS SYSTEM Nurses, midwives and phlebotomists caring for babies in hospitals, birthing centers, clinics and at home are at the front line of NBS. They are charged with the important task of drawing quality specimens with accurate demographic information for every infant and delivering them in a timely fashion to the screening laboratory. These four aspects are imperative to the success of any NBS program. Poor quality, inaccurate or incomplete demographics, babies without screens or slow delivery can significantly delay diagnosis and treatment of an affected baby with catastrophic consequences. For babies with certain disorders (classic galactosemia and salt-wasting congenital adrenal hyperplasia are examples), the delay of an extra day or two could mean the difference between life and death. Testing for the various conditions typically begins the same day the specimens are received by the NBS laboratory and most abnormal results are identified within 1 to 2 days of receipt. Results can be classified as screen negative (normal), screen positive (abnormal) and QNS (quality/quantity not suitable for testing), although terms vary between programs. NBS programs may employ a variety of reporting mechanisms for screen negative results: mail, fax or electronic posting to either the submitter or the primary care provider. Most programs stratify their response for a given screen positive result based on the likelihood of the infant truly having the condition. Extreme values trigger an aggressive follow-up that includes contacting the primary care provider directly with results and recommendations for diagnostic testing. For NBS programs that receive two routine specimens for each child, a mildly abnormal result (often called “borderline”) may require no action beyond monitoring to ensure that a second specimen is received for that baby. Poor quality specimens are unsuitable for testing and necessitate follow-up to ensure that each child has a valid specimen submitted for screening. Short-term follow-up services include efforts to ensure that every child has appropriate testing and proper response based on the results. A key component is the sometimes aggressive follow-up to provide timely diagnostic testing and linkage
2
Several NBS cost analyses have been performed. They have generally supported screening, although their individual assumptions and estimates vary considerably. A review of NBS cost analyses was published recently as an invited commentary by Dr. Scott Grosse, an economist at the Centers for Disease Control and Prevention (Grosse, 2005).
■
471
of affected babies and their families with specialty care providers and treatment facilities. NBS programs may have additional follow-up components that monitor the long-term health and developmental progress of affected individuals and helps provide assistance to families in need. Education plays several critical roles in NBS programs. Hospitals and clinics must understand the NBS system and how to best perform their role. Primary care providers and the families of affected babies need information about these rare conditions tailored to their different levels of understanding complex medical conditions. The general public can learn information about NBS from the various publications distributed by NBS programs and via the internet. Many excellent websites provide current information about NBS for practitioners, parents and the public (please see the recommended resources section at the end of this chapter). Success in NBS programs necessitates a seamless system from birth to specialty care treatment for affected babies. The NBS system is only as good as its weakest link, so constant efforts to strengthen and improve its individual components are critical to overall success. NBS laboratories and follow-up groups are constantly involved in improving the program through quality assurance and quality control measures. Laboratory staff ensure that instruments and testing systems are operating at peak performance on a daily basis. All programs in the United States and many others throughout the world participate in the proficiency testing program run by the Centers for Disease Control and Prevention (CDC) (CDC, 2006). Monitoring specimen quality and transit times and providing regular feedback to hospitals have improved compliance in Washington State (Resler et al., 2005). Testing (especially screening) methods inherently have imperfect sensitivities and specificities. An NBS test, however, needs great sensitivity to identify the affected newborns, and because of the low prevalence of these disorders, it needs exquisite specificity to manage the number of false positive results. For example, consider a hypothetical screening test that has a perfect (100%) sensitivity for a condition with a prevalence of one in 10,000. If the specificity of the screening test were 95%, there would be 500 false positives for every true positive. In reality, the specificities of NBS tests are typically well above 99%, but this example illustrates the importance of recognizing the imperfections of the screening process and the potential magnitude of false positive results, which can lead to misdiagnosis and parental anxiety (Tarini et al., 2006;Waisbren, Albers, Amato et al., 2003). Advances in technology have played a key role in improving the utility of NBS by providing tests that meet these demanding sensitivity and specificity requirements.
SCREENING TECHNOLOGY: SIMPLE IDEAS, COMPLEX REALITIES Technological advances have fostered new methods and also allowed expansion of the number of disorders available for
472
CHAPTER 41
■
Genomics and Population Screening: Example of Newborn Screening
screening. Historically, NBS programs have been very responsive to advances in technology. Examples include the quick uptake of the radioimmunoassay technique by NBS programs in the 1970s when a test for primary congenital hypothyroidism was described (Hannon et al., 1993) and the contemporary implementation of tandem mass spectrometry by many NBS programs (Sweetman, 2001). Table 41.1 provides a brief description of technologies commonly used in NBS and examples of the disorders they are used to detect. NBS expansion is driven by both advances in technology and by the desire to screen for certain conditions. Often it is difficult to distinguish the predominant factor in these improvements. However, disorders are not generally added to an NBS panel solely because a robust screening test is available. Candidate disorders must also meet the other inclusion criteria described earlier. Implementing screening for each new condition has presented unique challenges for public health programs. Most of the disorders have complex etiologies that are imperfectly understood at the initiation of screening, but benefit from the experience of early identification and treatment. The following four examples highlight how population-based NBS is much more complex than the simple idea of identifying affected babies and putting them on treatment to avoid negative outcomes. For each section, the impetus for NBS is explained, the screening platform is described, and some of the challenges and implications (including unintended consequences) of screening for the disorder are reviewed.
TABLE 41.1
Phenylketonuria Impetus: Phenylketonuria (PKU) is a metabolic disorder caused by deficiency in the enzyme that converts the amino acid phenylalanine to the amino acid tyrosine. Untreated PKU leads to a build-up of phenylalanine that causes central nervous system damage. In the 1950s, researchers found that people with PKU have improved health if placed on a phenylalanine-restricted diet, but that any neurological damage suffered by the individual is irreversible (Armstrong et al., 1957). The simple idea for NBS was to identify those babies with PKU early in infancy, remove phenylalanine from their diets and avoid mental retardation. Platform: Hospital-based screening for PKU was utilized in some locations during the 1950s, but it was performed by a ferric chloride urine “diaper test” that was unreliable, particularly with younger infants (Guthrie, 1961). In the early 1960s, Dr. Robert Guthrie developed a better method for testing newborns for PKU called the bacterial inhibition assay (BIA). In this test, punches from dried blood spots are placed on agar infused with bacteria spores and a chemical that inhibits their growth. Phenylalanine reverses the inhibition, allowing development of a zone of bacterial growth where phenylalanine has diffused into the agar. The size of the zone is proportional to the concentration of phenylalanine in the blood spot (Aldis et al., 1993; Guthrie et al., 1963). Challenges and Implications: In the beginning years of NBS, administering treatment for babies with PKU was a challenge (Kennedy et al., 1967). There were no comprehensive systems of screening and care for those diagnosed with PKU, so some
Newborn screening technologies
Technology
Brief description
Examplesa
Bacterial inhibition assay
Metabolic product or precursor in baby’s blood inactivates inhibitor resulting in bacterial growth zone proportional to concentration
Phenylketonuria, Maple syrup urine disease
DNA testing
Genotyping is used as a second-tier test to increase the screen’s sensitivity; multiplex platformb
Cystic fibrosis, Hemoglobinopathies
Enzyme assays
Enzyme ability to convert substrate into product is measured
Biotinidase deficiency, Galactosemia
High performance liquid chromatography
Proteins or amino acids are sorted and quantified using pressurized chromatography
Hemoglobinopathies, Phenylketonuria
Immunoassay
Amount of protein or hormone is measured by competitive binding with labeled protein for sites on specific antibodies
Congenital hypothyroidism, Cystic fibrosis
Isoelectric focusing
Proteins are sorted on a pH gradient gel within an electrical field based on their relative charge; multiplex platformb
Hemoglobinopathies
Tandem mass spectrometry
Chemicals characteristic of metabolic disorders are sorted and quantified based on mass and characteristic fragmentation products; multiplex platformb
Isovaleric acidemia, MCAD deficiency
a
This is not a comprehensive listing of disorders detected by the different technologies. Multiplex platform means that this technology can be used to screen for many different conditions or genotypes simultaneously using the same blood sample. b
Screening Technology: Simple Ideas, Complex Realities
babies slipped through the cracks and did not receive treatment. Some babies were given formula completely void of phenylalanine and became malnourished. The lesson learned was that people with PKU need a very small amount of phenylalanine in their diets for normal development. Also, many believed that the restricted diet only needed to be followed during early childhood when the nervous system was rapidly developing. Current recommendations are for patients with PKU to stay on the diet for life (National Institutes of Health, 2001). The variability of expression in people with PKU is not well understood. Some people with PKU must strictly adhere to the dietary regimen to keep their blood phenylalanine at acceptable values, while others do not need to follow the diet as carefully to achieve the same results. This has even been observed in siblings, with presumably the same PKU-causing mutations, who are affected differently by the disorder. A delicate dietary balance must be found for each patient with PKU. Typically, babies with PKU have their phenylalanine levels monitored as they grow to determine the proper level of treatment. NBS for PKU is a good representation of the intricacies involved in treating a metabolic disorder with significant phenotypic variability. The PKU screening story became more complex when women with PKU, identified through NBS, began to have children of their own. Experience revealed that babies born to PKU mothers with high phenylalanine levels during their pregnancies have an increased risk for microcephaly, mental retardation and congenital heart disease even though the babies do not have PKU. This condition was named maternal PKU (National Institutes of Health, 2001). Data from the Maternal PKU Collaborative Study show that there is a negative correlation between maternal phenylalanine levels and both cognitive and behavioral outcomes of their babies, meaning that women with PKU who do not maintain blood phenylalanine levels within recommended limits are more likely to have children with disabilities (Waisbren and Azen et al., 2003). Another complexity that only became apparent through screening experience is that approximately 1% of babies with positive PKU screens have a related disorder called biopterin deficiency. Tetrahydrobiopterin (BH4) is a cofactor for the enzyme that converts phenylalanine to tyrosine. It is impossible to distinguish between the two conditions based solely on NBS results, so part of the diagnostic testing for PKU includes a workup for biopterin deficiency. Complicating matters further, three different enzymes are responsible for BH4 biosynthesis. Treatment and success at managing biopterin deficiency vary based on which enzyme is deficient (Blau et al., 2001; Naylor, 1985). Because both the screening methodology and appropriate treatment have been greatly improved over 40 years of experience, NBS for PKU is widely regarded as a shining example of a population-based screening program. Dr. Guthrie and parent advocates were influential in mandating PKU screening in the beginning, even though the quality of the scientific evidence justifying the creation of screening programs was questioned at the time. However, they were right when they claimed that it would be beneficial to affected newborns. Today, thousands of
■
473
people around the world have been spared the negative effects of untreated PKU. Sickle-Cell Anemia Impetus: Since the 1960s, it has been known that babies and young children with sickle-cell anemia (hemoglobin SS or Hb SS) are particularly susceptible to life-threatening bacterial infections. In 1983, a randomized controlled trial was organized to examine the effect of prophylactic therapy with oral penicillin in children with Hb SS. The early data about the benefits of treatment were so compelling that the investigators stopped the study 8 months early (Gaston et al., 1986). Shortly thereafter, the National Institutes of Health (NIH) held a consensus conference and published its recommendations for universal neonatal screening for Hb SS with prophylactic treatment before 4 months of age (Consensus Conference, 1987). Sickle-carrier screening programs predated NBS for sicklecell disease, which was widely implemented only after a strong public health rationale was proven.3 Infants with Hb SS are susceptible to rapid onset of overwhelming pneumococcal sepsis that in some cases leads to death. These children are also at risk for sudden pooling of blood in the spleen, acute chest syndrome, and stroke, any of which can be life-threatening. Studies showed that parental education about early warning signs of sickling crises, such as enlarged spleen or temperature above 38.3°C (101°F), helped reduce the mortality associated with Hb SS (Vichinsky et al., 1988).The simple idea for NBS was to identify those infants with Hb SS and enroll them into comprehensive care, including prophylactic administration of penicillin and parental education about early warning signs of complications of the disease. This would save lives and improve outcomes for babies with Hb SS. Platform: At the time of the 1987 consensus conference, hemoglobin screening was performed by most laboratories using cellulose acetate and citrate-agar electrophoresis (Garrick et al., 1973). A newer technique called isoelectric focusing (IEF) provided better resolution of hemoglobins and was successfully adapted for mass screening (Kleman et al., 1989). Isoelectric focusing operates upon the principle that hemoglobin molecules migrate to specific positions on a gel pH gradient when placed within an electric field. This position is determined by the composition of amino acids present in the protein. Hemoglobin variants, such as the sickle cell mutation, migrate to a different location on the gel than the normal hemoglobin protein. Hemoglobin molecules of a specimen are separated by IEF and stained for visualization. Technicians make classifications based on where the molecules migrate within the gel. High performance liquid chromatography (HPLC) testing is an alternative method used by some laboratories
3 Sickle-cell carrier screening programs began in the 1970s, and in many instances, screening was required only in black populations. Generally, there were not good protections for confidentiality and adequate counseling was not always a part of these programs. This led to misinterpreted results, stigmatization, and in some cases discrimination. Many people who were carriers of the sickle-cell mutation thought they had sickle-cell disease (Hampton, et al., 1974; Markel, 1997).
474
CHAPTER 41
■
Genomics and Population Screening: Example of Newborn Screening
for hemoglobin screening, often as a second-tier test on the original blood specimen for babies with initial abnormal results. Also, a few laboratories employ direct DNA testing as a second-tier test to better refine the screening results. Challenges and Implications: From a genetic standpoint, Hb SS is the most simple of diseases: it is caused by a single nucleotide change in the DNA. This mutation substitutes a valine for glutamic acid at the sixth amino acid residue in the -globin protein and causes the hemoglobin to polymerize under conditions of low oxygen concentration. This makes the red blood cells change from their normal doughnut-like shape to a long, thin sickle shape. Sickled cells have difficulties passing through the body’s capillaries and can cause episodes of severe pain and other complications (Lane, 1996). Despite its genetic simplicity, Hb SS exhibits a wide phenotypic variability. Although everyone with Hb SS has the same single mutation in both copies of the β-globin gene, some suffer greatly from painful sickling episodes that cause repeated, frequent hospitalizations while others experience few consequences from the disorder. Screening for Hb SS using IEF comes with several unintended consequences because of the qualitative nature of the test. Babies (and families) with hemoglobin variants other than the sickle mutation, and both α and β thalassemia are identified. Another unavoidable consequence of using IEF is the identification of unaffected carriers of the sickle and other variant hemoglobins. The majority of hemoglobin variants are benign conditions, but a few of them are clinically significant, alone and in combination with the sickle mutation (e.g., SC disease, SE disease, Sβ thalassemia) and require specialized medical care from a hematologist. Only in rare instances are carriers of the sickle mutation or other variant hemoglobins affected clinically, although studies have shown that this concept can be difficult for families to understand (Hampton et al., 1974). Abnormal hemoglobin detection, including carrier status, is generally reported to the baby’s primary care provider, and genetic counseling services are recommended when a clinically significant mutation is detected (Council of Regional Networks for Genetic Services, 1997). Even though the first NBS programs started screening for hemoglobins in the 1970s, there remains much to learn about the clinical presentation of these disorders. Additionally, access to services for affected individuals continues to be a challenge in some places. Cystic Fibrosis Impetus: Cystic Fibrosis (CF) is a condition that affects the way the body regulates salt transport across cell membranes. It causes thick mucus to build up in the lungs, intestines and other organs, leading to failure to thrive, digestive problems, nutritional deficits and chronic lung problems. The simple idea for NBS was to identify babies with CF and start treatment early to improve clinical outcomes, with the hope that a cure will someday be found. Platform: An immunoassay was developed in 1979 to measure immunoreactive trypsinogen (IRT), a pancreatic protein that is elevated in the first months of life in babies with CF (Crossley
et al., 1979). NBS for CF began shortly thereafter as pilot studies and randomized control trials in parts of Australasia, Europe and the United States (Hammond et al., 1991; Heeley et al., 1982; Wilcken et al., 1983). A baby with an elevated IRT level would need a second NBS specimen collected at 2–4 weeks of age. If the second specimen also had elevated IRT levels, the baby would be referred for a diagnostic work-up (sweat chloride test). Because elevated IRT levels are not specific to babies with CF, the false positive rate on the initial blood spot is very high. The discovery of the cystic fibrosis transmembrane conductance regulator (CFTR) gene in 1989 and advances in genotyping technology have changed the way that most NBS for CF is performed. The trend among NBS programs has been to adopt DNA testing as a second-tier screen (after finding elevated IRT) using a multiple-mutation panel. These panels exist or can be created to test for the most common mutations in a specific population or subpopulations. The cost of this technology has dramatically decreased over the years and the methods have been adapted to mass screening. Challenges and Implications: Because this genotyping is performed from the original blood spots, using DNA testing offers the advantage of finding babies with two identified mutations and establishing a genetic diagnosis of CF without requiring a second blood specimen. This testing, however, also identifies many more babies who have only one mutation detected. For these infants, a diagnostic sweat test is necessary because even the most comprehensive DNA panels only include a fraction of the known CF-causing mutations. Most of these babies with only one mutation detected are simply unaffected carriers, meaning that one gene is normal and the other has a mutation. Another challenge arises from the differing severities of the CF mutations. In a panel of multiple mutations, some alleles may result in a late onset form of CF or may never be clinically significant. It is difficult to predict the effects of certain combinations of alleles. Furthermore, DNA testing results may be difficult for some families to understand. Cystic fibrosis is most prevalent among Caucasians and the majority of research has been done on the alleles most common in this population. The disease, however, is found among all ethnicities. The current DNA panels do a poor job of identifying CF in non-Caucasian babies (Comeau et al., 2004). Implementing direct-DNA testing of newborns for CF mutations has significant implications. One is the identification of unaffected carriers discussed earlier. During a pilot study in Massachusetts, eleven carriers were identified by screening for each true case of CF (Comeau et al., 2004). Because DNA testing has implications for reproductive decision-making for some families, genetic counseling is generally offered to those families that have babies with either one or two CF mutations detected. While programs may feel differently about whether carrier identification is a positive or negative consequence of screening, it is costly for society because of the increase in required diagnostic testing and recommended genetic counseling. New mutations added to NBS panels will improve the screening sensitivity. However, for each mutation added, the
Screening Technology: Simple Ideas, Complex Realities
number of unaffected carriers that will require sweat testing increases dramatically (Comeau et al., 2004). NBS programs using multiple mutation panels must balance the benefits and risks of increasing their detection rate because it comes at the cost of increased carrier identification. Recent data from the state of Colorado (Sontag et al., 2005) show that when a NBS program measures IRT levels on two specimens (one at 24–48 h after birth and another at about 14 days of age), similar sensitivity and negative predictive values can be achieved by screening for persistent IRT elevation in two specimens compared to using DNA with a multiple-mutation panel. This means that both algorithms would identify about the same number of babies with CF and miss about the same number of babies with CF over time (Thompson et al., 2005). These new data have some programs thinking carefully about the pros and cons of using direct DNA testing as a second-tier screen for CF (personal communications). MCAD Deficiency Impetus: Mass spectrometry is a powerful analytic tool to study mixtures of compounds. A group of scientists and metabolic specialists teamed in the 1980s to develop an application of the technology called tandem mass spectrometry (MS/MS) for NBS. They used MS/MS to measure acylcarnitines which accumulate in certain fatty-acid oxidation and amino acid disorders (Millington, 2002; Millington et al., 1990). Medium-chain acyl-CoA dehydrogenase (MCAD) deficiency is the most common and best understood of the fattyacid oxidation disorders. It is caused by a defect in one of the enzymes that participates in the beta oxidation cycle of fat breakdown. Carbohydrates are the only safe energy source for people with MCAD deficiency because toxic intermediate metabolites build up if they are forced to break down fats, for example during fasting or illness (Roe et al., 2001). The case fatality rate for clinically diagnosed MCAD deficiency is estimated to be about 16% (Pollitt et al., 1998). The simple idea for NBS was to identify babies with MCAD deficiency and put them on treatment (avoiding fasting or fasting-like states during illness) to prevent death and disability. Platform: Tandem mass spectrometers are complex instruments that separate the array of molecules in a blood sample. The molecules are first sorted by size in an initial mass spectrometer. The molecules then pass through a collision chamber where they are broken into pieces that are of characteristic size, dependent on their molecular structures. The fragments are then sorted by size in the second mass spectrometer and then quantified. The results of the analysis are plotted on a graph: the x-axis represents the mass of the different molecules, and the height of each peak on the y-axis represents the amount of that molecule present in the sample (Chace et al., 2003). Challenges and Implications:Tandem mass spectrometers can be used to examine a profile of metabolites for an individual, identifying abnormal levels and ratios of specific components in the blood characteristic of metabolic diseases. The MS/MS platform can be used to simultaneously screen for 30 or more disorders
■
475
using one blood specimen. While it is generally accepted that MCAD deficiency meets the traditional criteria for inclusion in NBS programs, some of the other disorders detectable through MS/MS clearly do not. There are yet other conditions that might not meet the criteria because of a very low prevalence. However, these disorders can ride the coattails of MCAD deficiency into NBS panels because the incremental cost of looking for additional disorders is minimal (Green et al., 1999). The MS/MS instruments can either be set to identify specific molecules associated with a particular disorder, or set of disorders, or they can be set to scan a range of many different molecules thus identifying a variety of metabolic conditions. Even if the MS/MS instrument is used narrowly to identify specific conditions, peaks representing masses characteristic of other disorders will occasionally be found during the screening process (Sweetman, 2001). Using the full-scan capacity of MS/MS will identify babies who have abnormal metabolic profiles that may not be specific to any disorder. Additionally, it appears that MS/MS identifies a subpopulation of people who may have remained asymptomatic without treatment. For example, many more cases of MCAD deficiency have been detected by NBS than is predicted by the prevalence of clinically identified populations. And, cases detected through NBS appear to have different genotype frequencies than those identified through clinical presentation (Andresen et al., 2001). In addition to identifying asymptomatic individuals, MS/ MS screening can identify babies with conditions that do not have treatments with proven benefits. This reality could lead parents and physicians to what Fielding et al. call “therapeutic odysseys,” where the doctors use what limited knowledge and experience they have about these rare disorders to give their best guess for treatments (Fielding et al., 2003). However, proponents of expanded screening argue that early identification is preferable to the “diagnostic odysseys” that usually accompany the identification of a rare metabolic disorder that presents clinically. Also, they argue, novel treatments may be discovered that are beneficial. Early diagnoses can also help some couples with reproductive decision-making. NBS programs with MS/MS capacity have either chosen to use this technology narrowly to identify a subset of disorders, or broadly in the full-scan mode to identify all of the detectable conditions. In 2004, a group of experts from the American College of Medical Genetics (ACMG), commissioned by the United States Health Resources and Services Administration (HRSA), recommended universal screening for a core panel of 29 disorders (20 of which are identified by MS/MS). The report suggests that another 25 conditions, largely conditions that are not well understood or do not have effective treatments, fall into a class called “secondary targets” for screening (22 of which are identifiable by MS/MS). The experts recommended that these conditions be reported but gave no guidance about what primary care providers should do with these results (NBS, 2005). Deciding to add these disorders to NBS panels is a contemporary challenge facing NBS programs. The trend in the United States has been to follow these guidelines by expanding
476
CHAPTER 41
■
Genomics and Population Screening: Example of Newborn Screening
NBS panels to the “ACMG 29.” Companion commentaries in the medical journal Pediatrics entitled “Newborn Screening Technology: Proceed with Caution” and “We Need Expanded Newborn Screening” offer clear arguments on both sides of the current debate (Botkin et al., 2006; Howell, 2006).
VARIABILITY AMONG NBS PROGRAMS The preceding four examples have highlighted some of the array of challenges that NBS presents. Considering these complexities, it is not surprising that states and countries have differences in their programs, both in the disorders they screen for and in the services they provide (Serving the family from birth to the medical home, 2000;Townes, 1986). Different groups of knowledgeable and well-intentioned individuals have come to different decisions about what to screen for and how to operate NBS programs. Most NBS programs operate at a regional level. In the United States, for example, each state and Washington DC operates its own NBS program. Among them there are variations in the disorders included in their panels, screening technologies used in the laboratory, funding mechanisms supporting the program and services available beyond screening in the laboratory (i.e., follow-up, genetic counseling, systems of care, etc.). Most states utilize one NBS laboratory, usually operating within the state health department. Because it has been suggested that laboratories should have a minimum of 50,000 specimens per year to maintain quality of screening (Guthrie, 1972), some smaller states collaborate by sending blood specimens to a regional laboratory for testing. A few programs contract with private laboratories to perform the screening. Historically, NBS has been a relatively low-profile public health program that has enjoyed a behind-the-scenes role of helping babies. Advocates for child health have long been instrumental in NBS expansion. Recently, parent advocacy groups, as well as non-profit health organizations, and the federal government of the United States have raised public awareness about NBS and pushed for national standards such as the ACMG report (Cunningham, 2002; NBS, 2005; Serving the family from birth to the medical home, 2000). Advocacy organizations such as Save Babies Through Screening, the March of Dimes, and the American Academy of Pediatrics were quick to voice their support of the report. Although a federal advisory committee accepted the recommendations on September 23, 2004, as of November 2006, the Department of Health and Human Services has taken no formal action regarding the ACMG report. The United States federal funding has increased in recent years supporting the development of new technologies and improvement of NBS services, in part to address issues of variability among NBS programs. Other federally funded projects, such as the Human Genome Project, have increased the general public awareness of genetics and brought hopes of cures and therapies for those who suffer from genetic disorders, including the prospect of personalized medicine where therapies are prescribed based on an individual’s genotype.
INFLUENCE OF GENETICS AND -OMIC TECHNOLOGIES ON NBS NBS is often regarded as a genetic screening program because most of the conditions included in NBS panels are monogenic, meaning they are disorders caused by mutations in a single gene. However, the fact that these conditions are genetic is not the reason that they are included in NBS panels (Motulsky, 1997). Although the genotype of the baby is not usually determined by NBS, genomic expression is analyzed. The majority of NBS assays either detect specific biomolecules (proteins, hormones, amino acids, organic acids or acylcarnitines) or measure the activity of certain enzymes. In some cases, affected babies are genotyped as part of the diagnostic work-up or have DNA testing done after diagnosis to improve clinical care. While most of these disorders exhibit a simple inheritance pattern, the physical manifestations of these conditions are complex and highly variable. The genotype/phenotype correlation, or how well an individual’s DNA predicts disease expression, is frequently weak because many of the disorders are affected by environmental influences and/or interactions with other genes. This complexity is compounded in some disorders because of allelic heterogeneity, meaning that more than one disease causing mutation exists in a given gene, as seen in more than 1400 mutations in the CFTR gene that are known to cause cystic fibrosis (Tsui et al., 2005). A related phenomenon is locus heterogeneity, where the same disease phenotype is caused by mutations in more than one gene. For example, defects in any of five different genes can cause congenital adrenal hyperplasia (Donohoue et al., 2001). While most of the conditions in NBS panels are monogenic disorders, direct DNA testing is not routinely employed. Currently, no NBS programs in the United States use direct DNA testing as a primary screening method, although the concept has been proposed (Green et al., 2006). While genotyping is becoming increasingly less costly, primary DNA testing for NBS is problematic because of the large amount of heterogeneity in the relevant genes and the inevitable high rates of carrier detection. Direct DNA analysis is used in some instances, as a second-tier test for specimens with abnormal results on the first screen. The largest use of this type of DNA testing in NBS is using panels of multiple mutations as part of the screen for CF. Second-tier genotyping for certain hemoglobin variants and mutations that cause galactosemia have proven useful for guiding follow-up in Washington State (Neier et al., 2005). For example, a baby with a very low galactose-1-phosphate uridyl transferase (GALT) enzyme activity can be genotyped for mutations that cause both the severe and mild forms of galactosemia. This allows NBS follow-up staff to temper their response based on the genotype: if a mild allele is detected, there is no need for immediate follow-up and diagnostic work-up. Research efforts in NBS are directed at improving current methodologies, designing new technologies and developing tests for “new” disorders not currently included in NBS panels. For example, new methods using MS/MS identify a class of disorders
References
known as lysosomal storage disorders that may become candidates for NBS (Li et al., 2004). Another condition being researched for NBS is type 1 diabetes. International efforts are being made with The Environmental Determinants of Diabetes in the Young (TEDDY) study. Families are recruited to include their newborns in the study. The initial screen uses the residual blood from the mandated NBS program. This blood is tested for certain genotypes that identify those babies with a higher risk of developing type 1 diabetes. The families of these babies are asked to participate in a long-term study on the environmental determinants of this complex childhood disease (Vogt et al., 2003). NBS has benefited modestly from the knowledge of genetics and genotyping technology. However, much of the “low-hanging fruit” of the human genome has already been picked (Risch, 2000). As reviewed in this chapter, even seemingly simple NBS conditions are often difficult to understand and treat. What remains is to tease out the genetics of common and complex conditions such as diabetes, obesity, heart disease, cancer and hypertension (Khoury et al., 2003). Advances in the fields of genomics, proteomics (de Hoog et al., 2004), metabolomics (Rochfort, 2005) and their integration into systems biology research (Weston et al., 2004) herald much promise for the future of medical science, including NBS. This type of integrated approach will be critical to better understand common and complex conditions that are probably influenced by multiple genes and the environment. Alexander and van Dyck outline a vision of the future of NBS in a recent article, calling for equity among states in the screening panels, testing for many more disorders using DNA technology, and improving NBS education and infrastructure (Alexander et al., 2006). They also advocate for “changing the dogma” of NBS, meaning that conditions for which effective treatment does not exist should not be excluded from NBS panels. They argue that this practice is outdated because it does not consider the benefits to families in avoiding diagnostic odysseys and for reproductive decision-making, and the benefits for the
■
477
child of early intervention and participating in research such as clinical trials of new treatments. NBS programs have generated more than 40 years of experience as a population-based genetic screening program. Lessons learned by NBS may be of benefit to policymakers and entrepreneurs as they consider other types of genetic screening programs: 1. In most cases, DNA testing is not an easy answer. Many disorders are influenced by more than one gene. Clinical heterogeneity of disease complicates matters further because a specific mutation may exhibit a spectrum of clinical outcomes, from severe disease to asymptomatic. 2. Experts are needed to interpret results to maximize the benefits of the genetic information (empowered patients and physicians) and to minimize the negative effects (poor decision-making based on misinformation and parental stress). The future holds many exciting advances in science, technology and medicine. These will include techniques to improve the screening and treatment of conditions currently included in NBS programs. Many new disorders will become candidates for NBS. Laboratories will adapt to these changes and learn new methods, but barring some global change in the way public health is viewed, it is likely that future expansion of NBS programs will be guided toward those conditions that meet the traditional inclusion criteria that were listed at the beginning of this chapter. In addition to advances in technology, improvements in some of the more mundane aspects of the NBS system, such as collecting good quality blood spots and decreasing specimen transit time, will surely yield great benefits to babies with NBS conditions.
ACKNOWLEDGEMENTS The authors thank Sheila Weiss and JoLyn Thompson for their helpful comments on draft versions of this chapter.
REFERENCES Aldis, B., Hoffman, F. and Therrell B.L. (1993). Laboratory methods for phenylalanine analysis on newborn screening specimens. In Laboratory Methods for Neonatal Screening (B.L. Therrell, Jr., ed.), American Public Health Association,Washington, DC, pp. 47–75. Alexander, D. and van Dyck, P.C. (2006). A Vision of the Future of Newborn Screening. Pediatrics 117(5), S350–S354. Andresen, B.S.,Dobrowolski, S.F.,O’Reilly, L.,Muenzer, J.,McCandless, S.E., Frazier, D.M., Udvari, S., Bross, P., Knudsen, I., Banas, R. et al. (2001). Medium-chain acyl-Coa dehydrogenase (MCAD) mutations identified by MS/MS-based prospective screening of newborns differ from those observed in patients with clinical symptoms: Identification and characterization of a new, prevalent mutation that results in mild MCAD Deficiency. Am J Hum Genet 68(6), 1408–1418. Armstrong, M.D., Low, N.L. and Bosma, J.F. (1957). Studies on phenylketonuria. Ix. Further observations on the effect of phenylalaninerestricted diet on patients with phenylketonuria. Am J Clin Nutr 5(5), 543–554.
Blau, N.,Thony, B., Cotton, R.G.H. and Hyland, K. (2001). Disorders of tetrahydrobiopterin and related biogenic amines. In The Metabolic and Molecular Bases of Inherited Disease (C.R. Scriver, A.L. Beaudet, W.S. Sly and D.Valle, eds), McGraw-Hill, New York, pp. 1725–1776. Botkin, J.R., Clayton, E.W., Fost, N.C., Burke, W., Murray, T.H., Baily, M.A., Wilfond, B., Berg, A. and Ross, L.F. (2006). Newborn screening technology: Proceed with caution. Pediatrics 117(5), 1793–1799. CDC (2006). Newborn Screening Quality Assurance Program. http:// www.cdc.gov/labstandards/nsqap.htm. Chace, D.H., Kalas, T.A. and Naylor, E.W. (2003). Use of tandem mass spectrometry for multianalyte screening of dried blood specimens from newborns. Clin Chem 49(11), 1797–1817. Clayton, E.W. (2005). Talking with parents before newborn screening. J Pediatr 147(Suppl. 3), S26–S29. Comeau, A.M., Parad, R.B., Dorkin, H.L., Dovey, M., Gerstle, R., Haver, K., Lapey, A., O’Sullivan, B.P.,Waltz, D.A., Zwerdling, R.G.
478
CHAPTER 41
■
Genomics and Population Screening: Example of Newborn Screening
et al. (2004). Population-based newborn screening for genetic disorders when multiple mutation DNA testing is incorporated: A cystic fibrosis newborn screening model demonstrating increased sensitivity but more carrier detections. Pediatrics 113(6), 1573–1581. Consensus Conference (1987). Newborn Screening for Sickle Cell Disease and Other Hemoglobinopathies. JAMA 258(9), 1205–1209. Council of Regional Networks for Genetic Services (1997). Guidelines for Follow-up of Carriers of Hemoglobin Variants Detected by Newborn Screening. Crossley, J.R., Elliott, R.B. and Smith, P.A. (1979). Dried-blood spot screening for cystic fibrosis in the newborn. Lancet 1(8114), 472–474. Cunningham, G. (2002). The science and politics of screening newborns. N Engl J Med 346(14), 1084–1085. de Hoog, C.L. and Mann, M. (2004). Proteomics. Annu Rev Genomics Hum Genet 5, 267–293. Donohoue, P.A., Parker, K.L. and Migeon, C.J. (2001). Congenital adrenal hyperplasia. In The Metabolic and Molecular Bases of Inherited Disease (C.R. Scriver, A.L. Beaudet, W.S. Sly and D. Valle, eds), McGraw-Hill, New York, pp. 4077–4115. Fielding, R., Lam, W. and Leung, G. (2003). Population screening. N Engl J Med 348(16), 1605. Garrick, M.D., Dembure, P. and Guthrie, R. (1973). Sickle-cell anemia and other hemoglobinopathies. Procedures and strategy for screening employing spots of blood on filter paper as specimens. N Engl J Med 288(24), 1265–1268. Gaston, M.H.,Verter, J.I.,Woods, G., Pegelow, C., Kelleher, J., Presbury, G., Zarkowsky, H., Vichinsky, E., Iyer, R., Lobel, J.S. et al. (1986). Prophylaxis with oral penicillin in children with sickle cell anemia. A randomized trial. N Engl J Med 314(25), 1593–1599. Green, A. and Pollitt, R.J. (1999). Population newborn screening for inherited metabolic disease: Current UK perspectives. J Inherit Metab Dis 22(4), 572–579. Green, N.S., Dolan, S.M. and Murray, T.H. (2006). Newborn screening: Complexities in universal genetic testing. Am J Public Health 96(11), 1955–1959. Grosse, S.D. (2005). Does newborn screening save money? The difference between cost-effective and cost-saving interventions. J Pediatr 146(2), 168–170. Grosse, S.D., Boyle, C.A., Kenneson, A., Khoury, M.J. and Wilfond, B.S. (2006). From public health emergency to public health service: The implications of evolving criteria for newborn screening panels. Pediatrics 117(3), 923–929. Guthrie, R. (1961). Blood screening for phenylketonuria. JAMA 178(863), 167. Guthrie, R. (1972). Mass screening for genetic disease. Hosp Pract, 93–100. Guthrie, R. and Susi, A. (1963). A simple phenylalanine method for detecting phenylketonuria in large populations of newborn infants. Pediatrics 32, 338–343. Hammond, K.B., Abman, S.H., Sokol, R.J. and Accurso, F.J. (1991). Efficacy of statewide neonatal screening for cystic fibrosis by assay of trypsinogen concentrations. N Engl J Med 325(11), 769–774. Hampton, M.L., Anderson, J., Lavizzo, B.S. and Bergmen, A.B. (1974). Sickle cell “nondisease” a potentially serious public health problem. Am J Dis Child 128(1), 58–61. Hannon, W.H. and Therrell,, B.L., (1993). Laboratory methods for detecting congenital hypothyroidism. In Laboratory Methods for Neonatal Screening (B.L. Therrell, Jr., ed.), American Public Health Association,Washington, DC, pp. 139–154.
Heeley, A.F., Heeley, M.E., King, D.N., Kuzemko, J.A. and Walsh, M.P. (1982). Screening for cystic fibrosis by dried blood spot trypsin assay. Arch Dis Child 57(1), 18–21. Howell, R.R. (2006). We need expanded newborn screening. Pediatrics 117(5), 1800–1805. Kennedy, J.L. Jr. Wertelecki, W., Gates, L., Sperry, B.P. and Cass, V.M. (1967). The early treatment of phenylketonuria. Am J Dis Child 113(1), 16–21. Khoury, M.J., McCabe, L.L. and McCabe, E.R. (2003). Population screening in the age of genomic medicine. N Engl J Med 348(1), 50–58. Kleman, K.M., Vichinsky, E. and Lubin, B.H. (1989). Experience with newborn screening using isoelectric focusing. Pediatrics 83(5 Pt 2), 852–854. Lane, P.A. (1996). Sickle cell disease. Pediatr Clin North Am 43(3), 639–664. Li,Y., Scott, C.R., Chamoles, N.A., Ghavami, A., Pinto, B.M., Turecek, F. and Gelb, M.H. (2004). Direct multiplex assay of lysosomal enzymes in dried blood spots for newborn screening. Clin Chem 50(10), 1785–1796. Markel, H. (1997). Appendix 6: Scientific advances and social risks: Historical perspectives of genetic screening programs for sickle cell disease, Tay-Sachs disease, neural tube defects and Down Syndrome, 1970–1997. In Promoting Safe and Effective Genetic Testing in the United States: Final Report of the Task Force on Genetic Testing (N. Holtzman and M. Watson, eds), National Institutes of Health, Bethesda, MD. Millington, D.S. (2002). Newborn screening for metabolic diseases. Am Sci 90, 40–47. Millington, D.S., Kodo, N., Norwood, D.L. and Roe, C.R. (1990). Tandem mass spectrometry: A new method for acylcarnitine profiling with potential for neonatal screening for inborn errors of metabolism. J Inherit Metab Dis 13(3), 321–324. Motulsky, A.G. (1997). Screening for genetic diseases. N Engl J Med 336(18), 1314–1316. National Institutes of Health (2001). Consensus Development Conference Statement: Phenylketonuria: Screening and Management, October 16–18, 2000. Pediatrics 108(4), 972–982. Naylor, E.W. (1985). Screening for Pku cofactor variants. Genetic disease: Screening and management. In Proceedings of the 1985 Albany Birth Defects Symposium Albany, NY, Alan R. Liss, Inc.: pp. 211–230. NBS (2005). Newborn Screening: Toward a Uniform Screening Panel and System Report for Public Comment. http://mchb.hrsa. gov/screening/. Neier, S., Neidt, K. and Davis, T. (2005). Integrating DNA to guide newborn screening follow-up. In Proceedings of the 2005 Newborn Screening and Genetic Testing Symposium Association of Public Health Laboratories, Washington, DC. Pollitt, R.J. and Leonard, J.V. (1998). Prospective surveillance study of medium chain acyl-Coa dehydrogenase deficiency in the UK. Arch Dis Child 79(2), 116–119. Resler, G., Masse, R. and Petrin, K. (2005). Monitoring and reducing specimen batching in Washington State. In Proceedings of the 2005 Newborn Screening and Genetic Testing Symposium Association of Public Health Laboratories, Washington, DC. Risch, N.J. (2000). Searching for genetic determinants in the new millennium. Nature 405(6788), 847–856. Rochfort, S. (2005). Metabolomics reviewed: A new “omics” platform technology for systems biology and implications for natural products research. J Nat Prod 68(12), 1813–1820. Roe, C.R. and Ding, J. (2001). Mitochondrial fatty acid oxidation disorders. In The Metabolic and Molecular Bases of Inherited Disease
Recommended Resources
(C.R. Scriver, A.L. Beaudet, W.S. Sly and D. Valle, eds), McGrawHill, New York, pp. 2297–2326. Serving the Family from Birth to the Medical Home. (2000). A Report from the Newborn Screening Task Force Convened in Washington, DC, May 10–11, 1999. Pediatrics 106(2 Pt 2), 383–427. Sontag, M.K., Hammond, K.B., Zielenski, J.,Wagener, J.S. and Accurso, F.J. (2005). Two-tiered immunoreactive trypsinogen-based newborn screening for cystic fibrosis in Colorado: Screening efficacy and diagnostic outcomes. J Pediatr 147(Suppl. 3), S83–S88. Sweetman, L. (2001). Newborn screening by tandem mass spectrometry: Gaining experience. Clin Chem 47(11), 1937–1938. Tarini, B.A., Christakis, D.A. and Welch, H.G. (2006). State newborn screening in the tandem mass spectrometry era: More tests, more false-positive results. Pediatrics 118(2), 448–456. Thompson, J.D. and Glass, M. (2005). Deciding on a newborn screening strategy for cystic fibrosis. In Proceedings of the 2005 Newborn Screening and Genetic Testing Symposium Association of Public Health Laboratories, Washington, DC. Townes, P.L. (1986). Newborn screening: A potpourri of policies. Am J Public Health 76(10), 1191–1192. Tsui, L.C. and Sielenski, J. (2005). Cystic Fibrosis Mutation Database. http://www.genet.sickkids.on.ca/cftr/.
■
479
Vichinsky, E., Hurst, D., Earles, A., Kleman, K. and Lubin, B. (1988). Newborn screening for sickle cell disease: Effect on mortality. Pediatrics 81(6), 749–755. Vogt, R.F., Meredith, N., Henderson, L.O. and Hannon, W.H. (2003). Newborn screening and type 1 diabetes: Historical perspective and current activities and the CDC division of laboratory sciences. Diabetes Technol Ther 5(6), 1017–1021. Waisbren, S.E., Albers, S., Amato, S., Ampola, M., Brewster, T.G., Demmer, L., Eaton, R.B., Greenstein, R., Korson, M., Larson, C. et al. (2003). Effect of expanded newborn screening for biochemical genetic disorders on child outcomes and parental stress. JAMA 290(19), 2564–2572. Waisbren, S.E. and Azen, C. (2003). Cognitive and behavioral development in maternal phenylketonuria offspring. Pediatrics 112(6 Pt 2), 1544–1547. Weston, A.D. and Hood, L. (2004). Systems biology, proteomics, and the future of health care: Toward predictive, preventative, and personalized medicine. J Proteome Res 3(2), 179–196. Wilcken, B., Brown, A.R., Urwin, R. and Brown, D.A. (1983). Cystic fibrosis screening by dried blood spot trypsin assay: Results in 75,000 newborn infants. J Pediatr 102(3), 383–387.
RECOMMENDED RESOURCES Centers for Disease Control and Prevention (2006). Office of Genomics and Disease Prevention Home Page. http://www.cdc.gov/genomics/. This website “provides information about human genomic discoveries and how they can be used to improve health and prevent disease in populations. It also provides links to CDC wide activities in public health genomics across the lifespan.” Health Resources and Services Administration (2006). Maternal and Child Health Bureau (MCHB) Home Page. http://mchb.hrsa. gov/default.htm. “MCHB seeks a nation where there is equal access for all to quality health care in a supportive, culturally competent, family and community setting.” This website provides information about MCHB programs and grants related to newborn screening. Health Resources and Services Administration (2006). National Newborn Screening and Genetics Resource Center Home Page. http://genes-r-us.uthscsa.edu/.This excellent resource center “provides information and resources in the area of newborn screening and genetics to benefit health professionals, the public health community, consumers and government officials.” Their website includes current information about newborn screening panels and links to each of the program websites in the United States. Levy, H.L. and Albers, S. (2000). Genetic Screening of Newborns. Annu Rev Genomics Hum Genet 1, 139–177. This comprehensive article reviews the genetic conditions widely included in newborn screening programs and covers recent advances in tandem mass spectrometry and DNA testing.
(HuGE) Review focuses on the sickle hemoglobin and other globin gene variants and contains data regarding morbidity and mortality rates and a section that discusses unique interactions with other genes. Centers for Disease Control and Prevention (2001). Using Tandem Mass Spectrometry for Metabolic Disease Screening among Newborns: A Report of a Work Group. MMWR 50(No. RR-3). “In June 2000, the National Newborn Screening and Genetics Resource Center, in collaboration with CDC and the Health Resources and Services Administration, convened a workshop…. This work group report contains proposals for planning, operating, and evaluating tandem mass spectrometry technology in newborn screening and maternal and child health programs.” Centers for Disease Control and Prevention (2004). Newborn Screening for Cystic Fibrosis: Evaluation of Benefits and Risks and Recommendations for State Newborn Screening Programs. MMWR Recomm Rep 53(RR-13), 1–36. “In November 2003, CDC and the Cystic Fibrosis Foundation cosponsored a workshop to review the benefits and risks associated with newborn screening for cystic fibrosis. This report describes new research findings and outlines the recommendations of the workshop.” Centerwall, S.A. and Centerwall, W.R. (2000). The Discovery of Phenylketonuria: The Story of a Young Couple, Two Retarded Children, and a Scientist. Pediatrics 105(1 Pt 1), 89–103. This article is a narrative about Asbjörn Fölling, the Norwegian physician that discovered PKU, and the two children with PKU under his care. It includes a section about the early treatment of PKU.
Disorders/Technologies Covered in Chapter
Advocacy Organizations
Ashley-Koch, A., Yang, Q. and Olney, R.S. (2000). Sickle Hemoglobin (Hbs) Allele and Sickle Cell Disease: A Huge Review. Am J Epidemiol 151(9), 839–845. This Human Genome Epidemiology
Save Babies Through Screening (2006). Save Babies Through Screening Home Page. http://www.savebabies.org/. “Save Babies Through Screening Foundation is a national non-profit public charity
Newborn Screening: General
480
CHAPTER 41
■
Genomics and Population Screening: Example of Newborn Screening
run by volunteers. Its mission is to improve the lives of babies by working to prevent disabilities and early death resulting from disorders detectable through newborn screening.” Save Babies Through Screening is a well-organized and active advocacy group composed largely of family members of individuals affected by congenital disorders. Their website offers a wide variety of resources for parents and the lay public including personal stories from families of affected individuals. This organization has been a strong and effective force advocating for expanded screening. March of Dimes Foundation (2006). March of Dimes Foundation Home Page. http://www.marchofdimes.com/. “Our mission is to improve the health of babies by preventing birth defects, premature birth, and infant mortality. We carry out this mission through research, community services, education and advocacy to save babies’ lives.” The March of Dimes has been a vocal advocate for expanded newborn screening.
New Technologies Guttmacher, A.E. and Collins, F.S. (2005). Realizing the Promise of Genomics in Biomedical Research. JAMA 294(11), 1399–1402. This commentary considers the future of genomic research and its relations to health care and society. Jones, P.M. and Bennett, M.J. (2002). The Changing Face of Newborn Screening: Diagnosis of Inborn Errors of Metabolism by Tandem Mass Spectrometry. Clin Chim Acta 324(1–2), 121–128. This article reviews changes in newborn screening for inborn errors of metabolism and discusses tandem mass spectrometry technology. Wilcken, B. (2003). Ethical Issues in Newborn Screening and the Impact of New Technologies. Eur J Pediatr 162 Suppl. 1, S62–66. This article applies the principles of medical ethics (autonomy, beneficence, non-maleficence and justice) to new technologies available for screening newborns.
CHAPTER
42 Family History: A Bridge Between Genomic Medicine and Disease Prevention Maren T. Scheuner and Paula W. Yoon
INTRODUCTION Understanding the genetic basis for disease may substantially improve health care by providing options for disease management and prevention that are targeted to an individual’s genetic make-up. This personalized approach can improve health outcomes and efficiency through earlier and more accurate diagnosis, enhanced prevention efforts, and choice of therapies that optimize response and avoid adverse drug reactions. Providing such personalized information to patients also fits well with the concept of shared decision-making, a communication process between a clinician and patient that recognizes that certain recommendations for disease management and prevention need to be individualized according to a patient’s special circumstances and preferences. Thus, genetic and genomic information have the potential to revolutionize the way health care is delivered – ensuring access to more appropriate and more effective patientcentered care. Family history is an important tool for identifying individuals and families with genetic susceptibility to common chronic diseases such as coronary heart disease, stroke, diabetes and most cancers, as well as the rarer single gene disorders like cystic fibrosis, sickle cell anemia, hereditary forms of breast and colorectal cancer, and hemochromatosis. Generally for rare genetic Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg
disorders, family history primarily reflects the genetic risk factors shared by affected family members. Usually these are single gene disorders that follow specific patterns of inheritance and can be recognized on that basis. For the common, chronic diseases that are multifactorial in nature, family history of disease reflects the complex interactions of genetic and non-genetic risk factors (e.g., exposures, behaviors, cultural factors) shared by affected family members. In this chapter we will describe the role of family history in health promotion and disease prevention from both a clinical and public health perspective. Both perspectives emphasize an individualized rather than a one-size-fits-all approach to health care, which has the potential to result in improved health outcomes.
CLINICAL APPROACH Familial Risk Assessment For many common chronic diseases, a positive family history can increase the risk of disease from two to more than ten times those of the general population, and this risk generally increases with an increasing number of affected relatives and earlier ages of disease onset (i.e., occurring about 10–20 years earlier than Copyright © 2009, Elsevier Inc. All rights reserved. 481
482
CHAPTER 42
TABLE 42.1
■
Family History and Disease Prevention
Risks associated with family history for selected common diseases
Disease
Risk due to family history
References
Coronary heart disease
OR 2.5 (one first-degree relative diagnosed at or after age 60) OR 5.0 (one first-degree relative diagnosed before age 60) OR 5.1 (two or more first-degree relatives at any age of onset) OR 4.6 (one second-degree relative diagnosed before age 60)
Scheuner et al. (2006)
Type 2 diabetes
RR 2.4 (mother) RR 4.0 (maternal and paternal relatives)
Klein et al. (1996) Bjornholt et al. (2000)
Osteoporosis
OR 2.0 (female first-degree relative) RR 2.4 (father)
Keen et al. (1999) Fox et al. (1998)
Breast cancer
RR 2.1 (one first-degree relative) RR 3.9 (three or more first-degree relatives)
Pharoah et al. (1997) Collaborative Group (2001)
Colorectal cancer
OR 1.7 (one first-degree relative) OR 4.9 (two first-degree relatives)
Fuchs et al. (1994) Sandhu et al. (2001)
Prostate cancer
RR 3.2 (one first-degree relative) RR 11.0 (three first-degree relatives)
Cerhan et al. (1999) Steinberg et al. (1990)
OR odds ratio; RR relative risk
typical) (Table 42.1). Additional characteristics of high-risk family histories include multifocal or bilateral disease, higher rates of disease recurrence, occurrence of disease in the less often affected sex, and the occurrence of related diagnoses in a pattern suggestive of a single gene disorder. By recognizing the magnitude of risk associated with these familial characteristics, as well as patterns of familial disease suggestive of an inherited susceptibility, stratification of familial risk into different risk groups (e.g., weak, moderate, and strong) is possible (see Box 1) (Hampel et al., 2004; Scheuner et al., 1997). For certain common diseases, quantitative assessment of familial risk can also be performed for specific conditions using mathematical models or published estimates (Amos et al., 1992; Claus et al., 1993, 1994; St John et al., 1993). For example, using tables published by Claus (1993, 1994) an estimate of a woman’s risk for breast cancer in the next 10 years or by age 80 can be BOX 1
determined on the basis of her family history, including age at diagnosis of breast or ovarian cancer in first- and second-degree relatives, and these estimates can be contrasted to the population risk estimates. Familial risk assessment methods can provide relatively good estimates of disease risk when family history is the predominant risk factor, and they can be used as a “screening test” to identify individuals who should be referred to genetic susceptibility models that estimate the probability of a gene mutation associated with an inherited susceptibility (e.g., the BRCAPRO model by Parmigiani et al. (1998) that estimates the probability of a BRCA1 or BRCA2 gene mutation). However, familial risk assessment methods and genetic susceptibility models are not ideal for assessing overall risk for complex multifactorial disease, since they do not consider other risk factors that might influence disease risk, such as clinical history, exposures, or behaviors.
Familial Risk Stratification Strong familial risk is assigned:
Generally, for many common chronic diseases the risks are as follows: Weak familial risk can be assigned if there is: ● ●
No family history of disease Disease in only one second-degree relative from one or both sides of the family Moderate familial risk is assigned if there is:
● ● ● ●
Only one first-degree relative with late-onset disease Only one first-degree relative with late-onset disease and one second-degree related with late-onset disease from the same lineage Only one second-degree relative with early-onset disease and one second-degree relative with late-onset disease from the same lineage Only two second-degree relatives with late-onset disease from the same lineage
●
●
For all other family histories of disease, including having at least one first-degree relative with early-onset disease and combinations of multiple affected family members within and across generations When family histories are suggestive of known single gene disorders
Early age of disease onset refers to disease that occurs about 10– 20 years earlier than typical. Lineage refers to the side of the family where disease is reported. Maternal lineage refers to an affected mother and/or her relatives, and paternal lineage an affected father and/or his relatives. Some individuals may have only siblings or children affected; this can be described as a “nuclear” pattern of lineage. Adapted from Scheuner et al. (1997).
Clinical Approach
Unfortunately, the available absolute disease risk prediction models such as the Gail model for breast cancer (Rockhill et al., 2001) or the Framingham risk equation for coronary heart disease (Wilson et al., 1998) that provide disease risk estimates based primarily on socio-demographic and personal risk factors (e.g., medical conditions, exposures, and habits) consider only limited family history or do not consider family history at all in deriving their risk estimates. These models were developed with data collected from large epidemiologic studies that did not ascertain comprehensive family history. The models work well in predicting risk for most people, but they can underestimate disease risk for people who have a strong familial risk. Ideally, future epidemiologic studies should collect comprehensive socio-demographic, clinical, and family history data so that absolute risk prediction models can be developed that adequately address family history as a risk factor. Until that time, disease risk assessment strategies should include a combination of absolute risk prediction models, familial risk assessment methods, and genetic susceptibility models, when available. The figure provides an illustration of this strategy (Figure 42.1). Family History and Disease Management and Prevention Interventions Knowledge of the familial risk level can guide risk-specific recommendations for disease management and prevention, including referral for genetic evaluation or testing (Table 42.2). Recognizing personal and family history characteristics is crucial for identifying individuals suspected of having rare Mendelian disorders and for determining which patients should be offered testing (Scheuner et al., 2004). Furthermore, available guidelines for genetic testing referral rely heavily on specific family history criteria. Examples include the BRCA1/2 testing guidelines developed by the United States Preventive Services Task Force (2005) and the National Comprehensive Cancer Network (2006). Prevention strategies for people with increased familial risk could include targeted lifestyle changes; screening at earlier ages, more frequently and with more intensive methods than used for average risk individuals; use of chemoprevention; and for those at highest risk, prophylactic procedures, and surgeries. Data are accumulating regarding the effectiveness of these strategies in high-risk individuals (Scheuner et al., 2004). For example, the United States Preventive Services Task Force has found fair to good evidence supporting family history as an important clinical consideration for screening and prevention recommendations for breast cancer, colorectal cancer, lipid disorders, coronary heart disease, and abdominal aortic aneurysm (Wattendorf and Hadley, 2005). Breast cancer screening and prevention strategies appropriate for women with genetic susceptibility to breast cancer provide some of the best examples demonstrating the clinical utility of a family history based approach to prevention (Table 42.3). Family History and Emerging Genomics Technologies With the completion of the human genome sequence in 2003, we can expect rapid progress in the characterization of the genetic basis of rare Mendelian disorders, which number over 6000. But perhaps more importantly from a public health perspective, we will be able to define the genetic basis for the common
■
483
chronic diseases that result from complex interactions between variations in multiple genes and the environment. These genomic discoveries will dramatically increase the number of available genetic tests. Today there are about 1000 clinical genetic tests, and most are for rare Mendelian disorders (GeneTests, 2007). In the next 10 years we can expect the development of new genetic tests – including concurrent testing of multiple genetic markers using microarray technologies (i.e., multiplex testing) – that will be used to help refine diagnoses, improve risk prediction, and target therapies for common, chronic diseases. Will the family history become obsolete and eventually be replaced by modern genomic testing technologies? Many experts are of the opinion that the family history will remain highly relevant for many years to come. Recognizing personal and family history characteristics is crucial for identifying individuals suspected of having rare single gene disorders and for determining which patients should be offered testing (Scheuner et al., 2004), and there is no reason to believe that this strategy will change. For the complex multifactorial common chronic diseases, until we completely understand the role of specific genetic factors and their interaction with each other and the environment in determining health and disease, the family history may still be the best approach to select individuals for whom testing is most appropriate, and it may be most effective to integrate these genetic test results with selected family history and personal risk factors to suggest how best to individualize care (Guttmacher et al., 2004). The Need for Family History Tools Results of a recent national survey show that most Americans believe that knowing family history is very (73%) or somewhat (24%) important to their personal health (Centers for Disease Control and Prevention, 2004). Yet additional studies have shown that having a close family member with disease does not automatically translate to improvements in rates of selfinitiated participation in screening or risk-reducing lifestyles (Kip et al., 2002; West et al., 2003). How can we explain this apparent inconsistency? It may be that consumers do not appreciate how family history relates to their own risk or they may not be aware that despite a familial predisposition there are actions they can take to lower their risk. Further complicating this limitation on the use of family history, the literature suggests that physicians perform poorly with respect to collection and interpretation of family history for risk stratification and recommendation of risk-specific interventions (Acheson et al., 2000; Frezzo et al., 2003; Hayflick et al., 1998; Koscica et al., 2001; Sifri et al., 2002; Sweet et al., 2002). The reluctance of clinicians to examine the role of family history appears to be due to concerns about the amount of time required to collect the information, their ability to interpret such information accurately, and fears that they will not be compensated for the time spent researching family history (Gramling et al., 2004; Rich et al., 2004; Suther and Goodson, 2003). Typically, an absolute risk prediction model would be appropriate for individuals with a weak or moderate familial risk for a common chronic disease, and individuals with a strong familial risk would be referred to high-risk clinics for genetic
484
CHAPTER 42
■
Family History and Disease Prevention
Familial risk assessment Familial risk result
Weak
Moderate
Absolute risk model
Strong
Genetic susceptibility model Probability of mutation
Disease risk Low
High Genetic testing
Average
Increased Familial mutation excluded
Standard preventive interventions
Familial mutation not excluded
Deleterious mutation found
Enhanced preventive interventions
Figure 42.1 Integrating familial risk assessment methods, genetic susceptibility models, and absolute risk prediction models for optimal guidance in disease prevention
TABLE 42.2
Prevention strategies for colorectal cancer according to familial risk level
Familial risk level
Prevention strategy
References
Weak
Beginning at age 50, annual fecal occult blood testing (FOBT) and/or flexible sigmoidoscopy every 5 years, or doublecontrast barium enema every 5–10 years, or colonoscopy every 10 years.
Winawer et al. (1997); Rex et al. (2000); Smith et al. (2002); USPSTF (2002a)
Moderate
Beginning at age 40, annual FOBT and/or flexible sigmoidoscopy every 5 years, or double-contrast barium enema every 5–10 years, or colonoscopy every 10 years.
Winawer et al. (1997); Rex et al. (2000)
Strong
Beginning at age 40 or 10 years before the earliest age at diagnosis in the family, colonoscopy every 3–5 years. Consider daily use of aspirin. If family history is suggestive of hereditary colorectal cancer, consider screening and preventive interventions for associated cancers (e.g., endometrial and ovarian cancer).
Rex et al. (2000); Baron et al. (2003)
evaluation and application of genetic susceptibility models, when available. Individuals with a strong familial risk but a low probability of carrying a mutation (e.g., 10%) would be less likely to pursue testing (or may not be offered testing). However, enhanced preventive strategies may still be appropriate given their increased familial risk; this increased risk might be due to another inherited susceptibility not evaluated by the available genetic susceptibility models or to familial aggregation of a multifactorial form of disease. Individuals with a high probability of a mutation (e.g., 10%) would be most likely to undergo testing. If such individuals decline testing, clinicians should recommend enhanced preventive
interventions given their increased familial risk. Enhanced preventive interventions would also be recommended for individuals who undergo testing and are found to carry a deleterious mutation, as well as for individuals with indeterminate or normal results when there is no prior knowledge of a familial mutation (i.e., familial mutation not excluded). If a deleterious mutation has been previously identified in another family member and is excluded in an at-risk family member, it would be most appropriate to refer these individuals to absolute risk prediction models since their family history is generally no longer a significant risk factor and other socio-demographic and clinical risk factors would predominate.
Clinical Approach
TABLE 42.3
Efficacy
Chemo-prevention
●
●
● ●
Prophylactic surgery
●
●
●
●
● ●
●
●
Lifestyle
485
Screening for and prevention of breast cancer among women with a genetic susceptibility
Intervention
Screening for early detection
■
● ●
●
References
Tamoxifen reduces the occurrence of breast cancer by about 50% in women with an increased risk due in part to family history. Raloxifene is effective in reducing breast cancer incidence in postmenopausal women at risk, especially for women with a family history. Tamoxifen may reduce by 50–75% the risk of contralateral breast cancer in women with BRCA mutations. Tamoxifen may reduce the incidence of breast cancer in women with BRCA2 gene mutations, but not in women with BRCA1 gene mutations.
Fisher et al. (1998)
Prophylactic bilateral mastectomy reduced the incidence of breast cancer by 90% in women with both moderate and high-risk family histories of breast cancer. For women with a personal and family history of breast cancer, contralateral prophylactic mastectomy reduced the incidence of a second primary breast cancer by 94% and 96% for both premenopausal and postmenopausal women, respectively. Bilateral prophylactic mastectomy reduces the risk of breast cancer in women with BRCA mutations by about 95% in women with prior or concurrent bilateral prophylactic oophorectomy (BPO) and by about 90% in women with intact ovaries. Women with a BRCA gene mutation who undergo prophylactic BPO prior to menopause benefit not only from reducing their ovarian cancer risk by 95% but also from reducing their breast cancer risk by about 50%.
Hartmann et al. (1999)
Vogel et al. (2006); Lippman et al. (2006)
Narod et al. (2000) King et al. (2001)
McDonnell et al. (2001)
Meijers-Heijboer et al. (2001); Rebbeck et al. (2004)
Rebbeck et al. (2002, 2005); Kauff et al. (2002); Eisen et al. (2005)
The predictive value of mammography may be increased three-fold if there is a family history of breast cancer. Half of the tumors in BRCA mutation carriers appear in the interval between annual mammograms. Thus, shorter mammogram screening intervals and/or the addition of other imaging techniques might be warranted. In general, beginning at age 25–35, monthly self-examination, clinical examination every 6 months, and annual mammography are suggested for women with BRCA gene mutations. Good evidence shows that MRI has higher sensitivity for detecting breast cancer among women with BRCA mutations than does mammography, clinical breast examination, or ultrasound; use of combination screening with MRI, ultrasound, and mammography has the highest sensitivity of 95%.
Kerlikowske et al. (1993); Brekelmans et al. (2001) Meijers-Heijboer et al. (2001); Brekelmans et al. (2001); Smith et al. (2003); Scheuer et al. (2002); Komenaka et al. (2004); Warner et al. (2004) Burke et al. (1997); Meijers-Heijboer et al. (2001); Vasen et al. (1998); Smith et al. (2003)
Weight loss in early adult life has been associated with a reduced breast cancer risk in BRCA mutation carriers. Breastfeeding for more than 1 year may reduce the risk of breast cancer for women with BRCA1 mutations, but not for women with BRCA2 mutations. Women with BRCA1 mutations who take oral contraceptives have an increased risk for breast cancer; however, oral contraceptives do not appear to increase the risk in BRCA2 carriers.
Kotsopoulos et al. (2005)
Warner et al. (2004); Kriege et al. (2004)
Jernström et al. (2004)
Narod et al. (2002)
(continued)
486
CHAPTER 42
TABLE 42.3
■
Family History and Disease Prevention
(Continued)
Intervention
Efficacy
Lifestyle (continued)
●
●
●
References
Women with family history of breast cancer may increase their risk for breast cancer by using hormone replacement therapy (HRT); however, the rate of overall mortality may not be increased. The protective effect of BPO on breast cancer risk appears to be sustained in women with BRCA mutations who choose to use short-term postmenopausal HRT. Use of HRT in women with BRCA mutations does not increase the risk of ovarian cancer.
The obstacle posed by the lack of sufficient time or expertise to obtain, organize, and analyze the family history could be overcome in large part through use of family history tools (e.g., stand-alone tools or applications integrated into electronic health record and personal health record systems) that have the capabilities to: (1) collect relevant personal and family health history in a structured format, (2) organize the data in a usable form such as a graphic display following pedigree drawing standards, (3) interpret the familial risk and recognize patterns of familial disease suggestive of inherited susceptibilities (i.e., pedigree analysis), and (4) provide recommendations about interventions that are tailored to the familial risk and personal factors. Such tools have the potential to: (1) aid in identifying individuals with increased risk for common chronic diseases and rare genetic conditions, (2) facilitate discussion about health issues between patients and their health care professionals and among family members, (3) improve management of disease and disease risk factors by patients (i.e., self-management) and their providers, (4) raise awareness of the importance of family as a social unit integral to health and well-being through educational messaging, and (5) increase accuracy and completeness of family health history information, because patients could enter the data in a non-clinical setting at their own pace, affording them more time to access records and consult with family members than they have during an office visit. Accuracy of the family history data is paramount particularly if clinical decisions will depend on the information. Many studies have shown that self-reports of family health history are relatively accurate for many common chronic conditions, such as coronary heart disease, stroke, diabetes, and many forms of cancer. Most sensitivity values for self-reports of a positive family history of these conditions in a first-degree relative range from 70% to 90% and specificity is usually 90% or greater (Bensen et al., 1999; Kee et al., 1993; Hastrup et al., 1985; Murabito et al., 2004; Murff et al., 2004; Silberberg et al., 1998;Watt et al., 2000; Ziogas and Anton-Culver, 2003). Accuracy, however, depends on the type of disease, and it is less for more distant relatives and when historians are older. Therefore, before clinical decisions are based on such information, confirmation of family health histories is advisable.
Steinberg et al. (1991); Sellers et al. (1997)
Rebbeck et al. (1999, 2005)
Kotsopoulos et al. (2006)
Currently such confirmation is handled by health professionals who request and review medical records, pathology reports, and death certificates of family members to verify self-reports. This can be time-consuming and costly. One genetics clinic reported that 3–5.5 h were spent per initial consultation, with over half of this time devoted to activities such as obtaining and verifying family history data (Bernhardt and Pyeritz, 1989). Family History Tools and Electronic Health Records If family history tools were integrated into electronic medical record (EMR) systems, they could offer clinicians the advantages of standardizing the collection and organization of family history data. Electronic clinical decision support could also provide familial risk assessment and pedigree analysis, standardized guidance for referral for genetics consultation and testing, and evidence-based recommendations for interventions specific to the familial risk. Additionally, family history tools integrated within personal health record (PHR) systems used by consumers have the potential to improve self-management of disease and disease risk factors through built-in messaging about health promotion and disease prevention activities tailored to an individual’s familial risk and personal characteristics. Such PHR products could further improve the accuracy of family history reports by capabilities that allow health information (including results of genetic testing) to be transferred directly to family members rather than relying on clinicians to obtain and review records and transmit information. This direct communication among family members would improve familial risk assessment and pedigree analysis. The privacy concerns that are often raised by consumers regarding familial or genetic information could also be addressed if these PHRs offer methods for giving consumers control over (1) who may receive their information, (2) what specific types of information these recipients may get, and (3) how the recipients are limited in further sharing of this information. Family history applications developed for EMR and PHR products must have common data requirements and technical standards to ensure optimal exchange of family history data and ultimately use of family history information for disease management and health promotion activities, including the ordering
Clinical Approach
and interpretation of genetic tests. At present, several different standards development organizations are working to create a framework for representing and exchanging the contents of EHRs (including the family health history). The two most prominent of these are Health Level 7’s Clinical Document Architecture and the American Society of Testing and Materials International Continuity of Care Record. Two in this case is not better than one, as the differing standards and data architectures may prove to be substantial obstacles to efficient data exchange. Fortunately, it appears that both organizations will work collaboratively to define a common standard (Ferranti et al., 2006). Population Approach Although family history is known to be a risk factor for many chronic diseases, its use in preventive medicine and public health has been de-emphasized compared with modifiable risk factors such as smoking and diet. However, recent studies show that a large fraction of the population is likely to have a family history of one or more common diseases. For example, a populationbased study of family history of cardiovascular disease in Utah showed that 72% of early coronary heart disease (diagnosed before age 55 years) in the population occurred in 14% of families and 86% of early stroke occurred in 11% of families (Hunt et al., 2003). In a recent analysis of the National Health and Nutrition Examination Survey 1999–2002, 48% of the study population reported having at least one relative with diabetes (Hariri et al., 2006). And in Michigan, 7% of the population reported having an immediate family member diagnosed with colorectal cancer according to the 2005 Behavioral Risk Factor Survey (Personal communication, Deb Duquette, Michigan Department of Community Health, 2006). These data suggest that implementation of family history based strategies to assess disease risk, influence early disease detection, and encourage lifestyle changes could lead to overall population health benefits. An advantage of family history based approaches to prevention is that they do not focus exclusively on an individual’s risk factors but can work within a framework of biologic and cultural relationships to affect risk factor reduction. Family-Based Risk Assessment and Prevention Modifiable risk factors for common chronic diseases aggregate in families. For example, in an analysis of the Third National Health and Nutrition Examination Survey, adults with a parental history of coronary heart disease were more likely to have multiple risk factors such as smoking, elevated blood pressure, and elevated cholesterol. The odds ratio for four or five risk factors compared with none was 2.9, 95% CI, 1.4–6.3 (Brown et al., 2002). Furthermore, studies have shown that disease risk is substantially increased for individuals having lifestyle risk factors and a family history of disease compared to having only lifestyle risk factors or only family history, suggesting that individuals with increased familial risk have the most to gain from interventions related to traditional, modifiable risk factors (Tavani et al., 2004a, b).
■
487
Therefore, modifying risk factors within a family rather than focusing on an individual ought to be an effective strategy, and this has been shown in a few studies looking at modification of cardiovascular disease risk factors (Knutsen and Knutsen, 1991; Pyke et al., 1997). Lifestyle changes, such as dietary modification, weight control, and smoking cessation are likely to be more effective when delivered to the family than to an individual because family members can influence each other and provide ongoing support to one another. Familial Risk Stratification: A Public Health Screening Tool The added value of family history as a public health screening tool or motivational strategy should be rigorously tested as an adjunct to population-level prevention activities. Comprehensive population-based data are needed to develop family history risk assessment tools and evaluate the validity and utility of this approach. For each disease of interest, data are needed to (1) determine the accuracy and reliability of reporting disease status for each relative (analytic validity), (2) estimate the risk associated with family history and how well it can predict future disease (clinical validity), (3) determine the impact and usefulness of family history screening for disease prevention and early detection (clinical utility), and (4) monitor the ethical, legal, and social implications of assessing family history and conveying disease risk. As with any public health strategy to prevent diseases, there is a series of activities that are involved in developing an effective intervention: (1) assessing the health issue, (2) prioritizing the health issues and related interventions, (3) setting objectives for the intervention, (4) selecting effective interventions, (5) implementing activities, and (6) evaluating the interventions (Brownson et al, 2003). Information obtained from population-based surveys and follow up studies of family history are needed for each of these steps. More specifically, data are needed to do the following: ●
●
●
●
●
●
●
●
●
Assess population characteristics such as gender, race, age, and socio-economic indicators that may affect attitudes, knowledge, and practice regarding family history Establish the prevalence of family history for each disease of interest Assess the association between family history and disease occurrence Validate algorithms and stratification schemes developed for individual familial risk assessment Determine what risk factors modify the effect of family history on disease occurrence Monitor patient and provider practices regarding the collection and use of family history information Examine associations between knowledge of familial risk and preventive behaviors Measure the impact of family history based interventions on reduction of risk behaviors and disease prevention Estimate the cost-effectiveness of family history based screening for early detection and prevention of disease
488
CHAPTER 42
■
Family History and Disease Prevention
There have been many studies that have shown that family history is an independent and often strong risk factor for common chronic diseases (Table 42.1). More recently studies are revealing the potential of family history as a screening tool to identify subclinical disease, for example, coronary artery calcification (Michos et al., 2005; Nasir et al., 2004) and undiagnosed diabetes (Hariri et al., 2006). From a public health perspective, family history may be most useful in settings where individuals don’t have access to more invasive tests such as glucose and cholesterol testing. Family history and other risk factors, such as body mass index (based on weight and height), that can be self-reported could be the basis for a low-cost low-tech screening tool to identify individuals and families at risk for diabetes and cardiovascular diseases (Hariri et al., 2006) and for whom preclinical disease detection strategies might be most appropriate. The use of cancer family history to identify individuals at risk and guide prevention activities is well established in clinical practice but has not been used widely as a population-level screening tool. Family history is a potentially valuable tool for identifying individuals in the population who might benefit from earlier or more intensive cancer screening. In addition, educating people about a familial risk of cancer might motivate them to adhere to screening guidelines. For example, only about 45% of men and 40% of women aged 50 years and older have undergone colon cancer screening (American Cancer Society, 2006). Therefore, any strategies that can improve participation in colorectal cancer screening are welcome. However, there is a paucity of literature describing the impact of familial risk assessment on health-related behaviors (Audrain et al. 2003). As an alternative, decision analysis methods have been used to estimate the impact of family-based screening (Ramsey et al., 2005;Tyagi and Morris, 2003). While results from these simulation studies are promising, further data are needed to determine the effectiveness of this strategy for disease prevention, as well as the likely burden on the medical system. Family History Public Health Initiatives There has been a renewed interest in using family health history for individualized risk assessment and disease prevention, and efforts such as the Surgeon General’s Family History Initiative have begun to raise awareness among the public and health professionals about the value of family history (see
Recommended Resources). In 2004, the US Surgeon General declared Thanksgiving as National Family History Day and encouraged Americans to collect their family members’ health histories and share the information with their health care providers (Guttmacher et al., 2004). A website was created with educational materials as well as a family history data collection tool (http://www.hhs.gov/familyhistory). Several federal, state, and professional organizations are developing new family history tools for data collection and risk assessment (see the National Office of Public Health Genomics website, listed under Recommended Resources) and community-based studies are under way to assess awareness, understanding, and self-efficacy among population subgroups for the purpose of developing more appropriate and effective educational materials and health promotion messages.
CONCLUSION Family history is an important risk factor for both common chronic diseases and rare genetic disorders. Recognizing patterns of familial disease that signify increased risk can help to identify individuals who may have the most to gain from preventive interventions including genetic testing technologies. Tools that aid in the collection, organization, and interpretation of family health information, especially applications for interoperable electronic health record systems, will be crucial to the success of this approach. Such tools can educate users about the important role of the family in defining health risks and prevention options, thereby facilitating discussion about health promotion activities between consumers and their family members and health care providers. Finally, to ensure widespread adoption of family-based risk assessment and prevention, evidence must be generated from studies evaluating the feasibility, validity, and utility of this approach.
ACKNOWLEDGEMENTS No conflict of interest to report. Dr. Scheuner has a patent pending for a familial risk stratification and pedigree analysis method and apparatus (US Patent Application Number: 20060173717, Publication Date: 8/3/06).
REFERENCES Acheson, L.S., Wiesner, G.L., Zyzanski, S.J., Goodwin, M.A. and Stange, K.C. (2000). Family history-taking in community family practice: implications for genetic screening. Genet Med 2, 180–185. American Cancer Society. (2006). Cancer Prevention and Early Detection: Facts and Figures 2006. American Cancer Society, Atlanta. Amos, C.I., Shaw, G.L., Tucker, M.A. and Hartge, P. (1992). Age at onset for familial epithelial ovarian cancer. JAMA 268, 1896–1899.
Audrain-McGovern, J., Hughes, C. and Patterson, F. (2003). Effecting behavior change: Awareness of family history. Am J Prev Med 24, 183–189. Baron, J.A., Cole, B.F., Sandler, R.S., Haile, R.W., Ahnen, D., Bresalier, R., McKeown-Eyssen, G., Summers, R.W., Rothstein, R., Burke, C.A. et al. (2003). A randomized trial of aspirin to prevent colorectal adenomas. N Engl J Med 348, 891–899. Bensen, J.T., Liese, A.D., Rushing, J.T., Province, M., Folsom, A.R., Rich, S.S. and Higgins, M. (1999). Accuracy of proband reported
References
family history: The NHLBI Family Heart Study (FHS). Genet Epidemiol 17, 141–150. Bernhardt, B.A. and Pyeritz, R.E. (1989). The economics of clinical genetics services. III. Cognitive genetics services are not selfsupporting. Am J Hum Genet 44, 288–293. Bjornholt, J.V., Erikssen, G., Liestol, K., Jervell, J., Thaulow, E. and Erikssen, J. (2000). Type 2 diabetes and maternal family history: An impact beyond slow glucose removal rate and fasting hyperglycemia in low-risk individuals? Results from 22.5 years of follow-up of healthy nondiabetic men. Diabetes Care 23, 1255–1259. Brekelmans, C.T.M., Seynaeve, C., Bartels, C.C.M., TilanusLinthorst, M.M.A., Meijers-Heijboer, E.J., Crepin, C.M.G., van Geel, A.N., Menke, M., Verhoog, L.C., van den Ouweland, A. et al. (2001). Effectiveness of breast cancer surveillance in BRCA1/2 gene mutation carriers and women with high familial risk. J Clin Oncol 19, 924–930. Brown, D.W., Giles, W.H., Burke, W., Greenlund, K.J. and Croft, J.B. (2002). Familial aggregation of early-onset myocardial infarction. Community Genet 5, 232–238. Brownson, R.C., Baker, E.A., Leet, T.L. and Gillespie, K.N. (2003). Evidence-Based Public Health. Oxford University Press, New York. Burke, W., Daly, M., Garber, J., Botkin, J., Ellis Kahn, M.J., Lynch, P., McTiernan, A., Offit, K., Perlman, J., Petersen, G. et al. (1997). Recommendations for follow-up care of individuals with an inherited predisposition to cancer. II. BRCA1 and BRCA2. JAMA 277, 997–1003. Centers for Disease Control and Prevention (2004). Awareness of family health history as a risk factor for disease – United States. MMWR 53, 1044–1047. Cerhan, J.R., Parker, A.S., Putnam, S.D., Chiu, B.C., Lynch, C.F., Cohen, M.B., Torner, J.C. and Cantor, K.P. (1999). Family history and prostate cancer risk in a population-based cohort of Iowa men. Cancer Epidemiol. Biomarkers Prev 8, 53–60. Claus, E.B., Risch, N. and Thompson, W.D. (1993). The calculation of breast cancer risk for women with a first degree family history of ovarian cancer. Breast Cancer Res Treat 28, 115–120. Claus, E.B., Risch, N. and Thompson, W.D. (1994). Autosomal dominant inheritance of early-onset breast cancer. Implications for risk prediction. Cancer 73, 643–651. Collaborative Group on Hormonal Factors (2001). Familial breast cancer: collaborative reanalysis of individual data from 52 epidemiological studies including 58,209 women with breast cancer and 101,986 women without the disease. Lancet 358, 1389–1399. Eisen, A., Lubinski, J., Klijn, J., Moller, P., Lynch, H.T., Offit, K.,Weber, B., Rebbeck, T., Neuhausen, S.L., Ghadirian, P. et al. (2005). Breast cancer risk following bilateral oophorectomy in BRCA1 and BRCA2 mutation carriers: An international case-control study. J Clin Oncol 23, 7491–7496. Family Heart Study Group (1994). Randomised controlled trial evaluating cardiovascular screening and intervention in general practice: principal results of British family heart study. BMJ 308, 313–320. Ferranti, J.M., Musser, R.C., Kawamoto, K. and Hammond, E. (2006). The Clinical Document Architecture and the Continuity of Care Record: A critical analysis. J Am Med Inform Assoc 13, 245–252. Fisher, B., Costantino, J.P.,Wickerham, D.L., Redmond, C.K., Kavanah, M., Cronin, W.M.,Vogel,V., Robidoux, A., Dimitrov, N., Atkins, J. et al. (1998). Tamoxifen for prevention of breast cancer: Report of the
■
489
National Surgical Adjuvant Breast and Bowel Project P-1 Study. J Natl Cancer Inst 90, 1371–1388. Fox, K.M., Cummings, S.R., Powell-Threets, K. and Stone, K. (1998). Family history and risk of osteoporotic fracture: Study of Osteoporotic Fractures Research Group. Osteoporos Int 8, 557–562. Frezzo, T.M., Rubenstein, W.S., Dunham, D. and Ormond, K.E. (2003). The genetic family history as a risk assessment tool in internal medicine. Genet Med. 5, 84–91. Fuchs, C.S., Giovannucci, E.L., Colditz, G.A., Hunter, D.J., Speizer, F.E. and Willett, W.C. (1994). A prospective study of family history and the risk of colorectal cancer. N Engl J Med 331, 1669–1674. GeneTests. (2007). GeneTests Home Page. http://www.genetests.org Gramling, R., Nash, J., Siren, K., Eaton, C. and Culpepper, L. (2004). Family physician self-efficacy with screening for inherited cancer risk. Ann Fam Med 2, 130–132. Guttmacher, A.E., Collins, F.S. and Carmona, R.H. (2004). The family history – more important than ever. N Engl J Med 351, 2333–2336. Hampel, H., Sweet, K., Westman, J.A., Offit, K. and Eng, C. (2004). Referral for cancer genetics consultation: A review and compilation of risk assessment criteria. J Med Genet 41, 81–91. Hariri, S.,Yoon, P.W., Moonesinghe, R.,Valdez, R., Khoury, M.J. (2006). Evaluation of family history as a risk factor and screening tool for detecting undiagnosed diabetes in a nationally representative survey population. Genet. Med. 8, 752–759. Hariri, S., Yoon, P.W., Qureshi, N., Valdez, R., Scheuner, M.T. and Khoury, M.J. (2006). Family history of type 2 diabetes: A population-based screening tool for prevention. Genet Med 8, 102–108. Hartmann, L.C., Schaid, D.J., Woods, J.E., Crotty, T.P., Myers, J.L., Arnold, P.G., Petty, P.M., Sellers,T.A., Johnson, J.L., McDonnell, S.K. et al. (1999). Efficacy of bilateral prophylactic mastectomy in women with a family history of breast cancer. N Engl J Med 340, 77–84. Hastrup, J.L., Hotchkiss, A.P. and Johnson, C.A. (1985). Accuracy of knowledge of family history of cardiovascular disorders. Health Psychol 4, 291–306. Hayflick, S.J. and Eiff, M.P. (1998). Role of primary care providers in the delivery of genetic services. Community Genet 1, 18–22. Hunt, S.C., Gwinn, M. and Adams, T. (2003). Family history assessment: Strategies for prevention of cardiovascular disease. Am J Prev Med 24, 136–142. Jernström, H., Lubinski, J., Lynch, H.T., Ghadirian, P., Neuhausen, S., Isaacs, C., Weber, B.L., Horsman, D., Rosen, B., Foulkes, W.D. et al. (2004). Breast-feeding and the risk of breast cancer in BRCA1 and BRCA2 mutation carriers. J Natl Cancer Inst 96, 1094–1098. Kauff , N.D., Satagopan, J.M., Robson, M.E., Scheuer, L., Hensley, M., Hudis, C.A., Ellis, N.A., Boyd, J., Borgen, P.I., Barakat, R.R. et al. (2002). Risk-reducing salpingo-oophorectomy in women with a BRCA1 or BRCA2 mutation. N Engl J Med 346, 1609–1615. Kee, F., Tiret, L., Robo, J.Y., Nicaud, V., McCrum, E., Evans, A. and Cambien, F. (1993). Reliability of reported family history of myocardial infarction. BMJ 307, 1528–1530. Keen, R.W., Hart, D.J., Arden, N.K., Doyle, D.V. and Spector, T.D. (1999). Family history of appendicular fracture and risk of osteoporosis: A population-based study. Osteoporos Int 10, 161–166. Kerlikowske, K., Grad, D., Barclay, J., Sickles, E.A., Eaton,A. and Ernster,V. (1993). Positive predictive value of screening mammography
490
CHAPTER 42
■
Family History and Disease Prevention
by age and family history of breast cancer. JAMA 270, 2444–2450. King, M-C., Wieand, S., Hale, K., Lee, M., Walsh, T., Owens, K., Tait, J., Ford, L., Dunn, B.K., Costantino, J. et al. (2001). Tamoxifen and breast cancer incidence among women with inherited mutations in BRCA1 and BRCA2. JAMA 286, 2251–2256. Kip, K.E., McCreath, H.E., Roseman, J.M., Hulley, S.B. and Schreiner, P.J. (2002). Absence of risk factor change in young adults after family heart attack or stroke: The CARDIA Study. Am J Prev Med 22, 258–266. Klein, B.E., Klein, R., Moss, S.E. and Cruickshanks, K.J. (1996). Parental history of diabetes in a population-based study. Diabetes Care 19, 827–830. Knutsen, S.F. and Knutsen, R. (1991). The Tromso Survey: The Family Intervention Study – the effect of intervention on some coronary risk factors and dietary habits, a 6-year follow-up. Prev Med 20, 197–212. Komenaka, I.K., Ditkoff, B.A., Joseph, K.A., Russo, D., Gorroochurn, P., Ward, M., Horowitz, E., El-Tamer, M.B. and Schnabel, F.R. (2004). The development of interval breast malignancies in patients with BRCA mutations. Cancer 100, 2079–2083. Koscica, K.L., Canterino, J.C., Harrigan, J.T., Dalaya, T., Ananth, C.V. and Vintzileos, A.M. (2001). Assessing genetic risk: Comparison between the referring obstetrician and genetic counselor. Am J Obstet Gynecol 185, 1032–1034. Kotsopoulos, J., Olopade, O.I., Ghadirian, P., Lubinski, J., Lynch, H.T., Isaacs, C.,Weber, B., Kim-Sing, C., Ainsworth, P., Foulkes,W.D. et al. (2005). Changes in body weight and the risk of breast cancer in BRCA1 and BRCA2 mutation carriers. Breast Cancer Res 7, R833–844. Kotsopoulos, J., Lubinski, J., Neuhausen, S.L., Lynch, H.T., Rosen, B., Ainsworth, P., Moller, P., Ghadirian, P., Isaacs, C., Karlan, B. et al. (2006). Hormone replacement therapy and the risk of ovarian cancer in BRCA1 and BRCA2 mutation carriers. Gynecol Oncol 100, 83–88. Kriege, M., Brekelmans, C.T.M., Boetes, C., Besnard, P.E., Zonderland, H.M., Obdeijn, I.M., Manoliu, R.A., Kok, T., Peterse, H., Tilanus-Linthorst, M.M.A. et al. (2004). Efficacy of MRI and mammography for breast-cancer screening in women with a familial or genetic predisposition. N Engl J Med 351, 427–437. Lippman, M.E., Cummings, S.R., Disch, D.P., Mershon, J.L., Dowsett, S.A., Cauley, J.A. and Martino, S. (2006). Effect of raloxifene on the incidence of invasive breast cancer in postmenopausal women with osteoporosis categorized by breast cancer risk. Clin Cancer Res 12, 5242–5247. McDonnell, S.K., Schaid, D.J., Myers, J.L., Grant, C.S., Donohue, J.H., Woods, J.E., Frost, M.H., Johnson, J.L., Sitta, D.L., Slezak, J.M. et al. (2001). Efficacy of contralateral prophylactic mastectomy in women with a personal and family history of breast cancer. J Clin Oncol 19, 3938–3943. Meijers-Heijboer, H., Van Geel, B., Van Putten, W.L.J., HenzenLogmans, S.C., Seynaeve, C., Menke-Pluymers, M.B.E., Bartels, C.C.M., Verhoog, L.C., Van Den Ouweland, A.M.W., Niermeijer, M.F. et al. (2001). Breast cancer after prophylactic bilateral mastecomy in women with a BRCA1 or BRCA2 mutation. N Engl J Med 345, 159–164. Michos, E.D., Vasamreddy, C.R., Becker, D.M., Yanek, L.R., Moy, T.F., Fishman, E.K., Becker, L.C. and Blumenthal, R.S. (2005). Women with a low Framingham risk score and a family history of premature
coronary heart disease have a high prevalence of subclinical coronary artherosclerosis. Am Heart J 150, 1276–1281. Murabito, J.M., Nam, B.H., D’Agostino, R.B., Sr., Lloyd-Jones, D.M., O’Donnell, C.J. and Wilson, P.W. (2004). Accuracy of offspring reports of parental cardiovascular disease history: The Framingham Offspring Study. Ann Intern Med 140, 434–440. Murff, H.J., Spiegel, D.R. and Syngal, S. (2004). Does this patient have a family history of cancer? An evidence-based analysis of the accuracy of family cancer history. JAMA 292, 1480–1489. Narod, S.A., Brunet, J.S., Ghadirian, P., Robson, M., Heimdal, K., Neuhausen, S.L., Stoppa-Lyonnet, D., Lerman, C., Pasini, B., de los Rios, P. et al. (2000). Tamoxifen and risk of contralateral breast cancer in BRCA1 and BRCA2 mutation carriers: A case-control study. Hereditary Breast Cancer Clinical Study Group. Lancet 356, 1876–1881. Narod, S.A., Dube, M.P., Klijn, J., Lubinski, J., Lynch, H.T., Ghadirian, P., Provencher, D., Heimdal, K., Moller, P., Robson, M. et al. (2002). Oral contraceptives and the risk of breast cancer in BRCA1 and BRCA2 mutation carriers. J Natl Cancer Inst 94, 1773–1779. Nasir, K., Michos, E.D., Rumberger, J.A., Braunstein, J.B., Post, W.S., Budoff, M.J. and Blumenthal, R.S. (2004). Coronary artery calcification and family history of premature coronary heart disease: Sibling history is more strongly associated than parental history. Circulation 110, 2150–2156. National Comprehensive Cancer Network. (2006). Genetic/familial high-risk assessment: breast and ovarian. http://www.nccn.org/ professionals/physician_gls/PDF/genetics_screening.pdf Parmigiani, G., Berry, D.A. and Aguilar, O. (1998). Determining carrier probabilities for breast cancer susceptibility genes BRCA1 and BRCA2. Am J Hum Genet 62, 145–158. Pharoah, P.D., Day, N.E., Duffy, S., Easton, D.F. and Ponder, B.A. (1997). Family history and the risk of breast cancer: A systematic review and meta-analysis. Int J Cancer 71, 800–809. Pyke, S.D.M., Wood, D.A., Kinmonth, A.L. and Thompson, S.G. (1997). Change in coronary risk and coronary risk factor levels in couples following lifestyle intervention. Arch Fam Med 6, 354–360. Ramsey, S.D., Burke, W., Pinsky, L., Clarke, L., Newcomb, P. and Khoury, M.J. (2005). Family history assessment to detect increased risk for colorectal cancer: Conceptual considerations and a preliminary economic analysis. Cancer Epidemiol Biomarkers Prev 14(11 Pt 1), 2494–2500. Rebbeck, T.R., Levin, A.M., Eisen, A., Snyder, C., Watson, P., CannonAlbright, L., Isaacs, C., Olopade, O., Garber, J.E., Godwin, A.K. et al. (1999). Breast cancer risk after bilateral prophylactic oophorectomy in BRCA1 carriers. J Natl Cancer Inst 91, 1475–1479. Rebbeck, T.R., Lynch, H.T., Neuhausen, S.L., Narod, S.A., van’t Veer, L., Garber, J.E., Evans, G., Isaacs, C., Daly, M.B., Matloff, E. et al. (2002). Prophylactic oophorectomy in carriers of BRCA1 or BRCA2 mutations. N Engl J Med 346, 1616–1622. Rebbeck, T.R., Friebel, T., Lynch, H.T., Neuhausen, S.L., van ‘t Veer, L., Garber, J.E., Evans, G.R., Narod, S.A., Isaacs, C., Matloff , E. et al. (2004). Bilateral prophylactic mastectomy reduces breast cancer risk in BRCA1 and BRCA2 mutation carriers: The PROSE Study Group. J Clin Oncol 22, 981–983. Rebbeck, T.R., Friebel, T., Wagner, T., Lynch, H.T., Garber, J.E., Daly, M.B., Isaacs, C., Olopade, O.I., Neuhausen, S.L., van ‘t Veer, L. et al. (2005). Effect of short-term hormone replacement therapy on breast cancer risk reduction after bilateral prophylactic oophorectomy in BRCA1 and BRCA2 mutation carriers: The PROSE Study Group. J Clin Oncol 23, 7772–7774.
References
Rex, D.K., Johnson, D.A., Lieberman, D.A., Burt, D.W. and Sonnenberg, A. (2000). Colorectal cancer prevention 2000: Screening recommendations of the American College of Gastrenterology. Am J Gastroenterol 95, 868–877. Rich, E.C., Burke, W., Heaton, C.J., Haga, S., Pinsky, L., Short, M.P. and Acheson, L. (2004). Reconsidering the family history in primary care. J Gen Intern Med 19, 273–280. Rockhill, B., Spiegelman, D., Byrne, C., Hunter, D.J. and Colditz, G.A. (2001). Validation of the Gail model of breast cancer risk prediction and implications for chemoprevention. J Natl Cancer Inst 93, 358–366. Sandhu, M.S., Luben, R. and Khaw, K.T. (2001). Prevalence and family history of colorectal cancer: Implications for screening. J Med Screen 8, 69–72. Scheuer, L., Kauff , N., Robson, M., Kelly, B., Barakat, R., Satagopan, J., Ellis, N., Hensley, M., Boyd, J., Borgen, P. et al. (2002). Outcome of preventive surgery and screening for breast and ovarian cancer in BRCA mutation carriers. J Clin Oncol 20, 1260–1268. Scheuner, M.T., Wang, S.J., Raffel, L.J., Larabell, S.K. and Rotter, J.I. (1997). The family history: A comprehensive genetic risk assessment method for the chronic conditions of adulthood. Am J Med Genet 71, 315–324. Scheuner, M.T., Yoon, P.W. and Khoury, M.J. (2004). Contribution of Mendelian disorders to common chronic disease: Opportunities for recognition, intervention and prevention. Am J Med Genet 125C, 50–65. Scheuner, M.T., Whitworth, W.C., McGruder, H., Yoon, P.W. and Khoury, M.J. (2006). Expanding the definition of family history for early-onset coronary heart disease. Genet Med 8, 491–501. Sellers, T.A., Mink, P.J., Cerhan, J.R., Zheng, W., Andeson, K.E., Kushi, L.H. and Folsom, A.R. (1997). The role of hormone replacement therapy in the risk for breast cancer and the total mortality in women with a family history of breast cancer. Ann Intern Med 127, 973–980. Sifri, R.D.,Wender, R. and Paynter, N. (2002). Cancer risk assessment from family history: Gaps in primary care practice. J Fam Pract 51, 856. Silberberg, J.S., Wlodarczyk, J., Fryer, J., Ray, C.D. and Hensley, M.J. (1998). Correction for biases in a population-based study of family history and coronary heart disease. The Newcastle Family I. Am J Epidemiol 147, 1123–1132. Smith, R.A., Cokkinides, V., von Eschenbach, A.C., Levin, B., Cohen, C., Runowicz, C.D., Sener, S., Saslow, D. and Eyre, H.J. (2002). American Cancer Society guidelines for the early detection of cancer. CA Cancer J Clin 52, 8–22. Smith, R.A., Saslow, D., Sawyer, K.A., Burke, W., Costanza, M.E., Evans, W.P., III, Foster, R.S., Jr., Hendrick, E., Eyre, H.J., Sener, S. et al. (2003). Update 2003. CA Cancer J Clin 53, 141–169. St John, D.J., McDermott, F.T., Hopper, J.L., Debney, E.A., Johnson, W.R. and Hughes, E.S. (1993). Cancer risk in relatives of patients with common colorectal cancer. Ann Intern Med, 785–790. Steinberg, G.D., Carter, B.S., Beaty, T.H., Childs, B. and Walsh, P.C. (1990). Family history and the risk of prostate cancer. Prostate 17, 337–347. Steinberg, K.K., Thacker, S.B., Smith, S.J., Stroup, D.F., Zack, M.M., Flanders, D. and Berkelman, R.L. (1991). A meta-analysis of the effect of estrogen replacement therapy on the risk of breast cancer. JAMA 265, 1985–1990. Suther, S. and Goodson, P. (2003). Barriers to the provision of genetic services by primary care physicians: A systematic review of the literature. Genet Med 5, 70–76.
■
491
Sweet, K.M., Bradley, T.L. and Westman, J.A. (2002). Identification and referral of families at high risk for cancer susceptibility. J Clin Oncol 20, 528–537. Tavani, A., Augustin, L., Bosetti, C., Giordano, L., Gallus, S., Jenkins, D.J. and La Vecchia, C. (2004a). Influence of selected lifestyle factors on risk of acute myocardial infarction in subjects with familial predisposition for the disease. Prev Med 38, 468–472. Tavani, A., Bosetti, C., Dal Maso, L., Giordano, L., Franceschi, S. and La Vecchia, C. (2004b). Influence of selected hormonal and lifestyle factors on familial propensity to ovarian cancer. Gynecol Oncol 92, 922–926. Tyagi, A. and Morris, J. (2003). Using decision analytic methods to assess the utility of family history tools. Am J Prev Med 24(2), 199–207. United States Preventive Services Task Force (2002). Clinical guidelines: Screening for colorectal cancer: recommendation and rationale. Ann Intern Med 137, 129–131. United States Preventive Services Task Force. (2005). Genetic risk assessment and BRCA mutation testing for breast and ovarian cancer susceptibility. http://www.ahrq.gov/clinic/uspstf/uspsbrgen. htm Vasen, H.F., Haites, N.E., Evans, D.G., Stell, C.M., Moller, P., Hodgson, S., Eccles, D., Morrison, P., Stoppa Lyonet, D., Chang-Claude, J. et al. (1998). Current policies for surveillance and management in women at risk of breast and ovarian cancer: A survey among 16 European family cancer clinics. European Familial Breast Cancer Collaborative Group. Eur J Cancer 34, 1922–1926. Vogel, V.G., Costantino, J.P., Wickerham, D.L., Cronin, W.M., Cecchini, R.S., Atkins, J.N., Bevers, T.B., Fehrenbacher, L., Pajon, E.R., Jr., Wade, J.L., III. et al. (2006). Effects of tamoxifen vs raloxifene on the risk of developing invasive breast cancer and other disease outcomes: The NSABP Study of Tamoxifen and Raloxifene (STAR) P-2 trial. JAMA 295, 2727–2741. Warner, E., Plewes, D.B., Hill, K.A., Causer, P.A., Zubovits, J.T., Jong, R.A., Cutrara, M.R., DeBoer, G., Yaffe, M.J., Messner, S.J. et al. (2004). Surveillance of BRCA1 and BRCA2 mutation carriers with magnetic resonance imaging, ultrasound, mammography, and clinical breast examination. JAMA 292, 1317–1325. Watt, G., Mcconnachie, A., Upton, M., Emslie, C. and Hunt, K. (2000). How accurately do adult sons and daughters report and perceive parental deaths from coronary disease. J Epidemiol Community Health 54, 859–863. Wattendorf , D.J. and Hadley, D.W. (2005). Family history: The threegeneration pedigree. Am Fam Physician 72, 441–448. West, D.S., Greene, P.G., Kratt, P.P., Pulley, L., Weiss, H.L., Siegfried, N. and Gore, S.A. (2003). The impact of a family history of breast cancer on screening practices and attitudes in low-income, rural, African American women. J Womens Health (Larchmt) 12, 779–787. Wilson, P.W.F., D’Agostino, R.B., Levy, D., Belanger,A.M., Silbershatz, H. and Kannel, W.B. (1998). Prediction of coronary heart disease using risk factor categories. Circulation 97, 1837–1847. Winawer, S.J., Fletcher, R.H., Miller, L., Godlee, F., Stolar, M.H., Mulrow, C.D.,Woolf, S.H., Glick, S.N., Ganiats,T.G. and Bond, J.H. et al. (1997). Colorectal cancer screening; Clinical guidelines, evidence and rationale. Gastroenterology 112, 594–642. Ziogas, A. and Anton-Culver, H. (2003). Validation of family history data in cancer family registries. Am J Prev Med 24, 190–198.
492
CHAPTER 42
■
Family History and Disease Prevention
RECOMMENDED RESOURCES National Office of Public Health Genomics, Centers for Disease Control and Prevention. 2006 Family History: Resources and Tools. (http://www.cdc.gov/genomics/public/famhistMain.htm)
Department of Health and Human Services. 2006 US Surgeon General’s Family History Initiative (http://www.hhs.gov/ familyhistory/)
Section
Clinical Technologies Supporting Personalized Medicine
7
50. 51. 52.
Molecular Imaging as a Paradigm for Genomic and Personalized Medicine PET Imaging in Genomic Medicine MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges Fluorescence Imaging: Overview and Applications in Biomedical Research Imaging Genetics: Integration of Neuroimaging and Genetics in the Search for Predictive Markers Viral Chip Technology in Genomic Medicine Vaccines Against Infectious Diseases: A Biotechnology-Driven Evolution Cancer Vaccines: Some Basic Considerations Biosensors for the Genomic Age Stem Cells
53.
Gene Therapy
43. 44. 45. 46. 47. 48. 49.
CHAPTER
43 Molecular Imaging as a Paradigm for Genomic and Personalized Medicine Ralph Weissleder
INTRODUCTION Many modern disease treatments require precise positional information. For example in cancer, critical questions are: where is a tumor located, how large is it, is it confined or has it spread to lymph nodes, does it involve any critical anatomical structures that would alter the treatment strategy? These and other questions are being answered, at ever-increasing spatial resolution, through application of traditional anatomical imaging methods such as computed X-ray tomography (CT), magnetic resonance imaging (MRI), and ultrasound (US). While these methods still represent the mainstay of clinical imaging, it has become clear that acquisition of molecular and physiological information by nuclear, optical, and magnetic resonance imaging technologies could vastly enhance our ability to treat many diseases more efficiently (Weissleder and Pittet, 2006; Jaffer and Weissleder, 2005; Juweid and Cheson, 2006;Weissleder, 2002). Table 43.1 provides an overview of these different imaging approaches (some of which are described in greater detail in other chapters in this section) and their spatial resolutions. The recent revolution in genomic and proteomic technologies has provided a wealth of new molecular information that may transform the way in which diseases are clinically managed. Molecular imaging is poised to play a central role in this transformation because it will allow integration of molecular and physiological information specific to each patient with
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 494
anatomical information obtained by conventional imaging methods. Depending on the imaging agents used, molecular imaging can already be used experimentally to monitor and quantify in vivo gene/protein expression as well as specific biological processes that determine disease behavior and drug response. The hope is that the information obtained by clinical molecular imaging will one day be used to (i) detect molecular or physiological alterations that signal the presence of disease when it is still at a curable stage, (ii) evaluate and adjust treatment protocols and choices in real time, and (iii) speed up the drug development process. Using cancer as a specific example, this chapter highlights some of the potential roles of molecular imaging. Similar advances are being made in cardiovascular (Table 43.2), inflammatory, autoimmune, neurologic, and other diseases (Jaffer and Weissleder, 2005; Jaffer et al., 2007).
MOLECULAR IMAGING AND CANCER DETECTION Detection of stage 1 cancers carries a 90% 5-year survival rate for the majority of malignancies (Etzioni et al., 2003), while removal of precancerous lesions is curative. Genetic, proteomic, and epidemiological studies largely aim at identifying individuals at high risk but the correct spatial localization of early cancers in a given patient often remains a challenge. Conventional anatomic imaging techniques typically detect cancers when they are
Copyright © 2009, Elsevier Inc. All rights reserved.
TABLE 43.1
Overview of imaging systems
Technique
Resolution
Depth
Time
Imaging agents
Target
Cost
Primary small animal use
Clinical use
MR imaging
10–100 μm
No limit
min– hours
Gadolinium, dysprosium, magnetic particles
A, P, M
$$$
Versatile imaging modality with high soft tissue contrast
Yes
CT imaging
50 μm
No limit
min
Iodine
A, P
$$
Primarily for lung and bone imaging
Yes
Ultrasound imaging
50 μm
mm
min
Microbubbles
A, P
$$
Vascular and interventional imaging
Yes
PET imaging
1–2 mm
No limit
min
F-18, Cu-64, C-11, O-15
P, M
$$$
Versatile imaging modality with many different tracers
Yes
SPECT imaging
1–2 mm
No limit
min
Tc-99 m, In-111 chelates
P, M
$$
Commonly used to image labeled antibodies, peptides, etc.
Yes
Fluorescence reflectance imaging (FRI)
2–3 mm
1 cm
sec–min
(NIR) Photoproteins (GFP) fluorochromes
P, M
$
Rapid screening of molecular events in surface-based disease
Yes
Fluorescence mediated tomography (FMT)
1 mm
10 cm
sec–min
NIR fluorochromes
P, M
$$
Quantitative imaging of targeted or “smart” fluorochrome reporters in deep tumors
In development
Bioluminescence imaging
Several mm
cm
min
Luciferins
M
$$
Gene expression, cell and bacterial tracking
No
Intravital microscopy (confocal, multiphoton)
1 m
400– 800 m
sec–min
Photoproteins (GFP)
P, M
$$$
All of the above at higher resolutions but at limited depths and coverage
Limited development (skin, endoscopy)
Adapted from Weissleder, 2002. The resolution and cost columns refer to high-resolution, small animal imaging systems and are different for clinical imaging systems. Primary area that a given imaging modality interrogates: A: anatomic, P: physiologic, M: molecular targets. Cost of system: $: 100K; $$: 100–300K; $$$: 300K.
496
CHAPTER 43
TABLE 43.2 as of 2007
■
Molecular Imaging as a Paradigm for Genomic and Personalized Medicine
Representative examples of clinically promising molecular imaging approaches in cardiovascular disease
Agent
Target
Platform
Clinically tested?
Magnetic nanoparticles (iron oxide)
Cellular Inflammation (Macrophage smooth muscle and endothelial cells)
MRI
Yes (40 patients)
18
Atherosclerosis
Glucose tranporter-1, hexokinase
PET
FDA approved
99m
F-fluorodeoxyglucose
Apoptosis/macrophages/intraplaque hemorrhage
SPECT
Yes (24 patients)
99m
Lymphocytes
SPECT
Yes (25 patients)
Prosense
Cysteine protease activity
NIRF
Planned
Paramagnetic nanoparticles
Angiogenesis (integrin vβ3)
MRI
Planned
99m
Platelet GP IIb/IIIa receptor
SPECT
FDA approved
EP-2104R
Fibrin
MRI
Yes (76 patients)
99m
Angiogenesis (integrin vβ3)
SPECT
Yes (10 patients)
Magnetic nanoparticles, 111 Indium-oxine
Stem Cells
MRI, SPECT
Yes (in cancer, 11 patients)
Tc-annexin Tc-interleukin-2
Thrombosis Tc-apcitide
Myocardial infarction Tc-NC100692
From Jaffer and Weissleder, 2005.
centimeter-sized, that is, already consist of 109 cells (including circulating and microscopic metastatic deposits). Molecular imaging is expected to play an important role in this setting by imaging key molecular targets and host responses associated with early cancer development. For example, a number of key oncogenes are activated in lung cancer (e.g., K-ras, EGFR, HER2/neu, 2, Bcl-2), but there are also a myriad of other targets, more ideally suited for imaging (Sweet-Cordero et al., 2005). Some of these targets (e.g., proteases from the cathepsin family) have recently been used for detecting early lung cancer (Grimm et al., 2005). Importantly, the imaging agents were fluorescencebased and can also be used for detection of microscopic foci of epithelial (pre)cancers using endoscopic microscopy (Alencar et al., 2005; Evans and Nishioka, 2005; Marten et al., 2002). This technological advance is important because it allows detection of microscopic lesions elusive to conventional imaging and could one day enable in vivo characterization of tumors without the need for multiple excisional biopsies. Several other imaging technologies play important roles in staging and re-staging. For example, the use of magnetic nanoparticles targeted to macrophages in lymph nodes has been used as a sensitive means to detect nodal metastases in clinically occult disease (Harisinghani and Weissleder, 2004). Because of the exquisite spatial resolution of MRI, mm-sized metastases are detectable in non-enlarged lymph nodes (Harisinghani et al., 2003) beyond the detection threshold of many other imaging techniques. This approach has already been validated for a number of genitourinary malignancies, head and neck, and breast cancer.
Positron Emission Tomography (PET) imaging has emerged as a clinical cornerstone in cancer staging and re-staging for a number of malignancies and remains one of the few FDA approved technologies to date (Juweid and Cheson, 2006). The most frequently used PET agent (90% of all cancer-related scans) is 18F-labeled 2-deoxy-d-glucose (FDG), a glucose analog that is taken up by cells in proportion to their rate of glucose metabolism (Quon and Gambhir, 2005). Glucose metabolism is markedly increased in malignant cells, and FDG-PET imaging has been approved for staging of breast, colorectal, esophageal, head and neck, non-small-cell lung cancers, melanoma, and lymphoma (Guller et al., 2002; Juweid and Cheson, 2006; Quon and Gambhir, 2005).
MOLECULAR IMAGING TO DETERMINE TREATMENT EFFICACY FDG-PET imaging has also emerged as a clinical tool for measuring efficacy of novel anticancer therapies and predicting outcome (responders versus non-responders). The technique is particularly established for lymphoma, gastrointestinal stromal tumors, esophageal, head and neck, ovarian cancer, and approved for breast cancer (Wieder et al., 2004). For example, in one recent study, sequential FDG-PET imaging predicted patient outcome as early as after the first cycle of neoadjuvant chemotherapy and was more accurate than clinical or histopathologic response criteria including changes in tumor marker CA-125
Near-Term Needs and Opportunities
(Avril et al., 2005). Furthermore, FDG-PET imaging is increasingly being used to image other inflammatory diseases and the efficacy of molecularly targeted drugs (Goerres et al., 2005; Radu et al., 2007; Tahara et al., 2006; Tawakol et al., 2006). While FDG-PET imaging has been quite successful clinically, there has been a continued search for other agents that more specifically monitor tumor growth and cell death to follow treatment response more accurately. A wealth of new targets has already emerged from expression profiling and proteomic studies and is expected to give rise to new imaging agents. For example, radiolabeled nucleoside analogs such as thymidine compounds which theoretically are incorporated into DNA and could serve as a marker of cell proliferation have been developed. Clinical trials attesting to the utility of such agents are currently ongoing. Monoclonal antibodies developed against specific antigen targets have also been developed (e.g., CEA, PSMA, her2/neu) but are often problematic because of their relatively slow clearance from blood often leading to high background signals, even up to 1 week after injection of the antibody (Quon and Gambhir, 2005). Efforts are under way to construct engineered antibody fragments more suited for imaging, such as minibodies and diabodies, which show much more rapid blood clearance albeit at the expense of affinity. Other targeted radiolabeled agents include engineered proteins such as annexin-V, nanoparticles targeted to αvβ3 or VCAM-1, peptides or small molecules including dihydrotestosterone, estrogen, kinase inhibitors or active site binders.
MOLECULAR IMAGING AND DRUG DEVELOPMENT The development of new cancer therapeutics is expensive, timeconsuming, and often requires vast numbers of patients – factors that all contribute to the final cost of the therapies once they are approved for clinical use (Kelloff and Sigman, 2005). On average, it takes 10–12 years to take a new drug from discovery to regulatory approval at costs that can exceed $880 million (BIAG, 2006). In addition, many newer drugs (cytotoxic, cyctostatic, and molecular targeted) are often efficacious only in subgroups of patients (Lynch et al., 2004), while others – despite robust scientific rationale and promising preclinical results – have failed to show efficacy in clinical trials (Park et al., 2004). Reducing the number and cost of failed projects (75% attrition rate) would benefit the pharmaceutical industry, the healthcare system, and most importantly the patients. Molecular imaging is widely viewed as one of the most promising tools to improve the efficiency and cost-effectiveness of drug development programs. Imaging-based biomarkers (specific molecular targets or biological cancer processes e.g., HER2/neu, apoptosis) have many potential uses in all phases of the cancer drug development process, from target discovery and validation to the pivotal clinical trials that precede drug approval (Rudin and Weissleder, 2003). Similarly, such biomarkers are being defined for cardiovascular (Jaffer and Weissleder, 2005), rheumatoid arthritis (Izmailova et al., 2007), and other disease processes. Genetic reporter strategies
■
497
involving bioluminescence and fluorescent-tagged proteins have been extremely valuable for the study of cancer biology and preclinical drug evaluation in mouse models (Contag, 2002; Gross and Piwnica-Worms, 2005; Herschman, 2003; Massoud and Gambhir, 2003;Weissleder and Ntziachristos, 2003; Ntziachristos, 2000). In recent examples, these types of imaging approaches have been used to test the anti-tumor efficacy of synthetic epothilone analogs (Wu et al., 2005) or novel TRAIL constructs (Shah et al., 2005). While the reporter gene strategy is not directly translatable clinically, high-resolution mouse imaging with injectable imaging agents can now provide a windows into the effect of drugs on specific targets (Tang et al., 2005; Weissleder and Pittet, 2008; Herschman, 2003; Rudin and Weissleder, 2003;Weissleder, 2002). Molecular imaging can help identify new efficacy endpoints that are more easily monitored and less labor-intensive than currently used endpoints such as histological analyses of tumor biopsies. For example, steady-state imaging of vascular tumor volumes can be obtained in minutes in live mice whereas CD31 microvascular density measurements are much more labor intensive. Furthermore, because molecular imaging is non-invasive, in the preclinical setting it allows for longitudinal studies in a single animal, which can reduce the number of animals required for an experiment without compromising statistical significance. In the clinical setting, molecular imaging endpoints could be used to identify the most appropriate patient populations in which to test new drugs. Imaging of cancer drugs that have been labeled with 11C 18 or F can facilitate clinical pharmacokinetic and pharmacodynamic assessments, dosing and comparative efficacy studies of different lead compounds. In particular, microdosing studies (defined as 1% of the therapeutic dose, which typically has negligible toxicities in patients) have been advocated as a way to quickly obtain data on drug absorption, distribution, metabolism, excretion, and toxicity (Lappin and Garner, 2003; Propper et al., 2003; Saleem et al., 2001).
NEAR-TERM NEEDS AND OPPORTUNITIES What’s needed to catalyze the field of molecular imaging and drive new imaging agents into the clinic at a faster pace? Of utmost importance is the need to discover and validate new biomarkers suited for imaging, particularly those with amplification potential such as internalizing cell surface receptors, enzymes, and abundant non-protein targets (e.g., growth factor receptors). Tighter integration of imaging and therapeutic agent developments and systems level analysis of pathways and networks are expected to significantly speed up the discovery processes. Recent technological developments that may further hasten the biomarker discovery process are “imaging filters” to screen existing databases for targets ideally suited for imaging, newer conjugation chemistries such as “click chemistry” (Kolb and Sharpless, 2003) and use of yeast/phage surface display to identify recombinant antibodies/peptides (Joyce et al.,
498
CHAPTER 43
■
Molecular Imaging as a Paradigm for Genomic and Personalized Medicine
2003). There is also a pressing need for synthesis of new imaging agents. Here, there is an opportunity to apply combinatorial methods, chemical biology, synthetic small molecule compounds, and newer nanomaterials as scaffolds to improve pharmacological behavior (Anderson et al., 2004;Weissleder et al., 2005). The continued development of “smart” imaging reagents whose signal depends on biochemical activities remains a top priority to improve target-to-background ratios in vivo. For example, small molecule prodrugs that change their imaging signal upon target interaction can dramatically boost target-to-background ratios (Meade et al., 2003; Querol et al., 2005). Additional opportunities exist for the development fluorescence-based imaging agents, one of the most important growth areas in molecular imaging. Fluorochromes such as indocyanines are inexpensive, stable, use no radiation, and have been used safely for the last 20 years. Just as in the in vitro setting where they have been key
tools in both genomics and proteomics, fluorochromes can be uniquely converted into sensing agents in vivo (Weissleder et al., 1999). Continuing improvements in instrumentation including high spatial resolution endomicroscopy, near infrared intraoperative reflectance imaging, and fluorescence tomography will also be important. The latter technology is of particular interest because it allows accurate in vivo quantitation of near infrared fluorochrome, important to differentiate target binding from pharmacokinetics (Ntziachristos et al., 2005).
ACKNOWLEDGEMENT Portions of this chapter are based on a previously published article (Weissleder R. (2006). Molecular Imaging in Cancer, Science 312, 1168–1171) and are used here with permission.
REFERENCES Alencar, H., Mahmood, U., Kawano, Y., Hirata, T. and Weissleder, R. (2005). Novel multiwavelength microscopic scanner for mouse imaging. Neoplasia 7, 977–983. Anderson, D.G., Levenberg, S. and Langer, R. (2004). Nanoliter-scale synthesis of arrayed biomaterials and application to human embryonic stem cells. Nat Biotechnol 22, 863–866. Avril, N., Sassen, S., Schmalfeldt, B., Naehrig, J., Rutke, S., Weber, W.A., Werner, M., Graeff, H., Schwaiger, M. and Kuhn, W. (2005). Prediction of response to neoadjuvant chemotherapy by sequential F-18-fluorodeoxyglucose positron emission tomography in patients with advanced-stage ovarian cancer. J Clin Oncol 23, 7445–7453. Contag, P.R. (2002). Whole-animal cellular and molecular imaging to accelerate drug development. Drug Discov Today 7, 555–562. Etzioni, R., Urban, N., Ramsey, S., McIntosh, M., Schwartz, S., Reid, B., Radich, J., Anderson, G. and Hartwell, L. (2003). The case for early detection. Nat Rev Cancer 3, 243–252. Evans, J.A. and Nishioka, N.S. (2005). Endoscopic confocal microscopy. Curr Opin Gastroenterol 21, 578–584. Goerres, G.W. et al. (2006). F-18 FDG whole-body PET for the assessment of disease activity in patients with rheumatoid arthritis. Clin Nucl Med 31, 386–390. Grimm, J., Kirsch, D.G., Windsor, S.D., Kim, C.F., Santiago, P.M., Ntziachristos, V., Jacks, T. and Weissleder, R. (2005). Use of gene expression profiling to direct in vivo molecular imaging of lung cancer. Proc Natl Acad Sci USA 102, 14404–14409. Gross, S. and Piwnica-Worms, D. (2005). Spying on cancer: molecular imaging in vivo with genetically encoded reporters. Cancer Cell 7, 5–15. Guller, U., Nitzsche, E.U., Schirp, U.,Viehl, C.T., Torhorst, J., Moch, H., Langer, I., Marti, W.R., Oertli, D., Harder, F. et al. (2002). Selective axillary surgery in breast cancer patients based on positron emission tomography with 18F-fluoro-2-deoxy-D-glucose: not yet!. Breast Cancer Res Treat 71, 171–173. Harisinghani, M.G. and Weissleder, R. (2004). Sensitive, noninvasive detection of lymph node metastases. PLoS Med 1, e66.
Harisinghani, M.G., Barentsz, J., Hahn, P.F., Deserno,W.M.,Tabatabaei, S., Van Dekaa, C.H., De La Rosette, J. and Weissleder, R. (2003). Noninvasive detection of clinically occult lymph-node metastases in prostate cancer. N Engl J Med 348, 2491–2499. Herschman, H.R. (2003). Molecular imaging: Looking at problems, seeing solutions. Science 302, 605–608. Izmailova, E.S. et al. (2007). Use of molecular imaging to quantify response to IKK-2 inhibitor treatment in murine arthritis. Arthritis Rheum 56, 117–128. Jaffer, F.A. and Weissleder, R. (2005). Molecular imaging in the clinical arena. JAMA 293, 855–862. Jaffer, F.A., Libby, P. and Weissleder, R. (2007). Molecular imaging of cardiovascular disease. Circulation 116(9), 1052–1061. Joyce, J.A., Laakkonen, P., Bernasconi, M., Bergers, G., Ruoslahti, E. and Hanahan, D. (2003). Stage-specific vascular markers revealed by phage display in a mouse model of pancreatic islet tumorigenesis. Cancer Cell 4, 393–403. Juweid, M.E. and Cheson, B.D. (2006). Positron-emission tomography and assessment of cancer therapy. N Engl J Med 354, 496–507. Kelloff , G.J. and Sigman, C.C. (2005). New science-based endpoints to accelerate oncology drug development. Eur J Cancer 41, 491–501. Kolb, H.C. and Sharpless, K.B. (2003). The growing impact of click chemistry on drug discovery. Drug Discov Today 8, 1128–1137. Lappin, G. and Garner, R.C. (2003). Big physics, small doses: The use of AMS and PET in human microdosing of development drugs. Nat Rev Drug Discov 2, 233–240. Lynch,T.J., Bell, D.W., Sordella, R., Gurubhagavatula, S., Okimoto, R.A., Brannigan, B.W., Harris, P.L., Haserlat, S.M., Supko, J.G., Haluska, F.G. et al. (2004). Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-smallcell lung cancer to gefitinib. N Engl J Med 350, 2129–2139. Marten, K., Bremer, C., Khazaie, K., Sameni, M., Sloane, B., Tung, C.H. and Weissleder, R. (2002). Detection of dysplastic intestinal adenomas using enzyme-sensing molecular beacons in mice. Gastroenterology 122, 406–414.
References
Massoud, T.F. and Gambhir, S.S. (2003). Molecular imaging in living subjects: Seeing fundamental biological processes in a new light. Genes Dev 17, 545–580. Meade, T.J., Taylor, A.K. and Bull, S.R. (2003). New magnetic resonance contrast agents as biochemical reporters. Curr Opin Neurobiol 13, 597–602. Nadler, E., Eckert, B. and Neumann, P.J. (2006). Do oncologists believe new cancer drugs offer good value?. Oncologist 11, 90–95. Ntziachristos, V., Yodh, A.G., Schnall, M. and Chance, B. (2000). Concurrent MRI and diffuse optical tomography of breast after indocyanine green enhancement. Proc Natl Acad Sci USA 97, 2767–2772. Ntziachristos, V., Ripoll, J., Wang, L.V. and Weissleder, R. (2005). Looking and listening to light: The evolution of whole-body photonic imaging. Nat Biotechnol 23, 313–320. Park, J.W., Kerbel, R.S., Kelloff, G.J., Barrett, J.C., Chabner, B.A., Parkinson, D.R., Peck, J., Ruddon, R.W., Sigman, C.C. and Slamon, D.J. (2004). Rationale for biomarkers and surrogate end points in mechanism-driven oncology drug development. Clin Cancer Res 10, 3885–3896. Propper, D.J., De Bono, J., Saleem, A., Ellard, S., Flanagan, E., Paul, J., Ganesan, T.S., Talbot, D.C., Aboagye, E.O., Price, P. et al. (2003). Use of positron emission tomography in pharmacokinetic studies to investigate therapeutic advantage in a phase I study of 120-hour intravenous infusion XR5000. J Clin Oncol 21, 203–210. Querol, M., Chen, J.W., Weissleder, R. and Bogdanov, A.J. (2005). DTPA-bisamide-based MR sensor agents for peroxidase imaging. Org Lett 7, 1719–1722. Quon, A. and Gambhir, S.S. (2005). FDG-PET and beyond: molecular breast cancer imaging. J Clin Oncol 23, 1664–1673. Radu, C.G., Shu, C.J., Shelly, S.M., Phelps, M.E. and Witte, O.N. (2007). Positron emission tomography with computed tomography imaging of neuroinflammation in experimental autoimmune encephalomyelitis. Proc Natl Acad Sci USA 104, 1937–1942. Rudin, M. and Weissleder, R. (2003). Molecular imaging in drug discovery and development. Nat Rev Drug Discov 2, 123–131. Saleem, A., Harte, R.J., Matthews, J.C., Osman, S., Brady, F., Luthra, S.K., Brown, G.D., Bleehen, N., Connors, T., Jones, T., Price, P.M. and Aboagye, E.O. (2001). Pharmacokinetic evaluation of N-[2-(dim ethylamino)ethyl]acridine-4-carboxamide in patients by positron emission tomography. J Clin Oncol 19, 1421–1429.
■
499
Shah, K., Tung, C.H., Breakefield, X.O. and Weissleder, R. (2005). In vivo imaging of S-TRAIL-mediated tumor regression and apoptosis. Mol Ther 11, 926–931. Sweet-Cordero, A., Mukherjee, S., Subramanian, A., You, H., Roix, J.J., Ladd-Acosta, C., Mesirov, J., Golub, T.R. and Jacks, T. (2005). An oncogenic KRAS2 expression signature identified by cross-species gene-expression analysis. Nat Genet 37, 48–55. Tahara, N. et al. (2006). Simvastatin attenuates plaque inflammation: evaluation by fluorodeoxyglucose positron emission tomography. J Am Coll Cardiol 48, 1825–1831. Tang, Y., Kim, M., Carrasco, D., Kung, A.L., Chin, L. and Weissleder, R. (2005). In vivo assessment of RAS-dependent maintenance of tumor angiogenesis by real-time magnetic resonance imaging. Cancer Res 65, 8324–8330. Tawakol, A., Migrino, R.Q., Bashian, G.G., Bedri, S., Vermylen, D., Cury, R.C., Yates, D., Lamuraglia, G.M., Furie, K., Houser, S. et al. (2006). In vivo 18F-fluorodeoxyglucose positron emission tomography imaging provides a noninvasive measure of carotid plaque inflammation in patients. J Am Coll Cardiol 48, 1818–1824. Weissleder, R. (2002). Scaling down imaging: Molecular mapping of cancer in mice. Nat Rev Cancer 2, 11–18. Weissleder, R. and Ntziachristos, V. (2003). Shedding light onto live molecular targets. Nat Med 9, 123–128. Weissleder, R. and Pittet, M.J. (2008). Imaging in the era of molecular oncology. Nature 452, 580–589. Weissleder, R., Tung, C.H., Mahmood, U. and Bogdanov, A., Jr. (1999). In vivo imaging of tumors with protease-activated near-infrared fluorescent probes. Nat Biotechnol 17, 375–378. Weissleder, R., Kelly, K., Sun, E.Y., Shtatland, T. and Josephson, L. (2005). Cell-specific targeting of nanoparticles by multivalent attachment of small molecules. Nat Biotechnol 23, 1418–1423. Wieder, H.A., Brucher, B.L., Zimmermann, F., Becker, K., Lordick, F., Beer, A., Schwaiger, M., Fink, U., Siewert, J.R., Stein, H.J. et al. (2004). Time course of tumor metabolic activity during chemoradiotherapy of esophageal squamous cell carcinoma and response to treatment. J Clin Oncol 22, 900–908. Wu, K.D., Cho, Y.S., Katz, J., Ponomarev, V., Chen-Kiang, S., Danishefsky, S.J. and Moore, M.A. (2005). Investigation of antitumor effects of synthetic epothilone analogs in human myeloma models in vitro and in vivo. Proc Natl Acad Sci USA 102, 10640–10645.
CHAPTER
44 PET Imaging in Genomic Medicine Vikas Kundra and Osama Mawlawi
INTRODUCTION This chapter describes the principles and techniques of PET imaging with an emphasis on applications for analysis of the genome and its derivative products, RNA and protein. Nuclear medicine, including PET, affords the possibility to image in vivo biologic processes in animals and patients that were heretofore confined to in vitro or ex vivo analysis. DNA synthesis, promoter function/gene expression, mRNA, protein expression and function, and protein–protein interactions may each be evaluated at a whole organism level. With appropriate experimental design, gene expression, for example, can be studied specifically in the organ of interest and compared to expression after an intervention, such as pharmacotherapy or radiation therapy. In the setting of gene therapy, one would be able to quantify the amount of expression, to design appropriate dosing schedules and compare with clinical outcome, and localize expression, to determine if it is appropriately located for therapeutic effect or if it is in an undesirable location that may be resulting in toxicity. Because the imaging agents are noninvasive, serial imaging may be performed. Thus, in biomarker discovery, PET enables the possibility to quantitatively recapitulate in vitro analyses from the level of DNA to protein function in the whole animal over time. Clinically, PET imaging has already proven utility in diagnosing and staging disease and monitoring treatment. Although the illustrations in this chapter utilize PET, other imaging modalities such as gamma camera (planar and/or single
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 500
photon emission computed tomography, SPECT) and lightbased techniques may also be applied to biological questions, in both research and medicine. A primary advantage of nuclear medicine–based techniques is superior tissue penetration when compared to light-based imaging, making nuclear medicine more amenable to clinical translation for percutaneous imaging. Sophisticated tomographic techniques for three-dimensional localization are currently utilized in PET and SPECT imaging. For PET imaging, coincidence imaging improves localization and sensitivity. Higher resolution may be achieved by gamma camera imaging using magnification techniques, which can be important for imaging small parts in patients and small animals. Nuclear medicine relies on radiopharmaceuticals to image targets and/or function. With the appropriate imaging agents and/ or reporter constructs, PET imaging enables imaging of fundamental cellular processes such as nucleotide utilization, transcription, metabolism, amino acid utilization, and protein function.
PHYSICS Basic Principles Before examining specific applications of PET imaging in genomic or personalized medicine, it is important to introduce the basic physics principles of this approach. PET is based on imaging radionuclides that decay by positron emission. The
Copyright © 2009, Elsevier Inc. All rights reserved.
Physics
radionuclides are usually introduced synthetically into a molecule of potential biological relevance and administered to a patient, usually intravenously. Positron emission is one of two decay schemes that characterizes radionuclides that possess an excess number of protons to neutrons in their nucleus. In this regard, the positron emission decay process reduces the number of protons by transforming a proton to a neutron along with the creation and emission of a positron () and a neutrino ( ) from the nucleus (Figure 44.1). A positron is a charged particle that has the same mass but opposite charge to an electron. Upon emission, positrons travel a short distance while losing their kinetic energy through multiple collisions with electrons in surrounding tissues. Eventually, the positron combines with an electron in an annihilation reaction that transforms their combined mass into energy according to E mc2, resulting in two 511 keV gamma ray photons traveling in opposite directions. In PET imaging, a ring of radiation detectors is placed around the subject to detect the emitted gamma rays. The simultaneous emission and subsequent detection of the paired gamma rays by opposing detectors, called coincidence detection, represents a unique feature of PET imaging for event localization, and is capitalized upon by placing the annihilation event along a line of response (LOR) that connects the two detected gamma rays. In this regard, PET, unlike SPECT imaging, does not require a collimator to help identify the source of activity, hence, can have higher sensitivity.
511 KeV
tro
n
ra
ng
e
e f
ar
si
No
nc
oli
ne
Po
ity
β
511 KeV
Figure 44.1 Positron () decay: transformation of a proton to a neutron with the creation and ejection of a positron and a neutrino from the nucleus. The positron then travels a short distance (positron range) before annihilating with an electron (e) to generate two 511 KeV gamma rays that are emitted at 180°. Positrons that annihilate prior to loosing all of their kinetic energy produce noncollinear gamma rays which result in loss of resolution indicated by f on the figure.
■
501
Two main criteria are tested every time a detector registers an event – energy and timing. The first test assesses the energy of the incident gamma ray. Scintillation detectors made from dense crystalline materials linearly converts gamma ray energy to visible light, which is subsequently converted into an electrical pulse and amplified using a photomultiplier tube (PMT). If the electrical pulse amplitude falls inside the 511 keV preset energy threshold window, it is counted; otherwise, it is discarded. The timing test assesses the difference in the gamma ray detection time between a coincidence detector pair. A valid coincidence event represents detection of two 511 keV photons practically simultaneously (within a few nanoseconds). Others are discarded. Valid events are stored as part of a sinogram; after completing the acquisition, the sinogram is reconstructed into the final PET image. PET Scanner Design In a PET scanner, several rings of detectors are placed adjacent to one another axially. Current clinical scanners have axial extents of 15–25 cm. Animal scanners have maximum axial extents of 2–13 cm. PET scanners acquire data in either twoor three-dimensional (2D or 3D) modes. In the 2D mode, thin septa of lead or tungsten separate each ring, and coincidences are only recorded between detectors within the same ring or adjacent rings (Figure 44.2a). In the 3D mode, the septa are removed, and coincidences are recorded between detectors lying in any ring combination (Figure 44.2b), which increases sensitivity, but also reduces contrast. PET Data Acquisition Modes PET data can be acquired in static, dynamic, gated, or list modes. In static mode, valid coincidence events are accumulated in the sinogram over a defined period of time. Dynamic mode is akin to stacking multiple static mode acquisitions over time and is primarily used to study the distribution of a new radiopharmaceutical over time, as well as for radiation dosimetry studies. Gated mode is primarily used to freeze motion of organs of interest. In this mode, the PET data is acquired into multiple bins during each motion cycle. This process is usually facilitated by external monitoring devices such as EKG or respiratory bellows. In list mode, events are stored in separate memory locations with time marks. The acquired list data can then be rebinned retrospectively, providing flexibility in motion correction and temporal resolution. Furthermore, if the patient moves during any portion of the scan time, that artifactual data can be discarded. PET Data Correction One of the main advantages of PET imaging is its ability to provide accurate quantification of radionuclide concentration. This advantage, however, is largely dependent on the application of numerous corrections to the acquired data including attenuation of the gamma rays (Carroll et al., 1983; Huang et al.,
502
CHAPTER 44
■
PET Imaging in Genomic Medicine
(a) 2D
Septa
(b) 3D
Figure 44.2 2D versus 3D PET imaging. (a) In 2D PET, lead or tungsten septa (red lines) are present in the field of view limiting the number of oblique line of response (LOR) that are detected along the axial direction of the scanner. (b) In 3D, the septa are removed allowing oblique LORs to be detected thereby increasing the scanner sensitivity but also increasing the amount of detected scatter events (noise).
1979; Kinahan et al., 1998; Phelps et al., 1975; Xu et al., 1991), random and scatter events (Bailey and Meikle, 1994; Casey and Hoffman, 1986; Cherry and Huang, 1995; Cooke et al., 1983; Hoffman et al., 1979; Ollinger, 1996; Shao et al., 1994), variation between detector responses (Bailey et al., 1996; Hoffman et al., 1989), detector dead time (Badawi and Marsden, 1999; DaubeWitherspoon and Carson, 1991), and decay of radioactivity. Attenuation correction deserves special mention. Attenuation correction accounts for the missing gamma rays absorbed by the patient to more accurately quantify the injected activity. This is commonly performed by acquiring transmission data (by rotating a gamma ray source around the patient) to generate an attenuation correction map to account for the different tissues. Because of their small size and hence low attenuation, small animal PET imaging is often performed without this correction. In hybrid PET–CT scanners, CT data are used for PET attenuation correction. CT-generated attenuation maps have low noise content. CT provides short acquisition time, which decreases patient motion and allows more accurate anatomic localization of the PET signal. Small animal PET–CT scanners have been recently developed. Image Reconstruction Image reconstruction by filtered back projection or iterative reconstruction follows data correction. In filtered back projection,
coincidence detectors acquire a series of LORs to form a profile of detected events versus distance for each angle. Parallel LORs are grouped to form a projection profile for each angle. (The sinogram is a collection of these projections acquired from different angles.) In back projection, each profile is then back projected along its acquired angle into the image grid. The summation has inherent blurring because the exact location of the annihilation event is not known; we only know that the event occurred somewhere along the path of the LOR. Filters attempt to minimize blurring. The choice of filters largely depends on the desired sharpness and noise in the final image. In iterative reconstruction, an initial guess of the activity distribution in the subject is made. Then, projection images are calculated from this guess and compared to the acquired projection data. This process is repeated until a match between the initial guess and the acquired data is found. Stop conditions can be based on residual difference between the measured and calculated projection images or number of iterations. This reconstruction approach is generally known as maximum likelihood expectation maximization (MLEM). The most widely used variant is ordered subset expectation maximization (OSEM) (Hudson and Larkin, 1994), which divides the projection images into several subsets, so that, rather than comparing the whole projection images to one another, subsets are compared, thereby expediting the reconstruction. Factors Affecting PET Image Resolution One of the most striking features of PET images is their characteristic low resolution. PET images cannot reproduce sharp transitions such as edges on CT and MR images. There are several reasons for the low resolution of PET images, including detector size, detector type/design, positron range, non-collinearity of the emitted 511 keV gamma rays, depth of interaction effects, linear and angular sampling, and image reconstruction parameters. All of these factors except for reconstruction parameters are intrinsic to the scanner and set the intrinsic scanner resolution limit. For current clinical scanners, this limit is about 4–7 mm, whereas for animal scanners, the limit is 1–3 mm. Positron range pertains to the fact that the annihilation event that generates the paired 511 keV gamma rays does not occur at the site of positron emission, but at some distance away depending on the kinetic energy of the emitted positron (Figure 44.1). The kinetic energy, hence the distance, is dependent on the radionuclide under investigation (Table 44.1) (Derenzo, 1985). Noncollinearity occurs when annihilation events occur before the positron looses all of its energy, resulting in slightly different than 180° (Cherry et al., 2003). The combined effects of the positron range and the noncollinearity suggest a theoretical spatial resolution limit for PET imaging. In contrast, decay from radionuclides used in gamma camera imaging originates from the location of the radionuclide itself and allows magnification to improve resolution. Due to depth-of-interaction, the resolution of PET images degrades as one moves radially from the center to the edge of
Imaging Agents and Methods in Analysis of Bilogical Samples
TABLE 44.1 Physical properties of positron-emitting radionuclides used in PET Radionuclide
Physical half-life T1/2
Maximum ⴙ energy (MeV)
ⴙ Range (Rrms) in water (mm)
Carbon-11
20.4 min
0.96
0.4
Nitrogen-13
9.96 min
1.2
0.6
Oxygen-15
2.05 min
1.7
0.9
Fluorine-18
1.83 h
0.64
0.2
Copper-62
9.74 min
2.9
1.6
Copper-64
12.7 h
0.58
0.2
Gallium-66
9.49 h
3.8
3.3
Gallium-68
1.14 h
1.9
1.2
Bromine-76
16.1 h
3.7
3.2
Rubidium-82
1.3 min
3.4
2.6
Yttrium-86
14.7 h
1.4
0.7
Iodine-124
4.18 days
1.5
0.8
Adapted from Derenzo, S.E. (1986).
PET scanner B A Patient
D A’ B’
Figure 44.3 Depth of interaction. The two gamma rays generated from the annihilation event (red spot) will be detected in detectors BB’ rather than AA’ due to scanner curvature. The location of the measured LOR (dotted line) will be different from the true LOR (dashed line). Depth of interaction results in a radial loss of spatial resolution indicated by D on the figure. Spatial resolution is optimal at the center of the scanner.
■
503
the field of view (FOV). At the edge of the FOV, a gamma ray has a higher probability of penetrating the first detector and being detected in an adjacent detector. This has greater effect on small animal scanners due to their small bore size (Figure 44.3) and practically suggests placing a subject in the center of the FOV for best resolution (Bartzakos and Thompson, 1991). The chosen reconstruction algorithm and its parameters significantly contribute to image resolution. In iterative reconstruction, as the number of iterations and subsets increase, the resolution improves, but the noise increases (Schoder et al., 2004). High frequency filters result in sharper images with higher resolution, but greater noise. Conversely, low frequency filters result in smooth images, but loss of detail. Clinically, two iterations and 20–30 subsets are commonly used as a trade-off between image sharpness and noise.
IMAGING AGENTS AND METHODS IN ANALYSIS OF BIOLOGICAL SAMPLES Radiopharmaceuticals In general, radiopharmaceuticals for gamma camera imaging consist of three parts, (i) the localizing agent, (ii) a linker, and (iii) a radionuclide. This has proven successful for many clinical applications. In comparison, radionuclides for PET commonly allow direct labeling of the localization agent without an intervening linker. For example, a hydroxyl group may be directly replaced by 18F. Other radionuclides, such as 11C, 15O, 13N, 82Rb, 64Cu, and 68Ga among others, may be more advantageous in particular situations. Many atoms that naturally occur in molecules of interest also have forms that are positron emitters, for example, 11 C, 15O, and 13N. Others, such as 18F, 64Cu, and 68Ga, may be used to label molecules while maintaining specificity and favorable biodistribution. In addition to chemistry, considerations for choosing a radionuclide include availability and half-life. One advantage of 68Ga is that it can be produced from a generator and thus may be more available than radionuclides that need to be produced in a cyclotron. 18F is commonly used because of its chemistry and its favorable 110-min half-life (Table 44.2). Imaging DNA DNA Replication For imaging DNA synthesis, radioactive nucleotide analogs have been used. Focus has centered on thymine analogs to avoid incorporation into RNA. For example, 18F labeled 3-deoxy-3fluorothymidine (18F-FLT) has been used as a marker of DNA replication. As with all drugs, one must be careful that the desired biologic event is the one actually being imaged. 18F-FLT is also retained intracellulary after phosphorylation by thymidine kinase 1 (Rasey et al., 2002); a significant percentage may not be incorporated during DNA synthesis, but may reflect the activity of this enzyme. Other thymidine analogs such as 18F-1-(2-deoxy-2fluoro--D-arabinofuranosyl)thymine (18F-FMAU) are also under investigation.
504
CHAPTER 44
TABLE 44.2 proteomics
■
PET Imaging in Genomic Medicine
PET applications in genomics and
Phosphodiester
2-O -methyl
Phosphorothioate
Evaluation of ●
promoter function
●
endogenous gene and exogenous gene expression
●
RNA transcripts
●
protein expression
●
protein function
●
protein–protein interaction
●
metabolism
Direct Methods For imaging specific sequences of genomic DNA, direct methods that take advantage of Watson–Crick base pairing have been proposed. Short stretches of DNA, oligonucleotides, produced in the antisense orientation may be used to bind specific stretches of DNA. For in vivo imaging, the oligonucleotide needs to be stable in the blood stream, have limited interaction with nonspecific macromolecules, be taken up by the cell, bind a specific stretch of DNA, and have less efflux from the cell with the complementary DNA compared to the cell without the complementary DNA. The native phosphodiester backbone of DNA is susceptible to degradation by nucleases found in plasma. Therefore, oligonucleotides with modified backbones have been created, such as phosphorothioates, methylphosphonates, and peptide nucleic acids, to enhance stability. Unlike therapy, degradation of the target DNA (e.g., by affecting RNAse H activity) is not required for imaging. Cellular uptake of oligonucleotides appears to be receptor-mediated since it is temperature dependent, appears to be saturable, and can be competed by other oligonucleotides (Iversen et al., 1992; Loke et al., 1989). This is a potential barrier for imaging, because it may limit the number of oligonucleotides that can enter the cell. Fortunately, subcellular fractionation studies indicate that once taken up, oligonucleotides are found where DNA is found, in nuclei and mitochondria (Iversen et al., 1992). For imaging, the oligonucleotides’ analogs have been linked to radiopharmaceuticals for both gamma camera and PET imaging (Figure 44.4). For example, labeling has been successfully performed using 18F. The PET was then used to follow in vivo kinetics and biodistribution of the labeled oligonucleotides (Tavitian et al., 1998). Once created, specificity needs to be demonstrated using controls such as a random sequence, a sequence with one or more base alterations, and imaging cells without the target DNA sequence or with different number of copies of the target sequence. Accumulation of modified oligonucleotides labeled with positron (Kobori et al., 1999) and gamma-emitters have been reported (Cammilleri et al., 1996; Dewanjee et al., 1994; Shi et al., 2000; Urbain et al., 1995). PET
0–5 minutes after injection
60–65 minutes after injection
Figure 44.4 Coronal PET imaging of the biodistribution of different 18F labeled oligonucleotide analogs in nonhuman primates. With phophodiester and 2-O-methyl analogs, uptake is seen primarily in the vasculature, lungs, liver, kidneys, nasal area (and bladder with the phosphodiester) at 5 min and kidneys and bladder at 1 h. With the phosphorothioate analog, uptake is seen primarily in the vasculature, lungs, liver, spleen, nasal area, and kidneys at 5 min and liver, spleen, kidneys, and bladder at 1 h. (Images courtesy of Bertrand Tavitian, Copyright 1998 Bertrand Tavitian/CEA-INSERM U803 modified from Nature Medicine 1998;4:467–471.)
imaging studies targeting specific nucleotide sequences have focused on mRNA instead of DNA itself. For example, 11Clabeled antisense phosphorothioate oligodeoxynucleotide has been used to image the mRNA of the glial fibrillary acidic protein (GFAP) gene. Gliomas expressing GFAP were imaged in rats by antisense, but not sense or 30% mismatch probes (Kobori et al., 1999). Modified oligonucleotides also have the potential to distinguish point mutations. For example, in a preliminary study, a tumor with a point mutation of the K-ras oncogene was distinguished from a tumor with wild-type K-ras using a 68Galabeled phosphorothioate (Roivainen et al., 2004). Indirect Methods Indirect methods for imaging DNA include reporter genes. The constructs may be based on simple designs, incorporating a promoter to drive expression of a gene whose product can
Imaging Agents and Methods in Analysis of Bilogical Samples
be imaged. Based on the central dogma, this scheme provides a built-in amplification system that increases the number of targets that can be imaged, that is, transcription of genomic DNA results in tens to hundreds of mRNA and translation of mRNA produces thousands of copies of the encoded protein. If the protein can be imaged either directly or indirectly, it may thus serve as a reporter of gene expression. A limited number of reporters have been proposed for PET imaging. Examples include receptors such as the somatostatin receptor type 2 (SSTR2) and dopamine-2 receptor (D2R), pumps such as the sodium iodide symporter, and enzymes such as herpes simplex virus type 1-thymidine kinase (HSV1-TK). One advantage that may be afforded by enzymes is that they may amplify signal by their activity. For example, phosphorylation of substrates by thymidine kinase adds negative charges that prevent the substrates from crossing the cell membrane, resulting in accumulation of the substrates within the cell. Considerations in choosing a reporter include strength of eventual signal to noise required, immunogenicity, effect of the reporter on the cell, potential interference with the model/molecule being studied, cost, ease of preparation of the radiopharmaceutical used for imaging, model to be studied, and availability of Food and Drug Administration (FDA)-approved radiopharmaceuticals for eventual translation into the clinic. For example, in many situations it would be desirable to have a human, signaling-deficient receptor, such as that based on SSTR2. It may be necessary to use two different reporters in some instances, such as when evaluating expression from two different promoters. Then, for example, one may image a first reporter followed by the second reporter as long as the two radiopharmaceuticals do not have crossreactivity with the two reporters. Unlike gamma camera imaging, where different energy windows may be used for simultaneous imaging of radionuclides that emit at different energy peaks, positron emissions occur at only one peak, 511 KeV. Thus, imaging of two or more reporters by PET may be separated instead in time. The simple promoter–reporter construct may be used for transcriptional analysis. Activation of the promoter will result in expression of the reporter. After delivery (usually intravenously) of the radioligand/radiosubstrate for the reporter, positron emission is imaged by PET. With such a paradigm, promoter sequence and structure can be analyzed. For example, if one performs in vitro transcription analysis on various promoter mutants and finds a more robust/interesting sequence than the wild type, the new and wild-type mutant can be compared for activity in the appropriate in vivo context. One iteration may be to use three constructs with control, wild type, or mutant promoter sequence driving the same reporter. Expression may be confirmed in vitro, for example, by Western blotting for the reporter or radiopharmaceutical accumulation/binding. Using a variety of techniques, the construct may then be transferred into the organ under study, that is, mutant in one group of mice, wild type in the second group of mice, and control in the third group. Approximately 2 days later, the animals may be injected with the radiopharmaceutical for imaging. Region of interest analysis for
Control
■
505
Hypoxia-inducible promoter–reporter construct
Figure 44.5 In vivo microPET imaging of a hypoxiainducible promoter–reporter construct. Eight repeats of the hypoxia responsive HIF-1 alpha promoter driving expression of herpes simplex virus thymidine kinase-green fluorescent protein fusion reporter gene. Expression of the reporter was visualized by the PET tracer 18F-29-fluoro-29-deoxy-1-b-D-b-arabinofuranosyl-5-ethyluracil, [18F]FEAU, which is a substrate for thymidine kinase that once phosphorylated is entrapped in the cell. Greater expression is seen in the hypoxic center of the larger tumor (bottom right). Two sagittal microPET images per side of wild-type C6 (control, left) and transfected C6 (reporter, right) tumor xenografts of different size are presented. (Images courtesy of Dr J. Gelovani modified from Cancer Research 2004; 64:6101–6108.)
uptake of the radiopharmaceutical may be used to reflect promoter activity. After the appropriate time for clearance/decay of the radiopharmaceutical, the animals may be reinjected with the radiopharmaceutical if serial imaging is needed. In this way, the function of the mutant promoter may be compared to wildtype promoter in vivo, and of particular importance, in the organ of interest. The promoter–reporter paradigm has wide applicability for the study of multiple different genes in vivo. It, however, works best for strong promoters such as CMV. It has been successfully used with other promoters such as the hypoxia-sensitive HIF-1 promoter and the albumin promoter. For example, in larger tumors with hypoxic centers, the amount of HSV1-TK expression driven by the HIF-1 promoter was shown to increase as revealed by PET imaging (Figure 44.5) (Serganova et al., 2004). Promoter– reporter constructs may also be used to create transgenic mice. For example, in vivo PET imaging revealed that the degree of HSV1-TK expression driven by the albumin promoter varied depending on the amount of protein in the diet of albumin promoter–HSV1-TK transgenic mice (Green et al., 2002). Thus, promoter function may be evaluated in vivo. However, many promoters tend to have weak activity. This is particularly problematic for tissue-specific promoters. For these, amplification schemes can be helpful and may be cis or trans. In the trans system, for example, a weak promoter may be
506
CHAPTER 44
■
PET Imaging in Genomic Medicine
used to drive the expression of multiple intermediary proteins that drive a stronger promoter. Specificity is maintained by the weak promoter that drives the expression of the intermediary proteins. Amplification occurs by creation of multiple intermediary proteins, use of multiple binding sites for the intermediary proteins, and activation of a second, stronger promoter. One model is the use of a trans activation system based on a GAL4– VP16 fusion protein, expression of which is driven by a weak promoter. The GAL4 portion of the fusion proteins then binds to multiple GAL4 binding sites. This brings the VP16 activation domain portion of the fusion proteins adjacent to the promoter consisting of the GAL4 binding sites and a minimal TATA box to drive expression of a reporter gene. Using a modified HSV1-TK reporter, such an amplification system has been used to image androgen-dependent expression from an androgenresponsive, enhanced prostate-specific antigen promoter, the PSE-BC promoter (Sato et al., 2005). Imaging RNA Direct methods for imaging specific RNAs include modified oligonucleotides as described above. There have only been a few examples of specific RNA sequence imaging in vivo. Limitations for oligonucleotide imaging are described above. A major problem is the relatively few RNA molecules of any one sequence that can bind to RNA, resulting in a relatively small amount of signal for each detection event. An interesting paradigm for increasing signal has been described using luciferase imaging. This indirect method potentially may be translatable to PET imaging. It uses spliceosome-mediated RNA trans-splicing to drive expression of a reporter (Bhaumik et al., 2004). Engineered pre-trans-splicing molecules (PTMs) are embedded with active splicing elements. These are recognized by the cell’s splicing machinery and promote trans-splicing of the PTM encoded exon into the target transcript. The normal cis-splicing within the target pre-mRNA is avoided. The PTM binding domain can be designed to be complementary to intronic sequences in the target of interest in order to confer specificity to the trans-splicing event. Thus, the PTM includes a binding domain that will bind to a specific RNA sequence and a 3 acceptor site for trans-splicing. Proof of principle for this methodology was provided using luciferase. Expression of either the N-terminal or C-terminal part of this protein does not confer enzymatic activity, and thus no signal in the presence of the substrate luciferin. A PTM containing the N-terminal portion of luciferase was constructed with a 5 splice site followed by sequences of human papilloma virus-16 (HPV-16) that contained 3 splice sites. A second construct contained a PTM with a binding site for HPV-16 sequences to confer specificity and to bring the 3 splice site followed by the C-terminus of luciferase closer to 5 splice site. The HPV-16 sequences and 3 splice sites were excluded upon trans-splicing resulting in mRNA containing the full luciferase transcript that was transcribed into the enzyme. Upon addition of luciferin, signal in cells was seen after transfection of
both constructs suggesting splicing to bring both the mRNA N-terminus and C-terminus of luciferase into a new mRNA for expression of the entire protein. The experiment has limitations such as imaging expression of exogenous, not endogenous genes and a schema that is not easily generalizable to all mRNA. It does provide a tool for studying splicing in vivo that may be applicable to PET imaging in the future. It also demonstrates the utility of using reporter technology for amplifying signal. Imaging Proteins and Their Function Labeled amino acids have been used to evaluate their metabolism, for example, in the setting of cancer. For genomics and proteomics, proteins are of greater interest. A simple approach to image proteins is to radiolabel the ligand so that it retains binding to the protein, but not to nontarget proteins. An approach for enzymes is to radiolabel the substrate so that it is altered by the enzyme under study in such a way that it becomes entrapped in the cell. For pumps, again, the substrate may be radiolabeled, but in this case, the retention in the cell will be due to the action of the pump concentrating the radiolabeled substrate in the cell and not necessarily any alteration of the radiolabeled substrate. Retention may be enhanced by binding of the radiolabeled substrate to proteins within the cell. For all of these methods, the ligand/substrate usually must be relatively easy and economical to produce; stable in vivo; usually have minimal nonspecific binding such as to albumin; able to leave the vasculature; bound by the cell, taken up by the cell and/or penetrate the cell, and have a favorable route of excretion in order to produce minimal background in the organ of interest. Most importantly, the rate of efflux of the ligand/substrate in the target tissue must be greater than in nontarget tissue in order to produce a specific signal. Ligands An example of imaging with a radioligand is evaluation of the dopamine receptor D2R. Benzamides are ligands for dopamine receptors. A substituted benzamide, raclopride was labeled with the positron emitter 11C. Binding studies demonstrated that the Kd of 11C-raclopride for binding the D2R was 1.1 nM with relatively less affinity for binding other dopamine receptor subtypes (Farde et al., 1985). In monkeys, uptake of the compound was high in the striatum, where D2 receptors are expected, compared to nonspecific cerebellar binding. Further, the uptake could be competed by haloperidol, which binds D2 receptors particularly as well as other sites. Once a specific agent is created, it can find a variety of uses, including in pharmacology. For example, PET imaging of displacement of 11C-raclopride from the basal ganglia by cocaine has been used to suggest that cocaine modulates the amount of dopamine available for D2R binding (Schlaepfer et al., 1997). PET studies using 11C-raclopride have found a discrepancy in the mean plasma half-life of two antipsychotics, olanzapine and risperidone (24.2 and 10.3 h, respectively) and time to decline to 50% of peak striatal D2R occupancy (75.2 and 66.6 h, respectively). This suggests that
Imaging Agents and Methods in Analysis of Bilogical Samples
kinetics of receptor occupancy, such as that evaluated by PET, may improve dosing schedules of psychotropic medications compared to plasma kinetics (Tauscher et al., 2002). Enzyme Substrates An example of an enzyme substrate for PET imaging is 18Ffluoro-2-deoxyglucose (18F-FDG). It mimics glucose. It enters cells primarily via the glucose transporters (GLUT). Once inside the cell, it serves as a substrate for hexokinase. Phosphorylation of 18F-FDG is a terminal event, since, unlike glucose-6-phosphate, phosphorylated FDG cannot continue along the glycolytic pathway. The negative charge added upon phosphorylation prevents efflux. Thus, phosphorylated 18F-FDG accumulates in the cell. Unlike glucose, FDG does not undergo tubular reabsorption and is more readily excreted through the urine, decreasing background activity. Uptake of 18F-FDG is felt to reflect the metabolic rate of the cell. Although its uptake is felt to primarily reflect hexokinase activity, it is a complex event. For example, 18F-FDG accumulates to a greater degree in many cancers compared with normal tissues and reasons may include increased transport due to overexpression of membrane GLUT transporters (primarily GLUT1) as well as increased rate of glycolysis, for example, due to increased metabolic demand, hypoxia, signaling from oncogenes, increased hexokinase expression (primarily HK-1 and HK-2) and/or activity, and decreased glucose6-phosphatase activity. Increased 18F-FDG can be found in not only cancer, but also in inflammation and infection. Increased 18 F-FDG uptake reflects a functional process, not a specific disease; therefore, has found multiple applications.
(a)
R
507
Currently, 18F-FDG imaging has its greatest impact in oncology for diagnosis and staging and now monitoring of many cancers (Figure 44.6). It has also had a significant impact in cardiology, primarily for assessing ischemia and infarction. One of its major influences in neurology has been in seizure focus imaging. The applications of 18F-FDG PET imaging illustrate the myriad of pathologic events that may be evaluated by developing a radiotracer for a central biologic event, such as metabolism; and at the molecular level, in studying a key enzyme such as hexokinase. Use of 18F-FDG also illustrates that the readout of a molecular event need not be the immediate molecule under study, but may be a downstream event. One of the most successful new targeted drugs is Gleevec. Gleevec inhibits tyrosine kinases containing a kinase insert domain, such as the c-kit receptor. This receptor is overexpressed in gastrointestinal stromal tumors and is thought to be the primary inciting event for its malignant phenotype. At presentation, these tumors and their metastases avidly accumulate 18F-FDG, thus, PET imaging with this tracer aids diagnosis and staging. With Gleevec therapy, the kinase activity of the c-kit receptor is inhibited and the metabolic activity of the tumor is curtailed. Within 48 h after therapy, uptake of 18 F-FDG wanes (Stroobants et al., 2003), far before there is any anatomic change in tumor size. Thus, metabolic change reflects upstream inhibition of tyrosine kinase activity. And, 18F-FDG PET imaging may be used to assess whether Gleevec therapy will be successful. Further, when a tumor mutates so that it no longer is responsive to Gleevec therapy, it again accumulates 18FFDG (Shankar et al., 2005).
(b)
L R
■
(c)
L
R
L
Figure 44.6 Imaging of metabolism/enzymatic activity. PET–CT using 18F-FDG. 18F-FDG mimics glucose and its mechanisms of uptake include glut receptor expression on the plasma membrane and glucose-6 phosphate enzymatic activity, which phosphorylates and thus entraps the radiopharmaceutical inside the cell. Imaging of a 53-year-old man with lymphoma demonstrates increased 18FFDG uptake in the spleen and multiple intra- and extra-thoracic nodes, consistent with multifocal malignancy. Among its applications, 18 F-FDG may be used to stage and assess therapeutic response of lymphoma. (a) CT, (b) PET, (c) PET–CT fusion image.
508
CHAPTER 44
PET Imaging in Genomic Medicine
■
Pumps An example of a pump that can be imaged is the sodium iodide symporter. This pump drives its substrate into the cell using energy from ATP. Its actions result in greater accumulation of the substrate in the cell compared to outside of the cell. Substrates may include pertechnetate, which contains 99mTc for gamma camera imaging, as well as iodide. The symporter is expressed primarily in the thyroid and its use has proven clinical value for imaging thyroid pathology. Pertechnetate is an example of a substrate that results is accumulation in cells solely due to the action of the pump. On the other hand, in the thyroid, iodide is accumulated into the cell by the sodium iodide symporter; and, it is then entrapped by incorporation into thyroglobulin resulting in organification and may be subsequently used to produce thyroid hormone. In cancer cells, organification may not occur, and then iodide accumulation is primarily via the action of the sodium iodide symporter. Because the more complex process of organification is not required for accumulation of iodide, the sodium iodide symporter has been used as a reporter of gene transfer. Different isotopes of iodine are available, including those that emit primarily gamma rays, particles, or positrons. 124I emits positrons and has been used primarily for PET-based reporter imaging in small animals. The different isotopes of iodine also illustrate that different isotopes may be used not only for imaging a protein, but for therapy as well.
(a)
Protein–Protein Interactions To study protein–protein interaction in vivo by PET, a modification of the yeast two-hybrid system may be employed (Figure 44.7) (Luker et al., 2002). For example, a bidirectional promoter can be used to express two fusion proteins. To control for the amount of expression, the promoter may be inducible, for example, by tetracycline. This will result in a fixed ratio of expression of the two fusion proteins. If the ratio of expression needs to be varied, the fusion proteins may be expressed under the control of two different promoters. The first fusion protein may consist of the protein of interest 1 fused to a GAL4 binding domain. The second fusion protein may consist of protein of interest 2 fused to VP16. The reporter construct consists of a promoter containing multiple GAL4 binding sites and a minimal TATA box. The first fusion protein localizes to the GAL4 binding sites via the GAL4 portion of the fusion protein. The protein of interest 1 portion of the first fusion protein can then interact with the second fusion protein. When such interaction occurs with the protein of interest 2 portion of the second fusion protein, the VP16 activation domain of the second fusion protein drives expression
(c)
Dox
rtTa Transactivator
Gal4-BD-p53
In addition to emission of gamma rays for gamma camera imaging, 131I emits particles that can be used for therapy, for example, for Graves’ disease or for thyroid cancers that take up iodide.
VP16-TAg
Tet Promoter
(b) (d) p53
TAg
(e)
VP16
Gal4-BD Gal4
Gal4
Gal4
Gal4
Gal4 promoter
Gal4
TATA TATA box
HSV-TK
GFP
Fusion reporter
Figure 44.7 In vivo PET-based imaging of protein–protein interaction. A two-hybrid system was modified for detecting protein– protein interactions by fluorescence imaging and by in vivo PET imaging. Upon interaction, fusion proteins bring together transactivator and DNA binding domains of a transcription factor to induce transcription of a reporter gene. Two-hybrid systems are applicable to protein interactions that can translocate to the nucleus. (a) Construct of a bidirectional, tetracycline (Tet) inducible promoter for driving equal expression of two interacting proteins such as p53 fused to a Gal4 binding domain and the oncogenic T-antigen (TAg) fused to a VP16 activator domain. As a control, a protein that does not bind p53 such as Coat Protein (CP)-VP16 may replace TAgVP16. (b) Upon promoter induction by the tetracycline analog doxycycline (Dox), the two interacting proteins bring together GAL4 and VP16 to drive expression of a reporter such as Herpes simplex virus-thymidine kinase fused to green fluorescent protein. (c) Fluorescence due to green fluorescent protein (GFP) indicating protein–protein interaction. (d) Photograph of the thorax of a nu/nu mouse with axillary xenograft tumors derived from cells stably transfected with the negative control p53-GAL4 CP-VP16 (arrowhead) or p53-GAL4 TAg-VP16 (arrow). (e) Coronal microPET image of the same mouse 48 h after Doxycycline induction. Uptake of the HSV-TK substrate 18F-FHBG is seen only in the tumor derived from cells transfected with p53-GAL4 TAg-VP16 (arrow), confirming interaction of p53 and TAg. Asterisk denotes excretion of radiotracer into the gallbladder. Intestinal activity (I) from normal hepatobiliary clearance of the radiotracer was observed in the lower portion of the image. (Image courtesy of David Piwnica-Worms and modified from Proceedings of the National Academy of Sciences USA 2002;99:6961–6966.)
Imaging Agents and Methods in Analysis of Bilogical Samples
of the reporter. A radiopharmaceutical can then be administered for PET imaging of the expressed reporter. Protein–protein interaction based on two-hybrid technology has been employed for in vivo PET imaging to demonstrate interaction of p53 and the large T antigen of simian virus-40 (Luker et al., 2002).
(a)
(b)
In vivo image (SPECT and MR) derived % injected dose/g
R 0.87
1.5
1
509
Multimodality Imaging One of the problems with functional imaging is the lack of anatomic detail. As tracers improve, signal will become more and more specific to the region of its origin resulting in a spot on a functional image without anatomic reference. Another problem is that function does not always equal form.The signal from a greater amount of radioactivity may appear larger than a signal from a smaller amount of radioactivity although the shape and size of the underlying object is the same. The apparent size and apparent signal of the object may also be influenced by volume averaging artifacts that may occur when the resolution of the object of interest is less than 2.7 times the resolution of the imaging system (Kessler et al., 1984). Reference to an anatomic image, such as that generated by CT or MR, can be helpful in understanding the underlying anatomy and size of the object of interest in order to improve quantification of the functional signal. Using MR and gamma camera imaging, this has been demonstrated for image-based quantification of gene expression of a somatostatin receptor–based reporter (Figure 44.8) (Yang et al., 2005). Thus, the functional
Prescan procedures
(c)
■
CT component
PET component
Injection of 18 F-FDG
Patient positioning
Couch moved to PET FOV
60–90 min uptake phase
CT scout acquired (confirm/adjust position)
Empty bladder
CT scan acquired
PET emission acquisition
CT reconstruction
PET data correction and image reconstruction
0.5
0 0
0.5 1 Ex vivo tumor tissue derived % injected dose/g
Figure 44.8 Gene expression in tumors can be quantified by combining functional and anatomic imaging. (a) Functional gamma camera–based tomographic (SPECT) imaging of a mouse with tumors expressing greater (arrow) and lesser (arrowhead) amounts of exogenously introduced hemaglutinin A–tagged somatostatin receptor type-2 reporter gene 1 day after intravenous injection of 111In-octreotide. (b) Anatomic T2-weighted MR imaging of the same tumors. Note that the internal morphology of the tumors can be identified (arrow). (c) Reporter gene expression assessed by in vivo image-based biodistribution (% injected dose per gram) correlates with that derived from excised tumors. (a) and (b) are axial images at the level of the shoulders. (Images modified from Radiology 2005;235:950–958.)
Attenuation maps generated
Viewable CT images
Occurs in parallel to PET component
Viewable PET images
Viewable fused PET/CT images
Figure 44.9 Flow chart of 18F-fluoro-deoxy glucose (18FFDG) PET–CT imaging sequence of a patient with cancer. See text for description.
510
CHAPTER 44
■
PET Imaging in Genomic Medicine
signal can be normalized to the size or weight of the object of interest. With new fusion system incorporating functional and anatomic imaging, such as PET–CT (Figure 44.6) or SPECT–CT, anatomic localization of the functional signal is improved. PET– MR and SPECT–MR systems are under development. As with PET imaging, care must be taken to use appropriate protocols for anatomic imaging to maximize signal and anatomic delineation. Fusion systems are available for small animal imaging. PET–CT has already proven valuable in evaluating patients. Sample PET–CT Imaging Protocol Figure 44.9 is a flow chart of an example 18F-FDG PET–CT imaging sequence of a patient with cancer. It may be of assistance in developing patient and/or animal PET studies. Critical parameters include radiopharmaceutical dose, delay (incubation phase), voiding, subject positioning, and duration of PET scanning (to obtain sufficient counts). Intravenous and oral contrast may be used to improve anatomic delineation by CT.
CONCLUSION After finding increased or decreased expression in a gene array or after proteomic profiling (see Chapters 13 and 14), one can potentially image the gene product/protein with an appropriate PET radiopharmaceutical, as outlined in this chapter. Evaluation of the promoter driving the expression may be evaluated using reporter technology. A promoter–reporter construct may be introduced exogenously by gene transfer, including creating transgenic animals. RNA transcripts may be imaged using labeled oligonucleotides and reporter systems. Protein function and protein–protein interactions are approachable. In short, PET imaging has the potential to quantitatively image processes from the level of DNA to the tissue in vivo. Clinically, it has already found significant utility via 18F-FDG for diagnosing and staging disease and monitoring treatment.
REFERENCES Badawi, R.D. and Marsden, P.K. (1999). Self-normalization of emission data in 3D PET. IEEE Trans Nucl Sci 46, 709–712. Bailey, D.L. and Meikle, S.R. (1994). A convolution-subtraction scatter correction method for 3D PET. Phys Med Biol 39, 411–424. Bailey, D.L., Townsend, D.W., Kinahan, P.E., Grootoonk, S. and Jones, T. (1996). An investigation of factors affecting detector and geometric correction in normalization of 3-d pet data. IEEE Trans Nucl Sci 43, 3300–3307. Bartzakos, P. and Thompson, C.J. (1991). A PET detector with depthof-interaction determination. Phys Med Biol 36, 735–748. Bhaumik, S., Walls, Z., Puttaraju, M., Mitchell, L.G. and Gambhir, S.S. (2004). Molecular imaging of gene expression in living subjects by spliceosome-mediated RNA trans-splicing. Proc Natl Acad Sci U S A 101, 8693–8698. Cammilleri, S., Sangrajrang, S., Perdereau, B. et al. (1996). Biodistribution of iodine-125 tyramine transforming growth factor alpha antisense oligonucleotide in athymic mice with a human mammary tumour xenograft following intratumoral injection. Eur J Nucl Med 23, 448–452. Carroll, L., Kertz, P. and Orcutt, G. (1983). The orbiting rod source: Improving performance in PET transmission correction scans. In Emission Computed Tomography: Current Trends (235–247) (P.D. Esser, ed.), Society of Nuclear Medicine, New York, NY. Casey, M.E. and Hoffman, E.J. (1986). Quantitation in positron emission computed tomography: 7. A technique to reduce noise in accidental coincidence measurements and coincidence efficiency calibration. J Comput Assist Tomogr 10, 845–850. Cherry, S.R. and Huang, S.-C. (1995). Effects of scatter on model parameter estimates in 3D PET studies of the human brain. IEEE Trans Nucl Sci 42, 1174–1179. Cherry, S.R., Sorenson, J.A., Phelps, M.E. and Sorenson, J.A. (2003). Physics in nuclear medicine, 3rd ed.. Saunders, Philadelphia, PA. xiii, 523 p. Cooke, B.E., Evans, A.C., Fanthome, E.A., Alarie, R. and Sendyk, A.M. (1983). Performance figures and images from the Therascan
3128 positron emission tomograph. IEEE Trans Nucl Sci NS-31, 640–644. Daube-Witherspoon, M.E. and Carson, R.E. (1991). Unified deadtime correction model for PET. IEEE Trans Med Imaging 10, 267–275. Derenzo, S.E. (1985). Mathematical removal of positron range blurring in high resolution tomography. IEEE Trans Nucl Sci NS-33. Dewanjee, M.K., Ghafouripour, A.K., Kapadvanjwala, M. et al. (1994). Noninvasive imaging of c-myc oncogene messenger RNA with indium-111-antisense probes in a mammary tumor-bearing mouse model. J Nucl Med 35, 1054–1063. Farde, L., Ehrin, E., Eriksson, L. et al. (1985). Substituted benzamides as ligands for visualization of dopamine receptor binding in the human brain by positron emission tomography. Proc Natl Acad Sci U S A 82, 3863–3867. Green, L.A.,Yap, C.S., Nguyen, K. et al. (2002). Indirect monitoring of endogenous gene expression by positron emission tomography (PET) imaging of reporter gene expression in transgenic mice. Mol Imaging Biol 4, 71–81. Hoffman, E.J., Guerrero, T.M., Germano, G., Digby, W.M. and Dahlbom, M. (1989). PET system calibrations and corrections for quantitative and spatially accurate images. IEEE Trans Nucl Sci 36, 1108–1112. Hoffman, E.J., Huang, S.C. and Phelps, M.E. (1979). Quantitation in positron emission computed tomography: 1. Effect of object size. J Comput Assist Tomogr 3, 299–308. Huang, S.C., Hoffman, E.J., Phelps, M.E. and Kuhl, D.E. (1979). Quantitation in positron emission computed tomography: 2. Effects of inaccurate attenuation correction. J Comput Assist Tomogr 3, 804–814. Hudson, H.M. and Larkin, R.S. (1994). Accelerated image reconstruction using ordered subsets of projection data. IEEE Trans Med Imaging 13, 601–609. Iversen, P.L., Zhu, S., Meyer, A. and Zon, G. (1992). Cellular uptake and subcellular distribution of phosphorothioate oligonucleotides into cultured cells. Antisense Res Dev 2, 211–222.
Recommended Resources
Kessler, R.M., Ellis, J.R., Jr. and Eden, M. (1984). Analysis of emission tomographic scan data: Limitations imposed by resolution and background. J Comput Assist Tomogr 8, 514–522. Kinahan, P.E., Townsend, D.W., Beyer, T. and Sashin, D. (1998). Attenuation correction for a combined 3D PET/CT scanner. Med Phys 25, 2046–2053. Kobori, N., Imahori, Y., Mineura, K., Ueda, S. and Fujii, R. (1999). Visualization of mRNA expression in CNS using 11Clabeled phosphorothioate oligodeoxynucleotide. Neuroreport 10, 2971–2974. Loke, S.L., Stein, C.A., Zhang, X.H. et al. (1989). Characterization of oligonucleotide transport into living cells. Proc Natl Acad Sci U S A 86, 3474–3478. Luker, G.D., Sharma, V., Pica, C.M. et al. (2002). Noninvasive imaging of protein-protein interactions in living animals. Proc Natl Acad Sci U S A 99, 6961–6966. Phelps, M.E., Hoffman, E.J., Mullani, N.A. and Ter-Pogossian, M. M. (1975). Application of annihilation coincidence detection to transaxial reconstruction tomography. J Nucl Med 16, 210–224. Ollinger, J.M. (1996). Model-based scatter correction for fully 3D PET. Phys Med Biol 41, 153–176. Rasey, J.S., Grierson, J.R., Wiens, L.W., Kolb, P.D. and Schwartz, J.L. (2002).Validation of FLT uptake as a measure of thymidine kinase-1 activity in A549 carcinoma cells. J Nucl Med 43, 1210–1217. Roivainen, A.,Tolvanen,T., Salomaki, S. et al. (2004). 68Ga-labeled oligonucleotides for in vivo imaging with PET. J Nucl Med 45, 347–355. Sato, M., Johnson, M., Zhang, L., Gambhir, S.S., Carey, M. and Wu, L. (2005). Functionality of androgen receptor-based gene expression imaging in hormone refractory prostate cancer. Clin Cancer Res 11, 3743–3749. Schlaepfer, T.E., Pearlson, G.D., Wong, D.F., Marenco, S. and Dannals, R. F. (1997). PET study of competition between intravenous cocaine and [11C]raclopride at dopamine receptors in human subjects. Am J Psychiatry 154, 1209–1213. Schoder, H., Erdi,Y.E., Chao, K., Gonen, M., Larson, S.M. and Yeung, H. W. (2004). Clinical implications of different image reconstruction
■
511
parameters for interpretation of whole-body PET studies in cancer patients. J Nucl Med 45, 559–566. Serganova, I., Doubrovin, M.,Vider, J. et al. (2004). Molecular imaging of temporal dynamics and spatial heterogeneity of hypoxia-inducible factor-1 signal transduction activity in tumors in living mice. Cancer Res 64, 6101–6108. Shao, L., Freifelder, R. and Karp, J.S. (1994). Triple energy window scatter correction technique in PET. IEEE Trans Medi Imaging 13, 641–648. Shankar, S., vanSonnenberg, E., Desai, J., Dipiro, P.J.,Van Den Abbeele, A. and Demetri, G.D. (2005). Gastrointestinal stromal tumor: New nodule-within-a-mass pattern of recurrence after partial response to imatinib mesylate. Radiology 235, 892–898. Shi, N., Boado, R.J. and Pardridge, W.M. (2000). Antisense imaging of gene expression in the brain in vivo. Proc Natl Acad Sci U S A 97, 14709–14714. Stroobants, S., Goeminne, J., Seegers, M. et al. (2003). 18FDG-Positron emission tomography for the early prediction of response in advanced soft tissue sarcoma treated with imatinib mesylate (Glivec). Eur J Cancer 39, 2012–2020. Tavitian, B., Terrazzino, S., Kuhnast, B. et al. (1998). In vivo imaging of oligonucleotides with positron emission tomography. Nat Med 4, 467–471. Tauscher, J., Jones, C., Remington, G., Zipursky, R.B. and Kapur, S. (2002). Significant dissociation of brain and plasma kinetics with antipsychotics. Mol Psychiatry 7, 317–321. Urbain, J.L., Shore, S.K., Vekemans, M.C. et al. (1995). Scintigraphic imaging of oncogenes with antisense probes: Does it make sense?.Eur J Nucl Med 22, 499–504. Yang, D., Han, L. and Kundra, V. (2005). Exogenous gene expression in tumors: Noninvasive quantification with functional and anatomic imaging in a mouse model. Radiology 235, 950–958. Xu, E.Z., Mullani, N.A., Gould, K.L. and Anderson, W.L. (1991). A segmented attenuation correction for PET. J Nucl Med 32, 161–165.
RECOMMENDED RESOURCES Bendriem, B. and Townsend, D.W. (1998). The Theory and Practice of 3D PET. Kluwer Academic, Dordrecht, London, xvi, 167 pp. Humm, J.L., Rosenfeld, A. and Del Guerra, A. (2003). From PET detectors to PET scanners. Eur J Nucl Med Mol Imaging 30, 1574–1597. Kelloff, G.J., Krohn, K.A., Larson, S.M., et al. (2005). The progress and promise of molecular imaging probes in oncologic drug development. Clin Cancer Res 11, 7967–7985. Mawlawi. O., Pan, T. and Macapinlac, H.A. (2006). PET/CT imaging techniques, considerations, and artifacts. J Thorac Imaging 21, 99–110. Nutt, R. (2002). 1999 ICP Distinguished Scientist Award. The history of positron emission tomography. Mol Imaging Biol 4, 11–26. Phelps, M.E. (2004). PET: Molecular Imaging and Its Biological Applications. Springer, New York, London, 615 pp. Valk, P.E. (2006). Positron Emission Tomography: Clinical Practice. Springer, London, xiv, 475 pp.
Zanzonico, P. (2004). Positron emission tomography: A review of basic principles, scanner design and performance, and current systems. Sem Nucl Med 34, 87–111.
Websites Badawi, R.D. (1999). Introduction to PET Physics. University of Washington Division of Nuclear Medicine. Retrieved on October 15, 2007 from http://depts.washington.edu/nucmed/IRL/pet_ intro/index.html Phelps, M.E., Gambhir, S.S., Mahoney, D.K. and Markham, J.A. (1993). Let’s Play PET. Regents of the University of California. Retrieved on October 15, 2007 from http://www.uib.no/med/avd/miapr/ arvid/MOD3_2002/Bildedannelse/lets_play_PET.pdf
CHAPTER
45 MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges Dmitri Artemov
INTRODUCTION Advances in molecular biology that include the decoding of the human genome have given a strong impetus to molecular-based and -targeted medicine. Genomic medicine is a novel concept that explores complex interactions within the genome and is focused on the role of multiple genes in human diseases (Guttmacher and Collins, 2002). These recent developments have resulted in an urgent need to measure noninvasively the expression pattern of a specific gene product(s) and to determine the response to molecular interventions, first in preclinical models of human disease that typically involve experimental animals such as mice and rats, but with the ultimate goal of clinical translation. Molecular imaging is a rapidly advancing area of biomedical research that can provide a matching set of tools to address the emerging demands of genomic and molecular medicine applications (Rome et al., 2007) (see Chapter 43). Monitoring of molecular events in vivo requires imaging modalities that permit repetitive in vivo measurement of these events with sufficient sensitivity and spatial resolution. Nuclear imaging techniques such as PET and SPECT are traditionally used for molecular imaging with highly specific probes and provide the highest Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 512
sensitivity of detection that allow detection of tracer concentrations of the probes combined with inherent three-dimensional (3D) imaging capabilities. The main problems of nuclear imaging are relatively low spatial resolution and the exposure of the subject to ionizing radiation. Optical imaging, an alternative technique that is widely used in preclinical models, allows detection of nanomolar concentrations of luminescent and fluorescent probes in vivo. Unfortunately optical imaging is restricted to imaging of relatively superficial tissue due to the strong absorption and scattering of light in living tissue.This also makes implementation of 3D imaging technology quite difficult especially for relatively large objects. Magnetic resonance imaging (MRI) is a unique imaging modality where the image intensity is a complex function of multiple parameters that can be effectively modified by choosing an appropriate imaging pulse sequence and by application of MR contrast agents. As a result MR images provide good delineation of morphological and functional parameters such as blood perfusion (Duyn et al., 2005) and tissue composition and micro-architecture (Mori et al., 2002). This inherently high spatial resolution of MRI is combined with a relatively low sensitivity, and typically concentrations in the millimolar range are required for robust MR detection. In a typical example of Copyright © 2009, Elsevier Inc. All rights reserved.
Basics of MRI Contrast
imaging of cell surface receptors, expressed at a density of one million receptors per cell, the resulting total concentration of the receptors is below 1 μmol/l. Therefore efficient strategies for signal enhancement are required to enable MRI of this important biological parameter. In this chapter, we will consider different approaches to improve MR capability for molecular imaging and will discuss in detail mechanisms of contrast generation such as T1 and T2 relaxation contrast, proton exchange, basic of MR spectroscopic imaging, and potential applications of the methods to clinically relevant problems. The ability to image the expression pattern of relevant genes noninvasively is an ultimate goal of molecular imaging applications in the field of genomic medicine.
BASICS OF MRI CONTRAST Source of the MR Signal MRI and spectroscopy are based on the detection of radiofrequency (RF) signals generated by magnetic nuclear spins precessing in the external magnetic field B0. The resonance frequency 0 of the MR linearly depends on B0 and the gyromagnetic ratio of the nucleus, as 0 B0. The intensity of the MR signal depends on (i) concentration of the nuclear spins and (ii) on the MR frequency or of this particular spin. These values are listed in Table 45.1 for nuclei used in biological MRI studies. Natural abundance shows the concentration of magnetic isotope or the percentage of the total number of nuclei for the given element with magnetic moment that is required to generate MR signal. From Table 45.1 it is apparent that proton (1H) is the most sensitive species and, due to the high content of protons in the form of water and/or fat in living tissues, proton MRI is most often used for in vivo studies. Therefore the discussion here is primarily limited to proton MRI and to various mechanisms, apart from proton concentration, that determine MRI signal intensity in the images. MRI Contrast Mechanisms Water constitutes more than 80% of living organisms and is uniformly distributed in soft tissues. Therefore, unless there are TABLE 45.1 in vivo MRI
Magnetic properties of nuclei used for
Nuclei
␥ [MHz/T]
u0 [MHz] @ B0 ⴝ 3T
Natural abundance (%)
1
H
42.58
128
100
3
He
32.44
97
1.37 E-4
13
10.71
32
1.11
19
40.08
120
100
23
11.27
34
100
31
17.25
52
100
C F Na P
■
513
differences in the concentration, MR images of proton concentration are usually featureless. On the other hand there are several other parameters that affect image intensity in MRI including relaxation, diffusion, flow rate, and water exchange. Relaxation A typical MRI experiment consists of a series of RF pulses applied to the spins of the object (excitation) and detection of the pulse response from the precessing spins (Gillies, 1994). After a single excitation cycle, the longitudinal magnetization (parallel to the external magnetic field B0) is reduced from its equilibrium value M0 to a new transient value Mt due to saturation effects of the RF pulses. If no further RF excitation is applied to the system during time t, then the magnetization M(t) relaxes back from Mt to its equilibrium value M0 according to the Bloch equation: M (t ) M t ⋅ exp(t /T1 ) M 0 ⋅ (1 exp(t /T1 )) T1 is called the longitudinal relaxation time of the spins, and for water in biological systems T1 typically is close to 1 s. Two important cases are usually defined: (i) when starting magnetization is completely saturated (Mt 0), so called “saturation recovery” and (ii) when initial magnetization is inverted (Mt M0), “inversion recovery.” Similarly, the excited precessing magnetization Mp decays to zero with a characteristic time T2, the transversal relaxation time, that generally is shorter than T1 due to (i) additional dephasing of rotating spins by local inhomogeneities of the magnetic field B0, (ii) diffusion, and (iii) additional relaxation mechanisms that are only effective for transversal relaxation (Slichter, 1990). The general equation for the transversal magnetization after complete excitation is: M p (t ) M p (0)exp(t /T2 ) An important case is complete excitation with Mp(0) M0. If local fields are refocused by the use of spin-echo (SE) pulse sequence, then we have true T2 decay. However, if the imaging pulse sequence is sensitive to dephasing processes due to the local field inhomogeneities such as gradient-echo (GE) imaging, a much faster decay of Mp is typically characterized by T2* relaxation time. A detailed discussion of T1 and T2 relaxation mechanisms is beyond the scope of this chapter and can be found in (Slichter, 1990). Further in the chapter we will discuss different classes of MR contrast agents and how they can modify inherent T1 and T2 to enhance imaging contrast. Diffusion Diffusion, or more correctly self-diffusion, of water molecules results in random changes of the position of the molecule with
514
CHAPTER 45
■
MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges
time. In an inhomogeneous magnetic field, B0, moving water molecules experience different strengths of the magnetic field in different positions. In this case the transversal magnetization is dephased by these randomly changed local fields at the current position of the molecule and cannot be completely refocused by RF pulses in a SE imaging sequence. For diffusion measurements a pair of variable diffusion weighting gradients is typically incorporated into a standard SE sequence. Signal is measured as a function of the gradient strength and, optionally, orientation. The magnetization decay depends on the diffusion coefficient of water molecules in the direction of the applied gradient. If a sufficient number of different gradient orientations are used then all the components of the diffusion tensor can be measured in each pixel of the MR image (Le Bihan et al., 1991). One example of application of this technique in MRI studies of human brain is determination of brain microfiber orientation that relates to white matter tracks as shown in Figure 45.1 (Mori et al., 2002). Water Exchange A novel contrast mechanism, CEST (chemical exchange saturation transfer), that utilizes proton exchange between bulk water and exchangeable chemical groups such as amides of proteins, imino- and hydroxyl protons (Guivel-Scharen et al., 1998) was proposed as an ultra-sensitive MR marker (McMahon et al., 2006; Snoussi et al., 2003). Briefly, when a prolonged RF pulse is applied to the system at the frequency of water resonance, water, immediately before imaging sequence, the resulting saturation of the water peak results in a very low signal ideally approaching zero for complete saturation. When the frequency of the saturation RF field shifts from the water
resonance, the efficiency of saturation decreases and the intensity of the water signal increases with the shift. However, if there is an exchangeable chemical group, group, then saturation on the resonance frequency of this group will lead to attenuation of the bulk water signal because magnetization of many exchanging water molecules will be saturated while their protons reside on the exchangeable group. Typically the intensity of the water resonance is measured for two saturation RF fields with frequencies equal to water , where group – water, and their difference is a measure of concentration of the exchangeable group. Two important properties of this contrast mechanism is a significant amplification of the signal because a single exchangeable group can affect signals of multiple water molecules and the possibility to express endogenous compounds that will produce a desirable CEST effect (McMahon et al., 2006). MRI Techniques Here we will discuss two most important modes of acquiring MR images that are based on (i) formation of spin-echo (SE) and (ii) gradient-echo (GE) imaging. Detailed description of the methods, as well as more modern MRI techniques that include fast and/or parallel imaging, can be found elsewhere (Brown, 2003). Diagrams of the SE and GE imaging pulse sequences are shown in Figure 45.2.
(a)
Excitation RF pulse
Refocusing RF pulse
Spin echo
Preparation RF t Gr t
Gp,s,d t (b)
Excitation RF pulse
Gradient echo
Preparation RF t Gr t Gp,s t
Figure 45.1 Diffusion Tension Imaging (DTI) uses water diffusion as a probe for white matter anatomy. Colors represent orientation of white matter tracks: red for right–left, green for anterior–posterior, and blue for superior–inferior orientation respectively (picture kindly provided by Dr Mori).
Figure 45.2 Time diagrams of 2D spin-echo (a) and gradientecho (b) MRI pulse sequences. Gr denotes read gradient and Gp,s,d phase encoding, slice selection, and diffusion weighting (solid blocks) gradients respectively.
MR Contrast Agents for Molecular Imaging Applications
The main features of the sequences are T2 and/or self-diffusion contrast for SE sequence that refocuses magnetic field inhomogeneities for stationary spins by echo formation with the refocusing RF pulse, and T2* contrast due to local magnetic fields for GE pulse sequence. The preparation period before the imaging sequence can be used to generate additional contrast. For example, T1 contrast can be generated by saturation or inversion of the initial magnetization and CEST contrast by selective saturation at frequencies of proton-exchangeable chemical groups as discussed above. Three-dimensional MRI sequences are analogous to 2D sequences discussed here and use a second phase-encoding gradient ramp. T2 and T2* contrast produced by shortening of T2 and T2* relaxation times respectively results in a negative contrast, that is, image intensity decreases in these regions. On the other hand, T1 contrast generated by shortening of T1 relaxation time produces positive contrast or brighter images that often are advantageous for image analysis. Recently, novel technologies were developed to generate positive contrast with T2* contrast agents by detecting off-resonance water signals (Cunningham et al., 2005) or use noncompensated local gradients generated by T2* contrast agent (Seppenwoolde et al., 2003).
MR CONTRAST AGENTS FOR MOLECULAR IMAGING APPLICATIONS Paramagnetic MR Contrast Agents The most important class of paramagnetic MR contrast agents is based on the transition metal, gadolinium. Gadolinium (Gd) has unique chemical and physical properties that make it an ideal MR tracer. It has nine coordination sites that form stable biologically inert complexes with linear and cyclic chelates such as DTPA and DOTA (Corot et al., 1998), it has an unpaired electron that has high magnetic moment (almost 700-fold higher than magnetic moment of a proton), inverse electron relaxation time is close to 0 for practically used MR magnetic fields, and correlation time of the complexes can be adjusted to 01, which makes it an efficient T1 contrast agent. Indeed, the strong magnetic moment of Gd fluctuates with a characteristic frequency e 0, and therefore significantly enhances T1 relaxation processes. To quantitatively measure the efficiency of a contrast agent relaxivity parameter, R, is often used. For T1 and T2 relaxation R is defined as:
R1, 2
1 1 T1, 2 T1, 20 [C ]
,
where T1,20 and T1,2 are corresponding relaxation times before and after addition of the contrast agent and [C] is the concentration of the contrast. Typical values of T1 relaxivity of lowmolecular weight GdDTPA/GdDOTA complexes are in the range 5–10 [s(mM Gd)]1. Usually a chelate group coordinates 8 of the 9 electrons of Gd, and one remaining coordination site
■
515
is used for rapidly exchanging water molecules that rapidly relax due to the fluctuating magnetic moment of the unpaired electron. It was suggested that the relaxivity of the complex could be modulated by blocking this site with a cleavable chemical group as originally proposed by Louie and colleagues (Louie et al., 2000). They designed a “smart” relaxation agent in which the access of water to the first coordination sphere was blocked with a -galactosidase substrate, galactopyranose, which can be removed by enzymatic cleavage. Following the cleavage of the blocking sugar group, the paramagnetic ion can interact directly with water protons to reduce their T1 by the inner sphere (contact) relaxation with a corresponding increase in the MR signal. Analogous approaches can be used to design CA with polypeptide blocking chains to probe protease activity. The major obstacle is intracellular expression of these enzymes therefore the contrast agent should be either microinjected to the cell(s) (Louie et al., 2000) or delivered to the cells by means of certain cellular transport systems such as amphiphilic Tat-type membrane translocation peptides (Bhorade et al., 2000). Another group of paramagnetic contrast agents uses manganese ion (Mn1), a biological analog of calcium, to visualize calcium influx and cell activation in vivo (Pautler et al., 1998) such as neuronal tract tracing (Koretsky et al., 2004; Wadghiri et al., 2004). To generate detectable T1 contrast in MR images (10% T1 change), the required concentration of Gd-based agents with T1 relaxivity of 苲10 [s · mM]1 should be in the range of 苲10 moles and above. Therefore, to detect molecular targets with submicromolar biological concentrations, multiple Gd groups should be attached to a single target molecule. One way to accomplish this is to use macromolecular carriers decorated with a large number of Gd chelate groups. Conjugation chemistry for this type of reaction is well established and typically a free amine group on the surface of the molecule is used to provide conjugation to derivatized Gd chelate complexes. Several imaging platforms were proposed as Gd carriers including proteins (albumin, avidin), dendrimers, poly-l-lysine, liposomes, and nanoemulsions (Artemov et al., 2002; Sipkins et al., 1998; Uzgiris et al., 2004; Winter et al., 2003). Generally, larger carriers provide proportionately larger T1 relaxivity per molecule (i.e., no saturation effects on T1 relaxivity is observed). However, there may be significant problems with the delivery of the large imaging probes to molecular targets that require extravasation of the molecule from blood capillaries and diffusion across tissue interstitium to access the binding site. Delivery issues will be discussed in more detail later in the chapter. Superparamagnetic MR Contrast Agents As was shown before, the large magnetic moment of the contrast agent induces very efficient relaxation, therefore superparamagnetic iron-oxide (SPIO)-based MR contrast agents were introduced, with an iron-oxide core protected by various polymeric coats. The magnetic core is typically composed from magnetite, that is an inverse spinel with formula Fe2+O · Fe23+O3 Fe3O4. Depending on the size, the core consists of several thousands of
516
CHAPTER 45
■
MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges
iron atoms and generates very high magnetic moment. Due to the relatively large molecular size of SPIO nanoparticles they have long rotational correlation time, c, and they act as efficient T1 contrast agents only at low magnetic fields where the condition 0 B0 c1 is fulfilled. On the other hand SPIOs generate strong low-frequency local magnetic fields that make them very sensitive T2 and T2* contrast agents with typical T2 relaxivity R2 in the order of 50–100 [s(mM Fe)]1. Because SPIO particles pack large amount of iron the resulting relaxivity of these nanoparticles is several orders of magnitude higher than for Gd compounds (per molecule of the contrast agent). Different preparations of SPIO were investigated for MRI applications including monocrystalline iron-oxide particles, MION, with a core diameter of 4.6 nm and nanoparticles diameter of about 20 nm, ultrasmall SPIO (USPIO) with 3–4 nm core and hydrodynamic diameter 25 nm, cross-linked iron-oxide (CLIO), classical SPIO such as Feridex (Berlex) with 5–6 nm polycrystalline core and dextran coating and 苲35 nm average diameter (Jung et al., 1995). Large SPIO particles with polystyrene coating and d 1 m are produced by Bangs, Inc. and are very useful for in vitro cell labeling. As with Gd-based contrast agent larger particles generally provide higher relaxivity. However, in vivo the delivery of this type of MR contrast agents may be not optimal for targeting tissue epitopes. Indeed, traditional use of SPIO relies on rapid internalization of the particles by macrophages in the liver, lymph nodes, and peripheral circulation. Biodistribution of CLIO particles in rats was studied by Moore and colleagues (Moore et al., 2000), and the results are summarized in Table 45.2, which is adapted from that paper. Nanoparticles have low distribution even in the tumors with presumably leaky vasculature, however, the high relaxivity of the agent enabled visualization of the brain tumor from normal brain (Moore et al., 2000). Delivery of monoclonal antibodies (mAb) and fluorescent nanospheres to Lewis lung carcinoma models in mice was studied by Nakahara et al. (Nakahara et al., 2006). Fluorescent micrographs of distribution of nonspecific fluorescent IgG and fluorescent 50-nm diameter nanospheres obtained 6 h after i.v. injection are shown in Figure 45.3. Extravasated IgG (arrows) had a patchy distribution with only vague association with CD31-stained blood vessels. On the other hand extravasated nanospheres were closely associated with focal regions of tumor vessels. Therefore contrast agents that can
specifically recognize tissue molecular markers and can efficiently diffuse to these markers should have a smaller molecular weight probably close to that of the mAb or about 150 kDa that corresponds to a molecular diameter less than 10 nm. While hyperpermeable tumors may have “hot spots” with increased extravasation of large macromolecules (Monsky et al., 1999), generally lower-molecular weight imaging markers are required for molecular MRI of tissue targets, unless the target is expressed by the vasculature. Interestingly, accumulation of endogenous iron by cells overexpressing iron-binding protein, ferritin can also lead to significant negative T2 contrast detectable in MR images. This method was proposed for molecular imaging of cancer cells transfected with ferritin, that can be used as a reporter gene for successful cell transfection in vivo (Cohen et al., 2005). CEST and PARACEST MR Agents This new class of MR contrast agent is designed to enable selective irradiation of exchangeable protons of these CA with an RF field to induce a drop in the MR signal of bulk water because of transfer of the saturated magnetization from the irradiated protons to water protons by chemical exchange. Originally proposed by the group of Balaban (Guivel-Scharen et al., 1998), CEST contrast was demonstrated for exchangeable amide protons of proteins (Zhou et al., 2003a, b), imino- and hydroxyl protons (Snoussi et al., 2003). A significant increase in the sensitivity of detection can be gained if a polymer with a large number of equivalent exchangeable groups is used. In experiments with polyuridine (2000 uridine units) the sensitivity gain of more than 5000 per imino-proton was demonstrated and the CA could be detected at concentration as low as 5 M (Snoussi et al., 2003). The efficiency of CEST contrast increases as the frequency offset, , between resonance frequency of the exchangeable group and water protons increases. The frequency difference can be significantly increased from several parts per million (ppm) to more than 20 ppm by attaching a paramagnetic shift reagent to the CEST agent. Experiments with a model poly-l-arginine/Tm(HDOTP)4 system lowered the detection limit to 2.8 M in a phantom at 7T magnetic field (Aime et al., 2003). This modification of the method was named PARACEST for paramagnetic enhanced CEST. The major potential advantages
TABLE 45.2 Biodistribution of long-circulating 125I-labeled CLIO nanoparticles in rats with 9L-GFP brain tumor (4 animals) 24 h post-injection (adapted from Moore et al. [2000]) Organ
Blood
Brain
Heart
Lymph nodes
Kidney
Liver
Lung
Muscle
Spleen
Tumor
Injected dose per gram of tissue (%)
0.5 0.2
0.01 0.01
0.2 0.1
25.0 3.2
0.8 0.1
1.9 0.3
0.2 0.1
0.1 0.1
9.8 0.8
0.1 0.1
Molecular Imaging Applications of MRI
of the CEST/PARACEST methods to generate MR contrast are: (i) the ability to “turn” the contrast “on” and “off ” by applying saturating RF field at the resonance frequency of the exchangeable group; (ii) high sensitivity that can be achieved by using polymer probes and/or PARACEST agents with large shifts and fast exchange rates; (iii) certain types of CEST agents such as poly-amino acids can be expressed in vivo by target cells with an appropriate reporter vector; (iv) it is possible to design PARACEST probes with different values and image them independently using RF saturation field with the appropriate frequency offset.
MOLECULAR IMAGING APPLICATIONS OF MRI MRI Cell Tracking A typical experiment for MR cell trafficking includes in vitro labeling of cell population with an iron-oxide nanoparticle preparation using standard transfection reagents such as poly-l-lysine (Bulte et al., 2004) and/or electroporation (Walczak et al., 2005). The magnetically labeled cells are administered systemically or via local injection, and T2- or T2*-weighted MRI is used for longitudinal imaging of the cell migration and accumulation to the target site. T2*-weighted GE MRI is sensitive to “long range” disturbances in the magnetic field produced by the concentrated SPIO particles that affect T2* relaxation of distant protons within an area much larger than the size of the nanoparticles. Therefore this technique provides significant amplification of the contrast and largest MR sensitivity (Bulte et al., 1992; Dodd et al., 1999). In several excellent reviews, Bulte et al. discussed in detail MR tracking of magnetically labeled cells (Bulte et al., 2004; Karmarkar et al., 2004; Lee et al., 2004;Walter et al., 2004). SPIO provides very efficient contrast enhancement in MRI, and isolated cells labeled with iron-oxide nanoparticles can be detected at iron concentrations as low as 17 ng per 106 cells or 8.5 104 MION particles/cell (Weissleder et al., 1997). MR detection of a single magnetically labeled T cell was demonstrated
(a)
IgG (iv) CD31
■
517
by Dodd et al. (Dodd et al., 1999). Typically for reliable in vivo MR detection, several million SPIO or USPIO particles must be loaded within the cell. Large micron diameter SPIO particles (such as Bangs) provide sufficient T2* contrast to enable detection of a single cell labeled with just a single micron size iron-oxide particle (Shapiro et al., 2004). One potential benefit of this approach is that the label is not diluted by subsequent cell division and daughter cells can be detected for a long time post-implantation. To further improve cell uptake properties CA nanoparticles can be linked to the HIV viral transport Tat peptide that can channel the cargo molecules across the cell plasma membrane (Gammon et al., 2003). Efficient cellular uptake of Tat-CLIO as well as Tat-Macrocyclic gadolinium chelates was recently demonstrated in in vitro and in vivo systems (Bhorade et al., 2000; Prantner et al., 2003). This approach is especially useful for loading of cells with low endocytotic/pinocytotic activity such as lymphocytes. Tat-CLIO nanoparticles were used to load cytotoxic T lymphocytes (CTL) ex vivo and to track their docking to the antigen-expressing tumor xenografts in vivo by T2-weighted MRI (Kircher et al., 2003). Magnetic labeling of transplanted islets with SPIO was used to noninvasively monitor transplantation efficiency and graft survival in small animal models (Evgenov et al., 2006). Briefly, isolated human islets were loaded with dextran-coated SPIO by overnight incubation and implanted in the kidney capsule or hepatically by intraportal infusion. In vivo MRI was performed with a B0 4.7T animal scanner using T2*-weighted SE pulse sequence. Results obtained at different time points after intrahepatic transplantation are shown in Figure 45.4.
MR Receptor Imaging with Targeted Contrast Agents Various mechanisms of contrast generations were explored for MRI of cell surface receptors. Several attempts to directly label mAb with Gd chelate groups did not produce satisfactory results in solid tumors primarily due to the limited number of groups that can be conjugated to the mAb without diminishing its binding affinity and limited extravasation and diffusion of the large molecular complexes in the tumor microenvironment. On
(b)
Microspheres (iv) CD31
Figure 45.3 Confocal micrographs showing the distribution of extravasated nonspecific IgG and 50-nm microspheres at 6 h after injection in RIP-Tag2 tumors (adapted from Nakahara et al. [2006]).
518
CHAPTER 45
■
MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges
Day 3
Day 5
Day 7
SC A S
V P
Figure 45.4 In vivo MRI obtained at different time points after intrahepatic transplantation of labeled human islets. A representative slice (0.5 mm) shows islets scattered throughout the liver. Pancreatic islets appear as hypointense spots on T2*-weighted images (red ovals). A, aorta; P, portal vein; S, stomach; SC, spinal cord; V, caudal vena cava (adapted from Evgenov et al. [2006]).
the other hand, several reports demonstrated successful imaging of v3 integrins that are marker of the tumor neovasculature and are expressed in the angiogenic endothelium of the tumors. High accessibility of this marker enables the use of large blood-pool nanocomplexes such as paramagnetic polymerized liposomes (Sipkins et al., 1998) or Gd-perfluorocarbon nanoparticles (Anderson et al., 2000) that can contain several thousands Gd ions per particle. For MRI of tissue targets, macromolecular carriers such as protein or dendrimer Gd conjugates can be administered independently from the targeting antibodies following a multistep labeling approach (Paganelli et al., 1999). This approach should provide more efficient delivery of the components in comparison to a single large molecular weight targeted imaging agent. This protocol was used for in vivo imaging of HER-2/ neu receptors in a preclinical model of breast cancer using biotinylated primary anti-HER-2/neu mAb and avidin (GdDTPA) conjugate (Artemov et al., 2003). The combination of high receptor expression levels, relatively small molecular sizes of the components, and multiplication effect of multiple Gd ions attached to a single avidin molecule and several avidin-binding biotins per mAb resulted in detectable positive T1 MR contrast in HER-2/neu positive tumors (Artemov et al., 2003). Histological sections of HER-2/neu expressing BT-474 tumor model (Figure 45.5) demonstrated efficient and uniform labeling of the breast cancer cells with fluorescent-avidin (MW 60 kDa) and biotinylated-Herceptin 24 h after systemic administration of this antibody with molecular weight of 150 kDa. It is likely that the use of multistep avidin–biotin labeling system in addition to the specific recognition of biotinylated mAb by avidin-based contrast agent also induces crosslinking that results in a rapid internalization of the HER-2/neu receptors with the attached probe (Zhu et al., 2006). It is also possible that the higher sensitivity of this approach resulted from efficient loading of cells with the internalized contrast agent. Superparamagnetic iron-oxide nanoparticles were also explored as sensitive MR probes for targeted imaging of tumor cell surface receptors. Two alternative strategies for the use of SPIO in molecular MRI were proposed. The first approach
utilizes specific cellular uptake of the SPIO via plasma membrane specific transporters. Iron transport protein, transferrin, conjugated with MION iron-oxide nanoparticles was used to image 9L rat glioma cancer cell overexpressing engineered transferring receptor (ETR) (Weissleder et al., 2000). The ETR cancer cells internalized up to 8 106 of the TF-targeted CA within an hour and MION loaded cancer cells generated strong negative T2* contrast in vivo in gradient-echo MR images obtained at 1.5T 24 h after intravenous injection of 3 mg of TfMION to a nude mouse. No differential contrast was detected in the T1-weighted image, and strong negative contrast was detected in the T2*-weighted image of the ETR tumor. The second approach relies on the use of mAb (or other targeting molecules) conjugated SPIO nanoparticles. In a study by Weissleder et al., human polyclonal IgG were used to target MION to sites of inflammation (Weissleder et al., 1991). A similar approach was used for molecular MRI of apoptosis in EL4 solid tumor models exposed to a chemotherapy; there SPIO particles were conjugated to the C2 domain of the protein synaptotagmin that binds with high affinity to phosphatidylserine residues that translocate to the outer leaflet of the plasma membrane in apoptotic cells (Zhao et al., 2001). Fluorescent CLIO particles conjugated with the targeting peptide were used to image underglycosylated MUC-1 tumor antigen (Moore et al., 2004). Combined T2 MRI and near-infrared fluorescence imaging demonstrated a specific accumulation of the contrast in the tumor that expressed the antigen. One of the potential problems of using SPIO nanoparticles for detection of cellular epitopes is their relatively poor delivery properties. The situation can be more favorable in tumors where hyperpermeable vasculature in combination with long circulation time of the agent results in an efficient extravasation and contrast enhancement. However, the question still remains regarding uniform and reliable delivery of nanosize contrast agents to the specific receptors expressed in tissues other than endothelial lining of blood vessels. Accumulation of endogenous iron in ferritin-overexpressing cells can also be used as a sensitive MR marker as was demonstrated by Cohen et al. (Cohen et al., 2005). C6 cells stably
Molecular Imaging Applications of MRI
■
519
R2
TET
TET
Figure 45.5 Histological frozen section of BT-474 tumor xenograft in SCID mouse treated with biotin-Herceptin and avidin-red 24 h after the mAb. Cell nuclei are stained blue with Hoeschst-3334.
expressing a TET-EGFP-HA-ferritin construct enabled the dynamic detection of TET-regulated gene expression by MRI, also validated by fluorescence microscopy and histology. Induction of the gene significantly elevated both in vitro and in vivo MR relaxation rates that were consistent with induced expression of ferritin and increased iron uptake. Figure 45.6 demonstrates the effect of TET-induction on R2 relaxation rate in the transfected tumor. CEST and PARACEST MRI CEST and PARACEST MR contrast mechanisms can be used for the development of a new generation of highly sensitive MRI agents (Woods et al., 2006). Specific CEST agents for MRI of glucose and lactate were also proposed (Aime et al., 2002; Zhang et al., 2001). For molecular MRI two interesting applications of CEST MRI use endogenously expressed markers or transgenic constructs expressed in the target cells. CEST MRI of exchangeable amide protons of endogenous proteins and peptides was used to detect amide proton transfer (APT) in rat brain tumors in vivo (Zhou et al., 2003a, b). Highly efficient CEST agent poly-l-lysine (25 lysine amino acids) was expressed in 9L rat glioma tumors, and MRI detection of the CEST effect in the tumor expressing this reporter gene was demonstrated in vivo (Gilad et al., 2006). MR Spectroscopy MR spectroscopy (MRS) provides information about the chemical environment of the nuclear spin such as chemical bonds, neighboring nuclei, and chemical structure. The extension of MRS to imaging has resulted in the development of spectroscopic MRI [Chemical Shift Imaging (CSI), Brown et al. (1982)] where chemical information is spatially encoded and selective images of
15
s1
25
Figure 45.6 In vivo MRI detection of switchable ferritin expression in C6 tumors. Scalebar 2.5 mm (adapted from Cohen et al. [2005]).
distribution of specific chemical compounds such as metabolites, exogenous substances, drugs, etc. can be obtained. Representative applications of CSI that can provide molecular information with MRI are discussed here. In comparison to contrast-enhanced MRI, where micromolar concentrations of CA can be detected, the low sensitivity of MRS limits the detectable concentration of CSI agents at millimolar range for proton and 19F MRS; even higher concentrations are required for less-sensitive nuclei such as 31P, 23Na, and 13C. Proton CSI of choline is an important tool for monitoring concentrations of this important metabolite that is a marker of malignancy in breast, prostate, brain, and possibly other tumors (Katz-Brull et al., 2002; Kurhanewicz et al., 2002; McKnight, 2004). Fluorine-19 is an attractive nucleus for CSI because of the high sensitivity provided by large magnetic moment of the spin, 100% natural abundance of the isotope, and the absence of any 19F background signals in biological systems in vivo. The feasibility of 19F MRI to track dendritic cells prelabeled with perfluoropolyether particles has been explored (Ahrens et al., 2005; Bulte, 2005). Hyperpolarization of the spins is a novel technique that produces a dramatic increase in MR signal (Kauczor et al., 1998). However, the lifetime of hyperpolarized state is limited by T1 relaxation time of the nucleus (usually in the range of few seconds) and typically first-pass MRI is performed with hyperpolarized 3He or 13C-labeled substrates (Golman et al., 2003, 2006).
520
CHAPTER 45
TABLE 45.3
MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges
■
Application of molecular MRI/MRS for biomedical research in genomic medicine
Method
Targets
Feasibility for preclinical studies with animal models
Clinical translation potential
Magnetic labeling of exogenous cells in vitro
Somatic/stem cells used for therapeutic transplantation
High, can be used for preclinical testing of novel therapeutic paradigms
Possible applications for stem cell therapy and islet transplantation using magnetically labeled cells
Specific MR contrast agents targeted to cell surface receptors
Cancer or endothelial cells expressing unique endogenous or engineered (reporter) cell surface receptors
High
Low due to potential toxicity of the contrast agent delivered at high dose
Endogenous CEST reporter genes
Cells engineered to express CEST-generating molecules under control of an appropriate promoter
High, can be used as sensitive reporter probes for noninvasive MR imaging
Possible application in transplanted cells as viability/differentiation reporters
MR spectroscopy
Endogenous and exogenous compounds. Later can be used with enriched magnetic isotope concentration (13C) and/or with hyperpolarization
High, used for metabolic studies in animal models as well as in reporter systems
Multiple applications are possible including 31P spectroscopy, 19F MRS of fluorine-containing drugs, 13C MRS of hyperpolarized compounds
Fluorine-19 MRS and CSI were proposed to detect activation of -galactosidase enzyme with a prototype reporter substrate, 4-fluoro-2-nitrophenyl-b-d-galactopyranoside. Enzymatic cleavage of the sugar changes the chemical shift of 19F nucleus on the nitrophenyl ring by 4–8 ppm depending on the pH of the environment. Phosphorus-31 MRS was used for the detection of creatine kinase activity as a model reporter enzyme. The enzyme was expressed in the mouse liver by adenoviral in vivo transduction and conversion of creatine to phosphocreatine was only detected in the liver of mice injected with the creatine kinase viral vector (Auricchio et al., 2001). The low sensitivity of 31P MRS translates to low spatial resolution of 31P CSI of the order of 10 cm3, which is a potential problem for practical applications of this technique. MRI and MRS applications to molecular imaging that are relevant for problems and challenges of genomic medicine are summarized in Table 45.3.
CONCLUSIONS In comparison to nuclear imaging MR, molecular imaging is still in its infancy. The main advantages of MRI are high spatial resolution, the ability to provide functional information such as in perfusion dynamic MR studies with clinically approved small molecular weight contrast agents, and chemical information that can be detected with MRS and/or spectroscopic imaging. While MRI provides excellent morphological resolution, soft tissue contrast, and functional information, its main problem is a relatively low sensitivity that requires some form of signal amplification to detect sparse molecular targets that are often expressed
in micromolar concentrations. We have reviewed several experimental approaches that may provide a framework for the development of novel sophisticated MRI technique that can bring receptor imaging into the realm of radiological testing. One of the important strategies is the development of novel targeted MR paramagnetic and superparamagnetic contrast agents with high specific relaxivity and favorable in vivo pharmacokinetics. These can be used for in vivo imaging of molecular targets and for in vitro labeling of cells for MRI cell tracking studies. Optimization of their relaxation properties, delivery efficiency, in vivo stability, and targeting mechanisms are required to facilitate efficient MRI of molecular targets. Specifically targeting strategies are not discussed here since an excellent comparison of intact mAb and their derivatives such as minibody, diabody, and single-chain fragments can be found in a recent review article (Wu et al., 2005). The development of a waterexchange type MR contrast mechanism (CEST) is also important as these imaging markers provide strong signal amplification and can be expressed endogenously by transfecting the target cell with an appropriate transfection vector. These may be feasible to monitor the efficiency of gene delivery to the cell for gene therapy. Imaging of molecular epitopes expressed on the surface of the target cell would require the use of contrast agents that can efficiently extravasate in the target organ and diffuse through the interstitium to their binding sites. To enable sufficient delivery, these agents should be of a relatively low-molecular weight and still have high MR relaxivity that translates to the high detection sensitivity in micromolar range of concentrations. Multicomponent contrast agents are an interesting alternative as the different subunits can be optimized for efficient
References
delivery, affinity to the target molecule, and high signal amplification. Biotin/streptavidin (or avidin) systems used to link different subunits have been tested extensively for nuclear imaging and therapy. For MRI applications, the required concentrations of the imaging agents are significantly higher in comparison to PET and SPECT, imaging and therefore potential issues of toxicity and immunogenecity have to be addressed accordingly. Clinical applications of molecular MRI are currently limited to in vivo tracking of prelabeled progenitor cells (Bulte et al., 2002). Another relatively straightforward application of molecular MRI is the imaging of vascular targets. These molecules are expressed in the lumen of blood vessels and therefore are accessible to large targeted MR contrast agents that can provide sufficient sensitivity of detection. The development of novel agents for imaging of solid tumor markers on the other hand requires extensive research. Major problems include accumulation of sufficient amount of the agent at the target site to generate MR
■
521
detectable imaging contrast, efficient delivery of macromolecular contrast agents to the target site, and associated toxicity and immune response of the host. The lower magnetic field of clinical MR scanners in comparison to small animal MRI research systems results in the reduced sensitivity of detection and amplifies these problems. Therefore several technical, biological, and methodological issues have to be resolved before moleculartargeted MRI can become a clinically feasible, routinely used, imaging technology.
ACKNOWLEDGEMENTS This publication was supported in part by NIH/NCI grants P50 CA103175 and RO1 CA97310. The author thanks Dr Z.M. Bhujwalla for helpful discussion and critically reading the manuscript.
REFERENCES Ahrens, E.T., Flores, R., Xu, H. and Morel, P.A. (2005). In vivo imaging platform for tracking immunotherapeutic cells. Nat Biotechnol 23, 983–987. Aime, S., Delli Castelli, D., Fedeli, F. and Terreno, E. (2002). A paramagnetic MRI-CEST agent responsive to lactate concentration. J Am Chem Soc 124, 9364–9365. Aime, S., Delli Castelli, D. and Terreno, E. (2003). Supramolecular adducts between poly-L-arginine and [TmIIIdotp]: A route to sensitivity-enhanced magnetic resonance imaging-chemical exchange saturation transfer agents. Angew Chem Int Ed Engl 42, 4527–4529. Anderson, S.A., Rader, R.K., Westlin, W.F., Null, C., Jackson, D., Lanza, G.M., Wickline, S.A. and Kotyk, J.J. (2000). Magnetic resonance contrast enhancement of neovasculature with alpha(v)beta(3)targeted nanoparticles. Magn Reson Med 44, 433–439. Artemov, D., Mori, N., Ravi, R. and Bhujwalla, Z. (2002). MR molecular imaging of Her-2/Neu receptor with Gd based targeted contrast agent. International Society for Magnetic Resonance in Medicine, Honolulu. Artemov, D., Mori, N., Ravi, R. and Bhujwalla, Z.M. (2003). Magnetic resonance molecular imaging of the Her-2/neu receptor. Cancer Res 63, 2723–2727. Auricchio, A., Zhou, R., Wilson, J.M. and Glickson, J.D. (2001). In vivo detection of gene expression in liver by 31P nuclear magnetic resonance spectroscopy employing creatine kinase as a marker gene. Proc Natl Acad Sci USA 98, 5205–5210. Bhorade, R., Weissleder, R., Nakakoshi, T., Moore, A. and Tung, C.H. (2000). Macrocyclic chelators with paramagnetic cations are internalized into mammalian cells via a HIV-tat derived membrane translocation peptide. Bioconjug Chem 11, 301–305. Brown, M.A. (2003). MRI: Basic principles and applications. Wiley-Liss, Hoboken, NJ. Brown, T.R., Kincaid, B.M. and Ugurbil, K. (1982). NMR chemical shift imaging in three dimensions. Proc Natl Acad Sci U S A 79, 3523–3526.
Bulte, J.W. (2005). Hot spot MRI emerges from the background. Nat Biotechnol 23, 945–946. Bulte, J.W., Hoekstra, Y., Kamman, R.L., Magin, R.L., Webb, A.G., Briggs, R.W., Go, K.G., Hulstaert, C.E., Miltenyi, S. and The, T.H. (1992). Specific MR imaging of human lymphocytes by monoclonal antibody-guided dextran-magnetite particles. Magn Reson Med 25, 148–157. Bulte, J.W., Duncan, I.D. and Frank, J.A. (2002). In vivo magnetic resonance tracking of magnetically labeled cells after transplantation. J Cereb Blood Flow Metab 22, 899–907. Bulte, J.W., Arbab, A.S., Douglas, T. and Frank, J.A. (2004). Preparation of magnetically labeled cells for cell tracking by magnetic resonance imaging. Meth Enzymol 386, 275–299. Cohen, B., Dafni, H., Meir, G., Harmelin, A. and Neeman, M. (2005). Ferritin as an endogenous MRI reporter for noninvasive imaging of gene expression in C6 glioma tumors. Neoplasia 7, 109–117. Corot, C., Idee, J.M., Hentsch, A.M., Santus, R., Mallet, C., Goulas, V., Bonnemain, B. and Meyer, D. (1998). Structure–activity relationship of macrocyclic and linear gadolinium chelates: investigation of transmetallation effect on the zinc-dependent metallopeptidase angiotensin-converting enzyme. J Magn Reson Imaging 8, 695–702. Cunningham, C.H., Arai, T., Yang, P.C., McConnell, M.V., Pauly, J.M. and Conolly, S.M. (2005). Positive contrast magnetic resonance imaging of cells labeled with magnetic nanoparticles. Magn Reson Med 53, 999–1005. Dodd, S.J., Williams, M., Suhan, J.P., Williams, D.S., Koretsky, A.P. and Ho, C. (1999). Detection of single mammalian cells by highresolution magnetic resonance imaging. Biophys J 76, 103–109. Duyn, J.H., van Gelderen, P., Talagala, L., Koretsky, A. and de Zwart, J. A. (2005). Technological advances in MRI measurement of brain perfusion. J Magn Reson Imaging 22, 751–753. Evgenov, N.V., Medarova, Z., Dai, G., Bonner-Weir, S. and Moore, A. (2006). In vivo imaging of islet transplantation. Nat Med 12, 144–148.
522
CHAPTER 45
■
MRI for Molecular Imaging Applications: Overview, Perspectives, and Challenges
Gammon, S.T., Villalobos, V.M., Prior, J.L., Sharma, V. and PiwnicaWorms, D. (2003). Quantitative analysis of permeation peptide complexes labeled with Technetium-99m: Chiral and sequencespecific effects on net cell uptake. Bioconjug Chem 14, 368–376. Gilad, A.A., McMahon, M.T., Walczak, P., Winnard, P.T.J., Raman, V., Bulte, J.W.M. and van Zijl, P.C.M. (2006). Artificial Reporter Gene Providing MRI Contrast In Vivo Based on Chemical Exchange. International Society for Magnetic Resonance in Medicine, Seattle, WA. Gillies, R.J. (1994). NMR in Physiology and Biomedicine. Academic Press, San Diego. Golman, K., Olsson, L.E., Axelsson, O., Mansson, S., Karlsson, M. and Petersson, J.S. (2003). Molecular imaging using hyperpolarized 13C. Br J Radiol 76(Spec No 2), S118–S127. Golman, K., Zandt, R.I., Lerche, M., Pehrson, R. and ArdenkjaerLarsen, J.H. (2006). Metabolic imaging by hyperpolarized 13C magnetic resonance imaging for in vivo tumor diagnosis. Cancer Res 66(22), 10855–10860. Guivel-Scharen, V., Sinnwell, T., Wolff , S.D. and Balaban, R.S. (1998). Detection of proton chemical exchange between metabolites and water in biological tissues. J Magn Reson 133, 36–45. Guttmacher, A.E. and Collins, F.S. (2002). Genomic medicine – Aprimer. New Engl J Med 19, 1512–1520. Jung, C.W. and Jacobs, P. (1995). Physical and chemical properties of superparamagnetic iron oxide MR contrast agents: Ferumoxides, ferumoxtran, ferumoxsil. Magn Reson Imaging 13, 661–674. Karmarkar, P.V., Kraitchman, D.L., Izbudak, I., Hofmann, L.V., Amado, L.C., Fritzges, D.,Young, R., Pittenger, M., Bulte, J.W. and Atalar, E. (2004). MR-trackable intramyocardial injection catheter. Magn Reson Med 51, 1163–1172. Katz-Brull, R., Lavin, P.T. and Lenkinski, R.E. (2002). Clinical utility of proton magnetic resonance spectroscopy in characterizing breast lesions. J Natl Cancer Inst 94, 1197–1203. Kauczor, H., Surkau, R. and Roberts, T. (1998). MRI using hyperpolarized noble gases. Eur Radiol 8, 820–827. Kircher, M.F., Mahmood, U., King, R.S., Weissleder, R. and Josephson, L. (2003). A multimodal nanoparticle for preoperative magnetic resonance imaging and intraoperative optical brain tumor delineation. Cancer Res 63, 8122–8125. Koretsky, A.P. and Silva, A.C. (2004). Manganese-enhanced magnetic resonance imaging (MEMRI). NMR Biomed 17, 527–531. Kurhanewicz, J., Swanson, M.G., Nelson, S.J. and Vigneron, D.B. (2002). Combined magnetic resonance imaging and spectroscopic imaging approach to molecular imaging of prostate cancer. J Magn Reson Imaging 16, 451–463. Le Bihan, D., Turner, R., Moonen, C.T. and Pekar, J. (1991). Imaging of diffusion and microcirculation with gradient sensitization: Design, strategy, and significance. J Magn Reson Imaging 1, 7–28. Lee, I.H., Bulte, J.W., Schweinhardt, P., Douglas, T., Trifunovski, A., Hofstetter, C., Olson, L. and Spenger, C. (2004). In vivo magnetic resonance tracking of olfactory ensheathing glia grafted into the rat spinal cord. Exp Neurol 187, 509–516. Louie, A.Y., Huber, M.M., Ahrens, E.T., Rothbacher, U., Moats, R., Jacobs, R.E., Fraser, S.E. and Meade, T.J. (2000). In vivo visualization of gene expression using magnetic resonance imaging. Nat Biotechnol 18, 321–325. McKnight, T.R. (2004). Proton magnetic resonance spectroscopic evaluation of brain tumor metabolism. Semin Oncol 31, 605–617. McMahon, M.T., Gilad, A.A., Zhou, J., Sun, P.Z., Bulte, J.W. and van Zijl, P.C. (2006). Quantifying exchange rates in chemical exchange
saturation transfer agents using the saturation time and saturation power dependencies of the magnetization transfer effect on the magnetic resonance imaging signal (QUEST and QUESP): Ph calibration for poly-l-lysine and a starburst dendrimer. Magn Reson Med 55, 836–847. Monsky,W.L., Fukumura, D., Gohongi,T., Ancukiewcz, M.,Weich, H.A., Torchilin, V.P., Yuan, F. and Jain, R.K. (1999). Augmentation of transvascular transport of macromolecules and nanoparticles in tumors using vascular endothelial growth factor. Cancer Res 59, 4129–4135. Moore, A., Marecos, E., Bogdanov, A., Jr. and Weissleder, R. (2000). Tumoral distribution of long-circulating dextran-coated iron oxide nanoparticles in a rodent model. Radiology 214, 568–574. Moore, A., Medarova, Z., Potthast, A. and Dai, G. (2004). In vivo targeting of underglycosylated MUC-1 tumor antigen using a multimodal imaging probe. Cancer Res 64, 1821–1827. Mori, S. and van Zijl, P.C. (2002). Fiber tracking: principles and strategies – a technical review. NMR Biomed 15, 468–480. Nakahara, T., Norberg, S.M., Shalinsky, D.R., Hu-Lowe, D.D. and McDonald, D.M. (2006). Effect of inhibition of vascular endothelial growth factor signaling on distribution of extravasated antibodies in tumors. Cancer Res 66, 1434–1445. Paganelli, G., Grana, C., Chinol, M., Cremonesi, M., De Cicco, C., De Braud, F., Robertson, C., Zurrida, S., Casadio, C., Zoboli, S. et al. (1999). Antibody-guided three-step therapy for high grade glioma with yttrium-90 biotin. Eur J Nucl Med 26, 348–357. Pautler, R.G., Silva, A.C. and Koretsky, A.P. (1998). In vivo neuronal tract tracing using manganese-enhanced magnetic resonance imaging. Magn Reson Med 40, 740–748. Prantner, A.M., Sharma,V., Garbow, J.R. and Piwnica-Worms, D. (2003). Synthesis and characterization of a Gd-DOTA-D-permeation peptide for magnetic resonance relaxation enhancement of intracellular targets. Mol Imaging 2, 333–341. Rome, C., Couillaud, F. and Moonen, C.T. (2007). Gene expression and gene therapy imaging. Eur Radiol 17, 305–319. Seppenwoolde, J.H., Viergever, M.A. and Bakker, C.J. (2003). Passive tracking exploiting local signal conservation: The white marker phenomenon. Magn Reson Med 50, 784–790. Shapiro, E.M., Skrtic, S., Sharer, K., Hill, J.M., Dunbar, C.E. and Koretsky, A.P. (2004). MRI detection of single particles for cellular imaging. Proc Natl Acad Sci USA 101, 10901–10906. Sipkins, D.A., Cheresh, D.A., Kazemi, M.R., Nevin, L.M., Bednarski, M.D. and Li, K.C. (1998). Detection of tumor angiogenesis in vivo by alphaVbeta3-targeted magnetic resonance imaging. Nat Med 4, 623–626. Slichter, C.P. (1990). Principles of Magnetic Resonance. Springer-Verlag, Berlin. Snoussi, K., Bulte, J.W., Gueron, M. and Van Zijl, P.C. (2003). Sensitive CEST agents based on nucleic acid imino proton exchange: Detection of poly(rU) and of a dendrimer-poly(rU) model for nucleic acid delivery and pharmacology. Magn Reson Med 49, 998–1005. Uzgiris, E.E., Cline, H., Moasser, B., Grimmond, B., Amaratunga, M., Smith, J.F. and Goddard, G. (2004). Conformation and structure of polymeric contrast agents for medical imaging. Biomacromolecules 5, 54–61. Wadghiri, Y.Z., Blind, J.A., Duan, X., Moreno, C., Yu, X., Joyner, A.L. and Turnbull, D.H. (2004). Manganese-enhanced magnetic resonance imaging (MEMRI) of mouse brain development. NMR Biomed 17, 613–619.
References
Walczak, P., Kedziorek, D.A., Gilad, A.A., Lin, S. and Bulte, J.W. (2005). Instant MR labeling of stem cells using magnetoelectroporation. Magn Reson Med 54, 769–774. Walter, G.A., Cahill, K.S., Huard, J., Feng, H., Douglas, T., Sweeney, H.L. and Bulte, J.W. (2004). Noninvasive monitoring of stem cell transfer for muscle disorders. Magn Reson Med 51, 273–277. Weissleder, R., Lee, A.S., Fischman, A.J., Reimer, P., Shen, T., Wilkinson, R., Callahan, R.J. and Brady, T.J. (1991). Polyclonal human immunoglobulin G labeled with polymeric iron oxide: Antibody MR imaging. Radiology 181, 245–249. Weissleder, R., Cheng, H.C., Bogdanova, A. and Bogdanov, A. (1997). Magnetically labeled cells can be detected by MR imaging. J Magn Reson Imaging 7, 258–263. Weissleder, R., Moore, A., Mahmood, U., Bhorade, R., Benveniste, H., Chiocca, E.A. and Basilion, J.P. (2000). In vivo magnetic resonance imaging of transgene expression. Nat Med 6, 351–355. Winter, P.M., Caruthers, S.D., Kassner, A., Harris, T.D., Chinen, L.K., Allen, J.S., Lacy, E.K., Zhang, H., Robertson, J.D., Wickline, S.A. et al. (2003). Molecular imaging of angiogenesis in nascent Vx-2 rabbit tumors using a novel alpha(nu)beta3-targeted nanoparticle and 1.5 tesla magnetic resonance imaging. Cancer Res 63, 5838–5843.
■
523
Woods, M., Woessner, D.E. and Sherry, A.D. (2006). Paramagnetic lanthanide complexes as PARACEST agents for medical imaging. Chem Soc Rev 35, 500–511. Wu, A.M. and Senter, P.D. (2005). Arming antibodies: prospects and challenges for immunoconjugates. Nat Biotechnol 23, 1137–1146. Zhang, S., Winter, P., Wu, K. and Sherry, A.D. (2001). A novel europium(III)-based MRI contrast agent. J Am Chem Soc 123, 1517–1518. Zhao, M., Beauregard, D.A., Loizou, L., Davletov, B. and Brindle, K.M. (2001). Non-invasive detection of apoptosis using magnetic resonance imaging and a targeted contrast agent. Nat Med 7, 1241–1244. Zhou, J., Lal, B., Wilson, D.A., Laterra, J. and van Zijl, P.C. (2003a). Amide proton transfer (APT) contrast for imaging of brain tumors. Magn Reson Med 50, 1120–1126. Zhou, J., Payen, J.F., Wilson, D.A., Traystman, R.J. and van Zijl, P.C. (2003b). Using the amide proton signals of intracellular proteins and peptides to detect pH effects in MRI. Nat Med 9, 1085–1090. Zhu, W., Okollie, B., Bhujwalla, Z.M. and Artemov, D. (2006). Controlled Internalization and Recycling of Her-2/neu by Cross-Linking with an Avidin/Streptavidin–Biotin System for MR Enhancement. International Society for Magnetic Resonance in Medicine, Seattle, WA.
CHAPTER
46 Fluorescence Imaging: Overview and Applications in Biomedical Research Vasilis Ntziachristos
INTRODUCTION Fluorescence methods have widely impacted biotechnology and genomics. From decoding the genome of various species to flow cytometry, high-throughput screening and uses in microscopy, fluorescence can easily qualify as one of the most common modalities in the biomedical laboratory (Giepmans et al., 2006; Herschman, 2003; Tsien, 2005; Weissleder and Ntziachristos, 2003). The high diversity in utilizing fluorescence molecules comes with three key advantages that make fluorescence an attractive tool. First, there is the high contrast that can be imparted with fluorescence approaches. A single molecule, a membrane receptor or several other cellular structures and organelles could go undetected under a camera or a microscope due to resolution and contrast limitations, but they can be identified with high sensitivity when they are tagged with a fluorescent agent. Second, fluorescent dyes come in many colors to simultaneously differentiate several parameters at moderate instrumentation complexity and cost. Finally, the use of non radio-decaying isotopes allows measurements over extended periods of time and enables longitudinal studies offering user-friendliness associated with the use of safe, nonionizing radiation. These fundamental principles that make fluorescence a powerful technique for in vitro assays also make it an attractive modality for in vivo intravital applications (Jain et al., 2002). Fluorescence Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 524
imaging of tissues, however, comes with two major complications. The first is that in vivo tissue staining necessitates elaborate methods for imparting high contrast and differentiate between the targeted moiety and nonspecific background fluorochrome bio-distribution that may be present (Achilefu et al., 2005; Tung, 2004; Tyagi et al., 2000). The second is that in contrast to clear solutions or ultra-thin (5–10 μm) tissue slices, intact tissue scatters light significantly and makes the formation of images challenging (Ntziachristos et al., 2005). Fluorescence imaging of tissues therefore refers to a group of methods that operate under high light-scattering conditions using appropriate in vivo staining methods and aims at resolving or sensing cellular and subcellular function from depths ranging from 10 μm to several centimeters. In this role, in vivo fluorescence imaging plays an essential role toward the study of genomes and proteomes in unperturbed environments in real time. In particular, fluorescence imaging can be pivotal in asserting dynamic interactions in the pathophysiology of multifactorial disease and enables monitoring of disease evolution and response to external factors on the same host organism over time (Brown et al., 2001; Gross and Piwnica-Worms, 2005). This, in contrast to the mainstream of laboratory research, pieces together the time evolution of events by in vitro analysis of multiple samples obtained at different time-points, which may offer inaccurate observations and time-inefficient procedures. Copyright © 2009, Elsevier Inc. All rights reserved.
Imaging Technology
Ultimately the same technology can be used for reading disease susceptibility and response from individuals, to study gene– environment interactions and overall lead personalized treatment and health care. Essential tools for achieving these goals include in vivo tissue staining techniques via reporter technology used to report on a specific molecular parameter or function (Shaner et al., 2004). A widely used example is the use of fluorescent proteins as endogenous markers to report on virtually any genomic process in vivo (Giepmans et al., 2006). In parallel, the advent of fluorescence agents, referred to as fluorescent probes (Weissleder and Ntziachristos, 2003), which attain specificity to particular biomarkers, further increase the application potential for fluorescence imaging to be applied beyond the biology bench to preclinical and clinical settings.
IMAGING TECHNOLOGY The following section summarizes the major fluorescence methodologies used for in vivo imaging. Penetration depth is an important distinguishing parameter that can be used to classify different fluorescence in vivo imaging technologies, in particular confocal and multiphoton microscopy, planar fluorescence macroscopic imaging and tomographic macroscopic imaging. These basic classes of fluorescence imaging and representative applications that exemplify its use are summarized in the following sections. Intravital Microscopy Microscopy has seen great advances in the recent years and new technologies continue to emerge at high rates. In vivo tissue microscopy employs methods that account for tissue scattering and
(a)
Confocal illumination path
(b)
■
525
produce high-resolution (diffraction-limited) images at depths of the order of a few hundred microns. A widely used approach is confocal microscopy. A schematic explaining the principle of operation is shown in Figure 46.1. Light is focused at an intended focal point inside the tissue. Upon incidence on the tissue surface however, this beam will illuminate an extended volume as shown in Figure 46.1a. The exact volume covered depends on the beam shape, the objective’s numerical aperture and the tissue’s optical characteristics. By consequence, all fluorochromes present in this volume are excited and emit fluorescence. If the entire light signal that reaches the microscope’s objective was to be collected, the image would appear blurred. This is because the light collected represents fluorescence activity from the entire volume illuminated and many of these photons have been scattered multiple times by cellular organelles so that they do not retain high-resolution information as to the place of origin. To improve the resolution, confocal microscopy uses a pinhole placed in front of the detector to reject the light emitted or scattered by depths other than the intended depth of the focal point inside the tissue, that is, the “confocal” depth. This is better explained in Figure 46.1b. At the confocal depth, the area excited has the narrower cross-section and therefore only fluorescence from a very confined area reaches the photon detector, most other fluorescence light hits the edges of the pinhole and it is not detected. Therefore resolution is retained at the expense of rejecting a significant part of the light emitted. To account for the low photon yield, light of high intensity is typically used, which however may lead to fluorochrome photobleaching or phototoxicity effects. Regardless of the experimental limitations however, confocal microscopy is highly utilized to reject the effects of tissue scattering and produce high-resolution fluorescence images.
Confocal detection path
(c)
Two-photon illumination path
z Excitation area
In focus Out of focus
Excitation area
Figure 46.1 Principle of operation for achieving depth (z-axis) resolution in (a and b) confocal and (c) two-photon (or multiphoton) microscopy. Spatial in-plane resolution, that is, the x–y resolution, is achieved by mechanically moving (scanning) the photon beam in relation to the sample imaged.
526
CHAPTER 46
■
Fluorescence Imaging: Overview and Applications in Biomedical Research
While confocal microscopy has been used to imaging fixed tissues (see Figure 46.2), it is increasingly employed for in vivo imaging at tissue depths of up to 200 μm also depending on the optical properties of the tissue investigated. To image tissues other than superficial epithelial structures, several invasive approaches have been developed such as the use of implantable skin window chambers (Jain et al., 2002) or invasive imaging by surgically exposing tissues for gaining access to different tissue types. Correspondingly, intravital imaging has been applied to many applications, for example, in tumor biology and the study of angiogenesis (Jain et al., 2002; Sipkins et al. 2005), pharmacological studies (Norman, 2005) or the microdistribution of fluorescent probes (Bogdanov et al., 2002). Most confocal microscopes employ lens-based noncontact imaging as shown in Figure 46.1. Fiber-based delivery and collection has been also considered, typically by direct contact of an optical fiber probe, consisting of several thousands of fibers tightly packed together, onto tissue. This approach allows access to an increased number of deep-seated organs and has been used endoscopically or through minimally invasive procedures (D’Hallewin et al., 2005). An imaging example obtained in vivo using a flexible endoscopic confocal microscope is shown in Figure 46.3. Another popular technique for rejecting tissue scattering is two- and multi-photon microscopy (Denk et al., 1990). In contrast to confocal detection, no pinhole is used here; instead the method selectively excites fluorochromes within a very small tissue volume at a predetermined location as shown on Figure 46.1c. Therefore, although multiple scattered photons are collected, high resolution is granted by the size of the volume excited, allowing for high sectioning capability. To achieve small volume excitation, the technique makes use of two- or multi-photon absorption phenomena, where the excitation of
a fluorochrome is based on the virtually simultaneous absorption of two or multiple photons with energy that in summation matches the energy required for fluorochrome excitation. Typically, photons that arrive within a femtosecond contribute to multiphoton absorption phenomena. To achieve conditions of virtually simultaneous photon arrival at a given tissue volume, high intensity needs to be delivered locally over short periods of time. This is typically implemented using a high numerical aperture objective and by tightly focusing ultra-short highpower pulses such as the ones produced by a Ti:Saphire laser. The resulting fluorescence detected has a quadratic nonlinear dependence on illumination strength; a twofold increase in illumination intensity results in a fourfold increase in fluorescence. Two- or multi-photon in vivo microscopy of living tissues can penetrate at depths beyond 500 μm inside tissue, which is significant penetration improvement over confocal microscopy (Helmchen and Denk, 2005), and they are gaining in popularity for biomedical in vivo imaging (Germain et al., 2005, Halin et al., 2005). Applications include the study of angiogenesis (Bird et al., 2005; Brown et al., 2001) and metastasis (Voura et al., 2004; Wang et al., 2002), immunology (Germain et al., 2005; Runnels et al., 2006), kidney physiology (Dunn et al., 2002; Molitoris and Sandoval, 2005) and neurological research (Majewska et al., 2006; Misgeld and Kerschensteiner, 2006). An example of twophoton microscopy applied to imaging a mammary adenocarcinoma is shown in Figure 46.4. Macroscopic Imaging In contrast to microscopy, macroscopic fluorescence imaging exchanges resolution for penetration depth and field of view. In macroscopy, the scattered light is not rejected as in
1 integrins/E-cadherin
50 μm
Figure 46.2 Collective cell movement from primary melanoma explants: plasticity of cell–cell interaction, 1 integrin function and migration strategies. Multicellular cell group emigrating from a primary melanoma explant after 6 days of culture in a 3D collagen lattice. 1 integrins (green) cluster and provide traction in a subset of cells at the leading edge only, while E-cadherin (red) shows linear staining along cell–cell junctions throughout the cluster. Images represent confocal sections of cells fixed in the process of migration, visualized by confocal microscopy. The arrow indicates the direction of migration. The asterisk shows the region of partial matrix degradation and remodeling at the trailing edge (matrix defect). (Modified from Hegerfeldt, Y., et al. (2002). Cancer Res 62:2125–2130. Courtesy of Professor P. Friedl, University of Wurzburg.)
Imaging Technology
■
527
(a)
42 m
(b)
Figure 46.4 Multiphoton intravital imaging of a TG1-1 mammary adenocarcinoma growing in a dorsal skinfold chamber of a TIE2-GFP mouse. The chamber places a window on the tumor to allow for direct visualization of tumor biology and function. In this case, GFP expressed by endothelial cells is shown in green while Tetramethlyrhodamine (TMR)-Dectran injected i.v. is shown in red. The image is a maximum intensity projection of ~150–200 μm under the surface, 250 μm across. (Courtesy of Ed Brown, University of Rochester.)
Figure 46.3 Confocal intravital images obtained in vivo with a flexible fiber probe of 650 m. (a) Normal human alveoli: visualization of normal distal lung, with distinct alveolar microarchitecture. The signal shown is tissue auto-fluorescence; no dye was applied in this case. The miniaturized fiber probes can be used in conjunction with a traditional bronchoscope since they are compatible with the working channel of conventional endoscopes. Field of view is 600 500 m and the slice thickness is 20 m obtained from the surface of the tissue. (b) In vivo angiogenesis imaging: visualization of tumoral vessels in a mouse prostate after FITC-Dextran (500 kDa) injection in the tail. The site was accessed through a micro-incision in the skin at the site of the tumor. Field of view is 400 280 m. An optical slice of 20 m is imaged at the surface of the tissue. (The images were obtained by Anne-Carole Duconseille and Olivier Clément, Université Paris V, Paris, France. Images courtesy of Mauna Kea Technologies (Paris, France). The composite image is by reproduction from Ntziachristos [2006].)
microscopy but is the major source of contrast. Images can be formed by direct collection of the light scattered from superficial and deeper tissues but accurate and high performing methods use elaborate theoretical models of photon propagation and inversion techniques to resolve depth and improve on the resolution and the quantification achieved. Fluorescence macroscopy is
separated to planar methods and tomographic methods respectively, as described in the following sections. Planar Imaging Simple “photographic methods” have been used to enable in vivo fluorescence imaging from small animals and tissues. The method operates using light with a defined bandwidth, such as a laser source, laser diode or an appropriately filtered bright lamp. This light source is expanded and emitted toward tissues in order to generate an illumination field. Light incident on the animal surface can propagate for several millimeters in tissues, especially in the far-red and near-infrared (NIR) (630–900 nm) because of the low light attenuation by tissues in this spectral range. The corresponding photon field established in the tissue has a depthdependent strength and can excite tissue fluorochromes in its path. Fluorescence signals generated by this process can similarly propagate to the tissue surface where they are recorded with a highly sensitive charge-coupled device (CCD) camera through appropriate filters. This approach can be applied in the epi-illumination or reflectance mode where illumination and detection are at the same side of the animal as shown in Figure 46.5a. An alternative imaging mode is transillumination, where illumination and detection are performed from opposite sides of the animal, as shown in Figure 46.5b. In both modes it is custom to acquire both a fluorescence image and a second image measured through different filters to obtain a “photograph” of the animal at the excitation
528
(a)
CHAPTER 46
■
Epi-illumination
Figure 46.5
Fluorescence Imaging: Overview and Applications in Biomedical Research
(b)
Transillumination
Modes of planar macroscopic imaging.
wavelength. An example of epi-illumination imaging is shown in Figure 46.6. Generally, epi-illumination imaging offers better sensitivity to surface activity whereas transillumination is more sensitive and accurate in detecting deeper-seated activity. Planar imaging comes with significant implementation simplicity and offers high-throughput imaging. However, it also has important shortcomings as it does not correct for signal intensity changes with varying tissue optical properties or with fluorochrome depth (Ntziachristos et al., 2005). Therefore the intensity reported is generally of qualitative nature, unless the depth of the fluorescence activity or the optical properties of tissue are not changing throughout the course of the measurements. In such case measurements can be related to each other in a relative, semiquantitative way. Other limitations include the single projection viewing that does not allow for resolving or localizing activity with depth. Fluorescence Molecular Tomography (FMT) In order to improve imaging performance, it is necessary to account for light intensity variations as a function of depth and tissue optical properties.To achieve this, it is necessary to resolve the position of the fluorescent lesions imaged. Three-dimensional information can be gained by illuminating tissue at different projections and correspondingly collect light around the animal, as shown schematically in Figure 46.7. Ideally, the animal is illuminated using projections over 360° angles but different implementations exist that utilize a limited number of projections, which may compromise the resolution achieved but attains implementation simplicity and improves imaging over planar imaging methods. Similar to other tomographic methods, FMT uses a forward model, that is, a theoretical prediction of the measurements expected assuming a known fluorochrome distribution. For image reconstruction, this forward model is used with the set of measurements collected at multiple projections to retrieve the unknown fluorochrome distribution. The development of appropriate forward problems plays an important role in FMT. Largely based on the theoretical mainframe of theory of photon diffusion (Gibson et al., 2005), in vivo fluorescence tomography needs to account for signal variations due to tissue absorption
(a)
Excitation
(b)
Emission
(c)
Merged
Figure 46.6 Imaging protease activity associated with HT1080 fibrosarcoma formation due to subcutaneous implantation of 106 HT1080 cells 2 weeks prior to imaging. (a) Planar image (photograph) of the animal at the excitation wavelength (672 nm), (b) planar image (photograph) at the emission (fluorescence; 700–750 nm) wavelength showing increased fluorescence activity from the tumor, due to increased cathepsin expression at the propagating tumor border. Cathepsins are tagged with an activatable fluorescence probe, intravenously administered 24 h prior to imaging. (c) Merged images.
Multiprojection illumination
Figure 46.7 Multiprojection imaging used for macroscopic fluorescence tomography. Projections are typically separated by time-sharing, that is, by sequentially switching the light on, in each projection. While projection is schematically shown here as plane propagation, in reality, for each light source turned on, there is a diffusive photon pattern propagating through tissue that covers a large part of the volume imaged.
and scattering heterogeneity (Ntziachristos et al., 2005). This is generally achieved by using measurements at the excitation wavelength to obtain information of the spatially dependent photon attenuation in tissues due to the variation in optical attenuation of the various organs (Figure 46.8).
Imaging Technology
(c)
(b)
(a) FMT
■
529
Merged
CT
Figure 46.8 FMT imaging of spontaneous lung cancer of a mouse model of conditional K-ras oncogene mutation. (a) Three consecutive slices from approximately the center of the animal in the coronal view, showing increased cathepsin activity associated with two spontaneous tumors developed. Contrast is due to fluorescence at 700–740 nm observed as a result of activation of quenched Cy5.5 dye carried on a biocompatible molecular probe with specificity to cathepsins. (b) Corresponding X-ray CT slice. Arrows demonstrate the anatomical presence of tumors, congruent with the fluorescent signals. (c) Co-registration and rendering of the three-dimensional fluorescence activity reconstructed by FMT (red) onto the anatomical X-ray CT volume (green).
Original tomographic systems utilized multiple fibers, brought in contact with tissue, coupled in some instances with matching fluids in order to perform tomographic measurements from animals. Using more elaborate forward problems, recent implementations operate on direct photon collection from CCD cameras and free tissue surfaces at the absence of matching media. Such approaches improve animal handling and offer high quality datasets for improved imaging performance. FMT, in its simplest implementation, needs only constant intensity light. Other approaches utilize ultra-short laser pulses for illumination and time-gated detection, or light of intensity that is modulated in the 100 MHz to 1 GHz range. These methods are appropriate for simultaneously resolving fluorochrome concentration and lifetime. For small animal imaging, resolution of the order to 0.5–1 mm is possible although the computational requirements for achieving this resolution in large volumes remain challenging. There is a diversity of applications for tomographic methods, due to the availability of an increasing number of fluorescent probes and the improved access to a larger number of organs granted by three-dimensional imaging (Weissleder and Ntziachristos, 2003). FMT has been used for example for assessing protease upregulation (Ntziachristos et al., 2002), monitoring apoptotic signatures based on annexin-V fluorescent conjugates (Ntziachristos et al., 2004) or imaging angiogenesis and receptor upregulation (Montet et al., 2005; Patwardhan et al., 2005). Hyper-Spectral and Multi-Spectral Imaging Hyper-spectral (HS) imaging generally refers to imaging methods that capture multiple images (n 10) at different spectral regions. Typically the spectral regions are selected consecutive over a wide spectral window and have a fine spectral step of a
few nanometers. In this way tens to hundreds of images may be used to cover the visible or NIR spectrum for example. In contrast, multi-spectral (MS) imaging typically refers to methods that capture a few and wider spectral regions. Hyper-spectral and multi-spectral methods can be applied in generic fluorescence applications such as plate and array readers, in microscopy applications and in macroscopic tissue imaging. Typically HS is used for determining spectral features in unknown fluorescence signatures. Applications include improved reading of microarrays when contaminating fluorescence signatures may be present (Martinez et al., 2003), in removing unknown auto-fluorescence signatures in planar fluorescence imaging and improve contrast (Mansfield et al., 2005) or in tissue diagnostics based on auto-fluorescence patterns, for example detecting cancer based on changes in auto-fluorescence emission (Demos et al., 2005; Farkas and Becker, 2001; Martin et al., 2006). The related instrumentation utilizes tunable filters, such as liquid crystal tunable filters (LCTF) or more commonly acousto-optic tunable filters (AOTF) to scan over wide spectral regions. Due to the wealth of spectral information, model-based fitting methods or decomposition methods, such as principle component analyses, can be used for data processing and obtaining meaningful answers. Conversely, MS is used when known and spectrally differentiated fluorescence signatures are imaged, for example, in microscopy or tissue imaging of selected fluorescent dyes or proteins. Then each of the spectral windows imaged carries information on the distribution of certain known fluorochromes. In this case, it is assumed that there is either little signal contamination and crosstalk or spectral un-mixing techniques can be otherwise employed. For example, one or few spectral channels can be assumed as background measurements, at spectral regions where the fluorochromes employed are not emitting, and used
530
CHAPTER 46
■
Fluorescence Imaging: Overview and Applications in Biomedical Research
for auto-fluorescence subtraction. While LCTF or AOTF solutions are employed in MS imaging, it is also common to use individual band-pass filters optimized for separating fluorescence light coming from different fluorochromes. Since a smaller number of individual spectral measurements is obtained, there is flexibility in selecting higher performance filters, than the spectral separation available in HS techniques. Better illumination systems can be also used, for example laser-based systems containing multiple excitation lines that are better tuned for each particular fluorochrome used, to achieve optimal signal-to-noise characteristics.
FLUORESCENCE APPLICATIONS IN GENOMIC MEDICINE A variety of targeted probes (NIR fluorochrome attached to affinity ligand) and activatable probes (based on quenching or fluorescence resonance energy transfer, FRET) have recently been developed for in vivo imaging of cellular and subcellular biomarker (Weissleder and Ntziachristos, 2003), which expands the repertoire of fluorescence applications to biomedical research and/or clinical medicine. These probes allow targeting of proteases, cellular receptors and other proteins in vivo. In addition, fluorescent proteins and bioluminescence approaches have revolutionized
modern biology and are used in reporting on molecular function in vivo (Gross and Piwnica-Worms, 2005; Massoud and Gambhir, 2003). A particular recent advance is the availability of red-shifted fluorescent proteins, which can significantly increase the detection sensitivity and penetration depth over the green (GFP), yellow (YFP) and common red proteins, such as DsRed2 (Giepmans et al., 2006;Tsien, 2005). This technology, combined with the imaging approaches presented above, offers a new dimension to the already widespread use of fluorescence methods in genomics, that is, fluorescence in situ hybridization, other genomic hybridization approaches and high-throughput microarray technology, as it raises the possibility of in vivo study of gene function in organisms. While gene reporter technologies (i.e., the use of fluorescent proteins) are more suitable for animal imaging, extrinsically administered fluorescence probes can be used in clinical applications as well. A particular benefit of in vivo imaging is that it does not study a single time point, as microarray studies, but can follow evolving function over time. In vivo imaging can therefore study the evolution and dynamic interactions of targets identified on high-throughput arrays. In this sense, fluorescence presents a highly involved modality for genomic imaging that can link in vitro to in vivo for understanding dynamic interactions and in diagnostic and therapeutic preclinical and clinical applications.
REFERENCES Achilefu, S., Bloch, S., Markiewicz, M.A., Zhong, T.X., Ye, Y.P., Dorshow, R.B., Chance, B. and Liang, K.X. (2005). Synergistic effects of light-emitting probes and peptides for targeting and monitoring integrin expression. Proc Natl Acad Sci USA 102, 7976–7981. Bird, D.K., Yan, L., Vrotsos, K.M., Eliceiri, K.W., Vaughan, E.M., Keely, P.J., White, J.G. and Ramanujam, N. (2005). Metabolic mapping of MCF10A human breast cells via multiphoton fluorescence lifetime imaging of the coenzyme NADH. Cancer Res 65, 8766–8773. Bogdanov, A.A., Lin, C.P., Simonova, M., Matuszewski, L. and Weissleder, R. (2002). Cellular activation of the self-quenched fluorescent reporter probe in tumor microenvironment. Neoplasia 4, 228–236. Brown, E.B., Campbell, R.B., Tsuzuki, Y., Xu, L., Carmeliet, P., Fukumura, D. and Jain, R.K. (2001). In vivo measurement of gene expression, angiogenesis and physiological function in tumors using multiphoton laser scanning microscopy. Nat Med 7, 864–868. [erratum appears in Nat Med 2001 Sep; 7(9): 1069] Demos, S.G., Lieber, C.A., Lin, B. and Ramsamooj, R. (2005). Imaging of tissue microstructures using a multimodal microscope design. IEEE J Sel Top Quant Electron 11, 752–758. Denk, W., Strickler, J.H. and Webb, W.W. (1990). 2-Photon Laser Scanning Fluorescence Microscopy. Science 248, 73–76.
D’Hallewin, M.A., El Khatib, S., Leroux, A., Bezdetnaya, L. and Guillemin, F. (2005). Endoscopic confocal fluorescence microscopy of normal and tumor bearing rat bladder. J Urol 174, 736–740. Dunn, K.W., Sandoval, R.M., Kelly, K.J., Dagher, P.C., Tanner, G.A., Atkinson, S.J., Bacallao, R.L. and Molitoris, B.A. (2002). Functional studies of the kidney of living animals using multicolor twophoton microscopy. Am J Physiol Cell Physiol 283, C905–C916. Farkas, D.L. and Becker, D. (2001). Applications of spectral imaging: Detection and analysis of human melanoma and its precursors. Pigment Cell Res 14, 2–8. Germain, R.N., Castellino, F., Chieppa, M.C., Egen, J.G., Huang, A.Y.C., Koo, L.Y. and Hai, Q. (2005). An extended vision for dynamic high-resolution intravital immune imaging. Semin Immunol 17, 431–441. Gibson, A.P., Hebden, J.C. and Arridge, S.R. (2005). Recent advances in diffuse optical imaging. Phys Med Biol 50, R1–R43. Giepmans, B.N.G., Adams, S.R., Ellisman, M.H. and Tsien, R.Y. (2006). Review: The fluorescent toolbox for assessing protein location and function. Science 312, 217–224. Gross, S. and Piwnica-Worms, D. (2005). Spying on cancer: Molecular imaging in vivo with genetically encoded reporters. Cancer Cell 7, 5–15. Halin, C., Mora, J.R., Sumen, C. and von Andrian, U.H. (2005). In vivo imaging of lymphocyte trafficking. Ann Rev Cell Dev Biol 21, 581–603.
References
Helmchen, F. and Denk, W. (2005). Deep tissue two-photon microscopy. Nat Methods 2, 932–940. Herschman, H.R. (2003). Molecular imaging: Looking at problems, seeing solutions. Science 302, 605–608. Jain, R.K., Munn, L.L. and Fukumura, D. (2002). Dissecting tumour pathophysiology using intravital microscopy. Nat Rev Cancer 2, 266–276. Mansfield, J.R., Gossage, K.W., Hoyt, C.C. and Levenson, R.M. (2005). Autofluorescence removal, multiplexing, and automated analysis methods for in-vivo fluorescence imaging. J Biomed Opt 10. Martin, M.E., Wabuyele, M., Panjehpour, M., Overholt, B., DeNovo, R., Kennel, S., Cunningham, G. and Vo-Dinh, T. (2006). An AOTFbased dual-modality hyperspectral imaging system (DMHSI) capable of simultaneous fluorescence and reflectance imaging. Med Eng Phys 28, 149–155. Martinez, M.J., Aragon, A.D., Rodriguez, A.L., Weber, J.M., Timlin, J.A., Sinclair, M.B., Haaland, D.M. and Werner-Washburne, M. (2003). Identification and removal of contaminating fluorescence from commercial and in-house printed DNA microarrays. Nucleic Acids Res 31. Majewska, A., Newton, J. and Sur, M. (2006). Remodeling of synaptic structure in sensory cortical areas in vivo. J Neurosci 26, 3021–3029. Massoud, T.F. and Gambhir, S.S. (2003). Molecular imaging in living subjects: Seeing fundamental biological processes in a new light. Genes Dev 17, 545–580. Misgeld, T. and Kerschensteiner, M. (2006). In vivo imaging of the diseased nervous system. Nat Rev Neurosci 7, 449–463. Molitoris, B.A. and Sandoval, R.M. (2005). Intravital multiphoton microscopy of dynamic renal processes. Am J Physiol Renal Physiol 288, F1084–F1089. Montet, X., Ntziachristos, V., Grimm, J. and Weissleder, R. (2005). Tomographic fluorescence mapping of tumor targets. Cancer Res 65, 6330–6336. Norman, K. (2005). Techniques: Intravital microscopy – a method for investigating disseminated intravascular coagulation?. Trends Pharmacol Sci 26, 327–332. Ntziachristos, V. (2006). Fluorescence molecular imaging. Ann Rev Biomed Eng 8, 1–33. Ntziachristos, V., Tung, C., Bremer, C. and Weissleder, R. (2002). Fluorescence-mediated tomography resolves protease activity in vivo. Nature Med 8, 757–760.
■
531
Ntziachristos,V., Schellenberger, E.A., Ripoll, J.,Yessayan, D., Graves, E., Bogdanov, A., Josephson, L. and Weissleder, R. (2004).Visualization of antitumor treatment by means of fluorescence molecular tomography with an annexin V-Cy5.5 conjugate. Proc Natl Acad Sci USA 101, 12294–12299. Ntziachristos, V., Ripoll, J., Wang, L.H.V. and Weissleder, R. (2005). Looking and listening to light: The evolution of whole-body photonic imaging. Nat Biotechnol 23, 313–320. Patwardhan, S.V., Bloch, S.R., Achilefu, S. and Culver, J.P. (2005). Timedependent whole-body fluorescence tomography of probe biodistributions in mice. Opt Express 13, 2564–2577. Runnels, J.M., Zamiri, P., Spencer, J.A., Veilleux, I., Wei, X.B., Bogdanov, A. and Lin, C.P. (2006). Imaging molecular expression on vascular endothelial cells by in vivo immunofluorescence microscopy. Mol Imaging 5, 31–40. Shaner, N.C., Campbell, R.E., Steinbach, P.A., Giepmans, B.N.G., Palmer, A.E. and Tsien, R.Y. (2004). Improved monomeric red, orange and yellow fluorescent proteins derived from Discosoma sp red fluorescent protein. Nat Biotechnol 22, 1567–1572. Sipkins, D.A., Wei, X.B.,Wu, J.W., Runnels, J.M., Cote, D., Means, T.K., Luster, A.D., Scadden, D.T. and Lin, C.P. (2005). In vivo imaging of specialized bone marrow endothelial microdomains for tumour engraftment. Nature 435, 969–973. Tsien, R.Y. (2005). Building and breeding molecules to spy on cells and tumors. FEBS Lett 579, 927–932. Tung, C.H. (2004). Fluorescent peptide probes for in vivo diagnostic imaging. Biopolymers 76, 391–403. Tyagi, S., Marras, S.A.E. and Kramer, F.R. (2000). Wavelength-shifting molecular beacons. Nat Biotechnol 18, 1191–1196. Voura, E.B., Jaiswal, J.K., Mattoussi, H. and Simon, S.M. (2004). Tracking metastatic tumor cell extravasation with quantum dot nanocrystals and fluorescence emission-scanning microscopy. Nat Med 10, 993–998. Wang, W.G., Wyckoff, J.B., Frohlich, V.C., Oleynikov, Y., Huttelmaier, S., Zavadil, J., Cermak, L., Bottinger, E.P., Singer, R.H., White, J.G. et al. (2002). Single cell behavior in metastatic primary mammary tumors correlated with gene expression patterns revealed by molecular profiling. Cancer Res 62, 6278–6288. Weissleder, R. and Ntziachristos, V. (2003). Shedding light onto live molecular targets. Nat Med 9, 123–128.
CHAPTER
47 Imaging Genetics: Integration of Neuroimaging and Genetics in the Search for Predictive Markers Ahmad R. Hariri
INTRODUCTION Individual differences in stable and enduring aspects of behavior – traits such as personality and temperament – are important predictors of vulnerability to neuropsychiatric disorders, including depression, anxiety, and addiction. Accordingly, identifying the biological mechanisms that give rise to trait individual differences affords a unique opportunity to develop both predictive markers of disease liability and novel targets for individualized treatment. In the past 5 years, human neuroimaging studies, especially those employing functional MRI (fMRI), which provides an indirect measure of brain activation (see Chapter 43), have begun to reveal the neural substrates of interindividual variability in these and related constructs (Drabant et al., 2006; Pezawas et al., 2005). Moreover, recent studies have established that blood oxygen level-dependent (BOLD) fMRI measures represent temporally stable and reliable indices of brain function (Johnstone et al., 2005; Manuck et al., 2007). Thus, much like their behavioral counterparts, patterns of brain activation represent enduring, trait-like phenomena, which in and of themselves may serve as important markers of liability and pathophysiology.
Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 532
As neuroimaging studies continue to illustrate the predictive relationship between regional brain activation and trait-like behaviors (e.g., increased amygdala reactivity predicts core features of anxious temperament (Haas et al., 2007; Most et al., 2006; Ray et al., 2005; Somerville et al., 2004; van Reekum et al., 2007)), an important next step is to systematically identify the underlying mechanisms driving variability in brain circuit function. In this regard, recent neuroimaging studies employing pharmacological challenge paradigms, principally targeting monoamine neurotransmission, have revealed that even subtle alterations in dopaminergic, noradrenergic, and serotonergic signaling can have profound impact on the functional response of brain circuitries supporting affect, personality and temperament (Bigos et al., 2008; Hariri et al., 2002a, b; Harmer et al., 2006; Tessitore et al., 2002; Strange and Dolan, 2004). Similarly, multimodal neuroimaging approaches have provided evidence for directionally specific relationships between key components of monoaminergic signaling cascades, assessed with radiotracer positron emission tomography (PET), and brain function, assessed with BOLD fMRI. Collectively, pharmacological challenge neuroimaging and multimodal PET/ fMRI are revealing how variability in behaviorally relevant brain activation emerges as a function of underlying variability in key
Copyright © 2009, Elsevier Inc. All rights reserved.
Conceptual Basis of Imaging Genetics
brain neurotransmission systems (e.g., increased serotonin signaling predicting increased amygdala reactivity (Fisher et al., 2006)). The next logical step is to identify the sources of interindividual variability in these key neurochemical signaling mechanisms. In the modern genomic era of human genetics, this step is firmly planted in the direction of identifying the relationships between common variation in the genes encoding components of these signaling cascades, their protein products and, subsequently, brain circuit function (Figure 47.1). As sequence variation across individuals represents the ultimate wellspring of variability in emergent neurobiological and related behavioral processes, understanding the relationships between genes, brain and behavior is critical for establishing the etiology and pathophysiology of psychiatric disease. The emerging field of imaging genetics seeks to establish a principled framework for the integration of modern genetics and neuroimaging technologies toward the ultimate goal of identifying truly predictive markers of disease vulnerability (Hariri and Weinberger, 2003a, b) (Figure 47.1). The current chapter provides a brief review of the basic principles of imaging genetics and highlights the vast potential of such an integrated approach by briefly reviewing recent studies of the neural correlates of genetically driven variability in serotonin (5-HT) neurotransmission. The collective results of such studies demonstrate that common sequence variation in the human serotonin transporter gene is associated with downstream alterations in serotonin signaling cascades, resulting in relatively increased serotonin signaling and, eventually, increased amygdala reactivity to environmental threat (Hariri and Holmes, 2006).
■
533
This genetically driven variability in serotonin neurotransmission and threat-related amygdala reactivity likely represent a key mechanism of increased temperamental anxiety and risk for depression, especially in the context of environmental adversity. With increased utilization of such imaging genetics strategies and their continued expansion to include pharmacological and multimodal neuroimaging techniques, many more behaviorally and clinically relevant neurobiological pathways and predictive markers will be illuminated in forthcoming years. Such discoveries, in turn, will directly inform ongoing efforts to establish genomic and personalized medicine.
CONCEPTUAL BASIS OF IMAGING GENETICS Genes have potential impact on all levels of biology. In the context of disease states, particularly behavioral disorders, genes represent the cornerstone of mechanisms that – either directly or in concert with environmental events – ultimately result in disease. Moreover, genetic and genomic analyses offer the potential to identify both at-risk individuals and biological pathways for the development of new treatments. While most human behaviors cannot be explained by genes alone (and certainly much variance in aspects of brain information processing will not be genetically determined directly), it is anticipated that variations in the genome sequence that impact gene function will contribute an appreciable amount of variance to these resultant complex behavioral phenomena. This conclusion is implicit
Neurobiology of Inter-Individual Variability in Behavior
Genes: multiple alleles each of small effect
Cells: subtle molecular alterations
Systems: response bias to environmental cues
Behavior: complex functional interactions and emergent phenomena
Figure 47.1 Schematic of the complex path from the genome to behavior. Imaging genetics allows for the discovery of genetic effects at the level of brain information processing, which represents a proximate biological link to the genome as well as an intermediate of behavior. (Adapted with permission from Hariri et al., 2006.)
534
CHAPTER 47
■
Imaging Genetics: Integration of Neuroimaging and Genetics in the Search for Predictive Markers
in the results of studies of twins that have revealed heritabilities ranging from 40% to 70% for various aspects of cognition, temperament, and personality (Plomin et al., 1994). In the case of psychiatric illness, genes appear to be the only consistent risk factors that have been identified across populations, and the majority of susceptibility for major psychiatric disorders is accounted for by inheritance (Moldin and Gottesman, 1997) (see Chapter 104). Traditionally, the impact of genetic polymorphisms on human behavior has been examined directly using clinical evaluations, personality questionnaires, and neuropsychological batteries. Genetic epidemiological investigations have directly examined the relationship between specific genetic polymorphisms and behaviors and have reported equivocal results (Malhotra and Goldman, 1999). This is not surprising for at least two reasons. First, there is considerable individual variability in dimensions of observable behavior as well as subjectivity in the assessment of behavior, necessitating very large samples – often exceeding several hundred subjects – to identify even small gene effects (Glatt and Freimer, 2002). Moreover, it is apparent that there are etiological subgroups within any given disease that obscure effects at the broader group level. Second and perhaps most importantly, the effects of genes are not expressed directly at the level of behavior. As discussed in detail below, gene effects on behavior are mediated by their molecular and cellular effects on information processing in the brain. Thus, examining gene effects on the brain represents a critical step in understanding their ultimate contribution to variability in behavior. Since genes are involved directly in the development and function of brain regions subserving specific cognitive and emotional processes, functional polymorphisms in genes may be strongly related to the function of these specific neural systems, in turn, may mediate the genes involvement in behavioral outcomes. This is the underlying assumption of our investigations examining the relation between genes and neural systems – what we initially called “imaging genomics” (Hariri and Weinberger, 2003a, b). More recently, we have described this approach as “imaging genetics” (Neumann et al., 2006), because it is typically utilized to explore variation in specific genes and not the genome broadly. The potential for marked differences at the neurobiological level underscores the need for a direct assay of brain function. Accordingly, imaging genetics within the context of a candidate gene association approach (see Chapter 8) provides an ideal opportunity to further our understanding of biological mechanisms potentially contributing to individual differences in behavior and personality. Moreover, imaging genetics provides a unique tool with which to explore and evaluate the functional impact of brain-relevant genetic polymorphisms with the potential to understand their impact on behavior (Figure 47.1). Of course, the relevance of imaging genetics findings for disease vulnerability will only be established if the variants under study are further associated with disease risk directly or if their impact on brain function is manifest, or even exaggerated, in the diseases of interest.
BASIC PRINCIPLES OF IMAGING GENETICS Selection of Candidate Genes The protocol for imaging genetics typically involves first identifying a meaningful variation in the genomic sequence within a candidate gene. For the variant to be meaningful, it should have an impact at the molecular and cellular level in gene or protein function (i.e., a functional variant), and the distribution of such effects at the level of brain systems involved in specific forms of information processing should be predictable. Consistent with the goal of genetic association studies to identify variation impacting individual differences in behavior and related risk for disease in the general population, candidate polymorphisms in imaging genetics studies should be relatively common. That is, the frequency of the less common or minor allele should ideally be greater than 5%. In light of the considerable costs of neuroimaging in comparison with behavioral studies and the related sample size limitations, the minor allele frequency of candidate polymorphisms in imaging genetics studies should ideally be greater than 30%. In such cases, the relative contributions of all potential genotypes (i.e., homozygotes for both major and minor alleles, as well as heterozygotes) to brain function can be examined using imaging genetics. When frequently occurring genetic variants have been demonstrated to affect specific physiological processes at the molecular and cellular level in distinct brain regions and circuits, neuroimaging can be employed to explore their effects on these substrates in both normal and impaired populations. Short of well-defined functional polymorphisms, candidate genes with identified single nucleotide polymorphisms (SNPs) or other allele variants in coding or promoter regions with likely functional implications (e.g., nonconservative amino acid substitution or missense mutation in a promoter consensus sequence) involving circumscribed neuroanatomical systems would also be attractive substrates for imaging genetics. In fact, recent imaging genetics studies have taken the lead in exploring the functionality of candidate variants by first describing in vivo effects at the level of brain systems (Brown et al., 2005). As such, imaging genetics can provide the initial impetus for further characterization of molecular and functional effects of specific candidate genes in brain systems involved in regulating behavior. In this manner, the contributions of abnormalities in these systems to complex behaviors and emergent phenomena, possibly including psychiatric illnesses, can then be understood from the perspective of their neurobiological origins. Control for Non-genetic Factors The contribution of single genes to the structural and functional integrity of brain systems, while putatively more substantial than that to emergent behavioral phenomena, is still presumably small. Furthermore, typically large effects of age, sex, and IQ as well as environmental factors such as illness, trauma, or substance abuse
Imaging Genetics and the Neurobiology of the 5-HTTLPR
on phenotypic variance can easily obscure these small potential gene effects. Since association studies in imaging genetics, as in any case–control association study, are susceptible to population stratification artifacts (i.e., differing ancestral genetic backgrounds for specific candidate genotypes; see Chapter 2), ethnic matching within genotype groups and genomic controls is also potentially critical (Devlin et al., 2001). Thus, the identification and contribution of genetic variation to specific phenotypes should be limited to studies in which other potential contributing and confounding factors are carefully matched across genotype groups. If the imaging protocol involves performance of a task, as is typical in fMRI, the groups should also be matched for level of performance, or at least any variability in perfor mance should be considered in the analysis and interpretation of the imaging data. This is because task performance and the fMRI BOLD signal are tightly linked, and systematic differences in performance between genotype groups could either obscure a true gene effect or masquerade for one. Task Selection There has been tremendous proliferation of functional neuroimaging studies accompanied by behavioral tasks designed specifically for this experimental setting. Many of these are modified versions of classic behavioral and neuropsychological tests (e.g., the Wisconsin Card Sorting Task (Axelrod, 2002)) designed to tap neural systems critical to particular behaviors. More recent paradigms have emerged that focus on interactions of specific behaviors and disease states as these questions have become newly accessible with noninvasive imaging (e.g., the emotion Stroop, which taps abilities to override emotionally provocative stimuli when performing simple behavioral tasks, and Obsessive Compulsive Disorder (Whalen et al., 2006)). Because of the relatively small effects of single genes, even after having controlled for non-genetic and other confounding variables, imaging tasks must maximize sensitivity and inferential value. As the interpretation of potential gene effects depends on the validity of the information processing paradigm, it is best to select well-characterized paradigms that are effective at engaging circumscribed brain regions and systems, and which produce robust signals in every individual and show variance across individuals. In short, imaging genetics studies are typically not the appropriate venue to design and test new functional tasks, and to do so might undermine their tremendous potential.
IMAGING GENETICS AND THE NEUROBIOLOGY OF THE 5-HTTLPR Converging evidence from rodents and non-human primates, as well as extensive human research, has implicated variability in 5-HT neurotransmission as a key predictor of individual differences in multiple, overlapping behavioral constructs related to trait negative affect (Lucki, 1998). Research employing pharmacologic challenge of the 5-HT system (via specific
■
535
receptor agonism/antagonism or general reuptake blockade) has consistently illustrated that manipulations resulting in relatively increased postsynaptic 5-HT neurotransmission produce potentiated responses in affective neural circuitries, peripheral stress responses, and subjective negative affect (Burghardt et al., 2004; Forster et al., 2006; Harmer et al., 2006; Maier and Watkins, 2005). These and other findings have subsequently spurred intensive efforts to identify genetic polymorphisms in 5-HT subsystems, which ultimately control the regulation of 5-HT neurotransmission as a function of both homeostatic drive and environmental feedback. Such polymorphisms might predict trait negative affect as well as differentiate relative risk for disease (Glatt and Freimer, 2002; Lesch, 2005; Stoltenberg and Burmeister, 2000). The 5-HT transporter (5-HTT), which is responsible for the active clearance of synaptic 5-HT and, thus, regulation of pre- and postsynaptic 5-HT receptor stimulation, is of particular importance in efforts to identify genetic polymorphisms in 5-HT subsystems that impact trait negative affect as well as differentiate relative risk for disease. In 1996, a relatively common functional promoter polymorphism in the human 5-HTT gene (SLC6A4) was linked to relatively increased trait anxiety in healthy adult volunteers (Lesch et al., 1996). The so-called 5-HTTLPR (5-HTT gene-linked polymorphic region) is typically defined by two variable nucleotide tandem repeat elements, a short (S) allele comprised of 14 copies of a 20–23 base pair repeat unit and a long (L) allele comprised of 16 copies (Wendland et al., 2006). While initial in vitro and in vivo assays revealed relatively diminished 5-HTT density associated with the S allele (Heinz et al., 1998;Lesch et al., 1996), recent work has indicated that more complex mechanisms (e.g., regional up- and down-regulation of specific 5-HT receptors) and not altered 5-HTT density may mediate the long-term impact of the 5-HTTLPR on 5-HT neurotransmission (David et al., 2005; Hariri and Holmes, 2006; Lee et al., 2005; Parsey et al., 2006). Regardless of the underlying mechanisms of action, a modest association has been widely reported between the 5-HTTLPR S allele and relatively increased trait negative affect (Munafo et al., 2005; Schinka et al., 2004; Sen et al., 2004). Moreover, the 5-HTTLPR S allele has been associated with increased risk for depression in the context of environmental adversity, a relationship that may be mediated by increased neuroticism, a psychometrically robust index of trait negative affect (Caspi et al., 2003; Munafo et al., 2005). Imaging genetics studies have provided a unique understanding of how the 5-HTTLPR may impact temperamental anxiety and risk for depression. In a landmark study, fMRI revealed that the reactivity of the amygdala, a core brain region responsible for triggering both physiological and behavioral arousal, to threat-related facial expressions was significantly exaggerated in S allele carriers (Hariri et al., 2002b). Since this original study, there have been multiple replications of the association between the S allele and relatively increased amygdala reactivity in both healthy volunteers and patients with mood disorders; a recent meta-analysis revealed that the 5-HTTLPR may
536
CHAPTER 47
■
Imaging Genetics: Integration of Neuroimaging and Genetics in the Search for Predictive Markers
account for up to 10% of the variability in amygdala reactivity (Munafo et al., 2007). In addition to this consistent effect on amygdala reactivity, the 5-HTTLPR S allele has been further linked with reduced gray matter volumes of and functional coupling between the amygdala and medial prefrontal cortex (Canli et al., 2005; Heinz et al., 2005; Pezawas et al., 2005). As the magnitude of amygdala reactivity (as well as its functional coupling with medial prefrontal cortex) is associated with temperamental anxiety, these imaging genetics findings suggest that the 5-HTTLPR S allele may be associated with increased risk for depression upon exposure to environmental stressors because of its effects on corticolimbic reactivity to potential threat.
FUTURE DIRECTIONS The emerging field of imaging genetics is in its infancy, as the number of genes explored is very few and the strategies for looking at gene effects in brain are relatively simplistic. The studies reviewed above provide compelling evidence that gene effects at the level of amygdala reactivity are more robust (i.e., “penetrant”) than at the level of manifest emotional behaviors. This is consistent with the conclusion that genes related to affect, mood, and temperament are not coding for behavior per se but rather impact on the development and function of neural systems that mediate emotional experience and behavior. Emotional responses are complex and not the result of variation in any single genes; futures studies will emphasize interactions of genes and interactions of genes with the environment. This is likely to add complexity but also a refined resolution to the analyses. Combining existing neuroimaging modalities is another important future direction for imaging genetics. Implementation of multimodal strategies is critical for identifying intermediate
mechanisms mediating the effects of genetic polymorphisms on neural circuit function and related behaviors. The potential of multimodal neuroimaging was recently demonstrated in a study employing both PET and fMRI to identify the impact of 5HT1A autoreceptor regulation of 5-HT release on amygdala reactivity (Fisher et al., 2006). In the study, adult volunteers underwent PET to determine levels of 5-HT1A autoreceptors using a specific radiotracer that binds to 5-HT1A and provides an in vivo index of receptor density. During the same day, all subjects also underwent fMRI, to determine the functional reactivity of the amygdala. Remarkably, the density of 5-HT1A autoreceptors accounted for 30–44% of the variability in amygdala reactivity. Downstream effects on 5-HT1A autoreceptors, notably reduced receptor density, have been hypothesized to mediate neural and behavioral changes associated with the 5-HTTLPR S allele (David et al., 2005; Hariri and Fisher, 2007; Lee et al., 2005). Thus, these findings suggest that 5-HT1A autoreceptor regulation of corticolimbic circuitry represents a key molecular mechanism mediating the effects of the 5-HTTLPR. Although such findings are not yet readily translated to the clinical practice of psychiatry today, they do provide examples of how the integration of advances in genetics and neuroimaging can lead to the eventual application of these technologies in the diagnosis and treatment of psychiatric illness. Individualized treatment strategies based on targeted neurogenetic pathways will very likely contribute to more effective and tolerated intervention platforms. Even more compelling is the potential of these approaches to establish predictive genetic and neurobiological markers of disease risk that will guide strategies for active prevention in the era of genomic and personalized medicine. Continued imaging genetics research at the interface of genes, brain, and behavior holds great promise for further explicating the neurobiological mechanisms through which variability in behavior emerges and affects risk for psychiatric disease in the context of environmental adversity.
REFERENCES Axelrod, B.N. (2002). Are normative data from the 64-card version of the WCST comparable to the full WCST?. Clin Neuropsychol 16(1), 7–11. Bigos, K.L., Pollock, B.G., Aizenstein, H.J., Fisher, P.M., Bies, R.R., Hariri, A.R. (2008). Acute 5-HT reuptake blockade potentiates human amygdala reactivity. Neuropsychopharmacology. [Epub ahead of print] Brown, S.M., Peet, E. et al. (2005). A regulatory variant of the human tryptophan hydroxylase-2 gene biases amygdala reactivity. Mol Psychiatry 10(9), 805. Burghardt, N.S., Sullivan, G.M. et al. (2004). The selective serotonin reuptake inhibitor citalopram increases fear after acute treatment but reduces fear with chronic treatment: A comparison with tianeptine. Biol Psychiatry 55(12), 1171–1178. Canli, T., Omura, K. et al. (2005). Beyond affect: A role for genetic variation of the serotonin transporter in neural activation during a cognitive attention task. Proc Natl Acad Sci USA 102(34), 12224–12229.
Caspi, A., Sugden, K. et al. (2003). Influence of life stress on depression: Moderation by a polymorphism in the 5-HTT gene. Science 301(5631), 386–389. David, S.P., Murthy, N.V. et al. (2005). A functional genetic variation of the serotonin (5-HT) transporter affects 5-HT1A receptor binding in humans. J Neurosci 25(10), 2586–2590. Devlin, B., Roeder, K. et al. (2001). Genomic control, a new approach to genetic-based association studies. Theor Popul Biol 60(3), 155–166. Drabant, E.M., Hariri, A.R. et al. (2006). Catechol O-methyltransferase val158met genotype and neural mechanisms related to affective arousal and regulation. Arch Gen Psychiatry 63(12), 1396–1406. Fisher, P.M., Meltzer, C.C. et al. (2006). Capacity for 5-HT1Amediated autoregulation predicts amygdala reactivity. Nat Neurosci 9(11), 1362–1363. Forster, G.L., Feng, N. et al. (2006). Corticotropin-releasing factor in the dorsal raphe elicits temporally distinct serotonergic responses
References
in the limbic system in relation to fear behavior. Neuroscience 141(2), 1047–1055. Glatt, C.E. and Freimer, N.B. (2002). Association analysis of candidate genes for neuropsychiatric disease: The perpetual campaign. Trends Genet 18(6), 307–312. Haas, B.W., Omura, K. et al. (2007). Emotional conflict and neuroticism: Personality-dependent activation in the amygdala and subgenual anterior cingulate. Behav Neurosci 121(2), 249–256. Hariri, A.R. and Fisher, P.M. (2007). Regulation of corticolimbic reactivity via the 5-HT1A autoreceptor in the pathophysiology and treatment of depression. Future Neurol 2(2), 121–124. Hariri, A.R. and Holmes, A. (2006). Genetics of emotional regulation: The role of the serotonin transporter in neural function. Trends Cogn Sci 10(4), 182–191. Hariri, A.R. and Weinberger, D.R. (2003a). Functional neuroimaging of genetic variation in serotonergic neurotransmission. Genes Brain Behav 2(6), 314–349. Hariri, A.R. and Weinberger, D.R. (2003b). Imaging genomics. Br Med Bull 65, 259–270. Hariri, A.R., Drabant, E.M. and Weinberger, D.R. (2006). Imaging genetics: Perspectives from studies of genetically driven variation in serotonin function and corticolimbic affective processing. Biol Psychiatry 59(10), 888–897. Hariri, A.R., Mattay, V.S. et al. (2002a). Dextroamphetamine modulates the response of the human amygdala. Neuropsychopharmacology 27(6), 1036–1040. Hariri, A.R., Mattay, V.S. et al. (2002b). Serotonin transporter genetic variation and the response of the human amygdala. Science 297(5580), 400–403. Harmer, C.J., Mackay, C.E. et al. (2006). Antidepressant drug treatment modifies the neural processing of nonconscious threat cues. Biol Psychiatry 59(9), 816–820. Heinz, A., Higley, J.D. et al. (1998). In vivo association between alcohol intoxication, aggression, and serotonin transporter availability in nonhuman primates. Am J Psychiatry 155(8), 1023–1028. Heinz, A., Braus, D.F. et al. (2005). Amygdala-prefrontal coupling depends on a genetic variation of the serotonin transporter. Nat Neurosci 8(1), 20–21. Johnstone, T., Somerville, L.H. et al. (2005). Stability of amygdala BOLD response to fearful faces over multiple scan sessions. Neuroimage 25(4), 1112–1123. Lee, M., Bailer, U.F. et al. (2005). Relationship of a 5-HT transporter functional polymorphism to 5-HT1A receptor binding in healthy women. Mol Psychiatry 10(8), 715–716. Lesch, K.P. (2005). Serotonergic gene inactivation in mice: Models for anxiety and aggression?. Novartis Found Symp 268, 111–140. discussion 140–146, 167–170. Lesch, K.P., Bengel, D. et al. (1996). Association of anxiety-related traits with a polymorphism in the serotonin transporter gene regulatory region. Science 274(5292), 1527–1531. Lucki, I. (1998). The spectrum of behaviors influenced by serotonin. Biol Psychiatry 44(3), 151–162. Maier, S.F. and Watkins, L.R. (2005). Stressor controllability and learned helplessness: The roles of the dorsal raphe nucleus, serotonin, and corticotropin-releasing factor. Neurosci Biobehav Rev 29(4–5), 829–841. Malhotra, A.K. and Goldman, D. (1999). Benefits and pitfalls encountered in psychiatric genetic association studies. Biol Psychiatry 45(5), 544–550.
■
537
Manuck, S.B., Brown, S.M. et al. (2007). Temporal stability of individual differences in amygdala reactivity. Am J Psychiatry 164(10), 1613–1614. Moldin, S.O. and Gottesman, I.I. (1997). At issue: Genes, experience, and chance in schizophrenia – positioning for the 21st century. Schizophr Bull 23(4), 547–561. Most, S.B., Chun, M.M. et al. (2006). Attentional modulation of the amygdala varies with personality. Neuroimage 31(2), 934–944. Munafo, M.R., Clark, T. et al. (2005). Does measurement instrument moderate the association between the serotonin transporter gene and anxiety-related personality traits? A meta-analysis. Mol Psychiatry 10(4), 415–419. Munafo, M.R., Brown, S.M. et al. (2007). Serotonin transporter (5-HTTLPR) genotype and amygdala activation: A meta-analysis. Biol Psychiatry. Neumann, S.A., Brown, S.M. et al. (2006). Human choline transporter gene variation is associated with corticolimbic reactivity and autonomic-cholinergic function. Biol Psychiatry 60(10), 1155–1162. Parsey, R.V., Hastings, R.S. et al. (2006). Effect of a triallelic functional polymorphism of the serotonin-transporter-linked promoter region on expression of serotonin transporter in the human brain. Am J Psychiatry 163(1), 48–51. Pezawas, L., Meyer-Lindenberg, A. et al. (2005). 5-HTTLPR polymorphism impacts human cingulate-amygdala interactions: A genetic susceptibility mechanism for depression. Nat Neurosci 8(6), 828–834. Plomin, R., Owen, M.J. et al. (1994). The genetic basis of complex human behaviors. Science 264(5166), 1733–1739. Ray, R.D., Ochsner, K.N. et al. (2005). Individual differences in trait rumination and the neural systems supporting cognitive reappraisal. Cogn Affect Behav Neurosci 5(2), 156–168. Schinka, J.A., Busch, R.M. et al. (2004). A meta-analysis of the association between the serotonin transporter gene polymorphism (5-HTTLPR) and trait anxiety. Mol Psychiatry 9(2), 197–202. Sen, S., Burmeister, M. et al. (2004). Meta-analysis of the association between a serotonin transporter promoter polymorphism (5-HTTLPR) and anxiety-related personality traits. Am J Med Genet 127B(1), 85–89. Somerville, L.H., Kim, H. et al. (2004). Human amygdala responses during presentation of happy and neutral faces: Correlations with state anxiety. Biol Psychiatry 55(9), 897–903. Strange, B.A. and Dolan, R.J. (2004). Beta-adrenergic modulation of emotional memory-evoked human amygdala and hippocampal responses. Proc Natl Acad Sci USA 101(31), 11454–11458. Stoltenberg, S.F. and Burmeister, M. (2000). Recent progress in psychiatric genetics-some hope but no hype. Hum Mol Genet 9(6), 927–935. Tessitore, A., Hariri, A.R. et al. (2002). Dopamine modulates the response of the human amygdala: A study in Parkinson’s disease. J Neurosci 22(20), 9099–9103. van Reekum, C.M., Urry, H.L. et al. (2007). Individual differences in amygdala and ventromedial prefrontal cortex activity are associated with evaluation speed and psychological well-being. J Cogn Neurosci 19(2), 237–248. Wendland, J.R., Martin, B.J. et al. (2006). Simultaneous genotyping of four functional loci of human SLC6A4, with a reappraisal of 5-HTTLPR and rs25531. Mol Psychiatry 11(3), 224–226. Whalen, P.J., Bush, G. et al. (2006). The emotional counting Stroop: A task for assessing emotional interference during brain imaging. Nat Protoc 1(1), 293–296.
CHAPTER
48 Viral Chip Technology in Genomic Medicine Zeno Földes-Papp
INTRODUCTION The advances discussed in this chapter should be relevant to a broad and heterogeneous readership because of their significance, novelty, or wide applicability of viral chip technology in genomic medicine. These advances are making an outstanding contribution to the development of an important field of research. The state-of-the-field, particularly as it has evolved over the last years, is surveyed. As the field is already too expansive for a single article, I apologize in advance for the many omissions, but hope that this chapter captures the excitement of recent achievements in viral chip technology for genomic medicine. Viruses are a heterogeneous group of infectious agents composed mainly of nucleic acids, either DNA or RNA, packed tightly inside the protein coat. The outer shell mainly protects virus from physical, chemical, or enzymatic damage. Following entry into the cell by specific receptors (Table 48.1), the virus uses the host cellular regulatory system and the host cell’s synthetic machinery to produce enzymes required for its replication, and in many cases the viral genome encodes proteins, which act, for example, as RNA-polymerases, transcriptases that are needed in replication. These components are then assembled and released as viral particles.Viruses also use host cell ribosomes to translate viral proteins. Viruses can only replicate within cells. Thus, viruses are an exception to what is understood as the concept of “life.” Genomic and Personalized Medicine, 2-vol set by Willard & Ginsburg 538
ROLE OF VIRUSES IN HUMAN INFECTIOUS DISEASE Viruses cause diseases ranging from acute infections (e.g., poliomyelitis) to chronic infections that are relatively benign (e.g., herpes) to life-threatening chronic infections (e.g., AIDS). As shown in Table 48.2 and in the recommended general references, different viruses lead to markedly different diseases, reflecting the diverse processes by which they damage human tissues and cells. Some common viruses are listed with typical examples of diseases caused in humans. The variety of viruses has required potential hosts to develop two crucial features of adaptive immunity. First, the need to recognize a wide range of different DNA and RNA pathogens had driven the development of receptors on B and T cells of equal or greater diversity. Second, the distinct habitants and life cycles have to be countered by a range of different effector mechanisms. Characteristic features of each virus are its mode of transmission, its mechanism of replication, its pathogenesis or the means by which it causes disease, and the immune response it elicits. Elimination of a viral infection generally requires the destruction of virally infected cells by osmotic lysis, which follows natural killer (NK) or cytotoxic T-cell (CTL) activation. Activation of these immune cells involves cytokines secreted by antigen presenting cells such as macrophages and dendritic cells, or Type I cytokine producing CD4 T cells. NK and CTL Copyright © 2009, Elsevier Inc. All rights reserved.
Role of Viruses in Human Infectious Disease
TABLE 48.1 receptors
Representative viruses and viral binding
Virus
Cell surface receptora
Influenza
Glycophorin A
Rhinovirus
Adhesion molecule ICAM-1
Reovirus
-adrenergic hormone receptor
Rabies
Acetylcholine receptor
Vaccinia
Epidermal growth factor receptor
Epstein-Bar
B cell complement receptor
HIV
T cell, macrophage CD4 receptor
a All viruses need to bind to a specific cell surface molecule in order to enter into the cell. The binding can be blocked by antibody, but there are cases in which the antibody helps the virus to get in.
cells differ in killing virus–infected cells mainly at the recognition site. NK and CTL cells active in these processes recognize unique surface features of virally infected cells, which are not present on non-infected cells. NK cells can use several different receptors that signal them to kill, including lectin-like receptors that recognize carbohydrate on self cells, killing inhibitory receptors (KIRs) that recognize “free” major histocompatibility complex (MHC) class I molecules encoded by genes of the human leukocyte antigen (HLA) C region and overrule the actions of the killing activation receptors (KARs). Once recognition has occurred, the cytotoxic mechanisms of the NK and CTL cells are similar. During the initial phase of viral infection and when the viruses bud from the cell, the particles can be targeted by IgM and IgG antibodies, leading to opsonin-mediated phagocytosis and/or neutralization. Phagocytosis and neutralization by antibodies are components of the natural and adaptive immune system in response to viral infection when the virus is disseminated via lymph, blood, or interstitial fluid, rather sequestered within a cell. Immune response to viruses also varies according to the site where the virus penetrates the human organisms and whether the infection is primary or secondary. General Problem of Needing to Identify the Relevant Virus Because viruses are sequestered within host cells and replicate rapidly, viruses produce mutations rapidly. When these mutations occur at the antigenic epitope, previously active immune effector cells may no longer recognize the altered virus. Immunological resistance develops. Many disease states have been associated with immune dysfunction of varying degrees of severity and significance, for example HIV/AIDS, flu, or colds. Antigenic diversity is the result of extensive antigenic variation in the past. Viruses interfere with the proper functioning of the cytokine network. For example, adenoviruses and rotaviruses inhibit
■
539
interferon formation, hepatitis B virus does not induce interferon, and the EBV releases inhibitory cytokines suchas IL-10. Not all virus infections cause disease. The objective of the virus is to survive and be transmitted. Sometimes this may cause pathology; for example, diarrhea helps the rotaviruses (see Table 48.2) to get into water and to infect another host. Some viruses are obliged to destroy cells in order to spread, but many viruses inhabit the human body without causing disease symptoms for a long time, and these could be regarded as the most successful viruses. EBV in the throat and HSV-1 in the ganglia are examples of this. How Were Viruses Identified Prior to the Development of Chip Technology? Diagnosis depends on the detection of antiviral antibody and/ or viral antigen, not on immunological markers. The nature of investigations undertaken should be selected accordingly. For example, immune cell proliferation is often abnormal in viral infections but does not help in identification of the relevant virus. Accurate monitoring of viral load is now performed by quantitative real-time PCR. In general, there are three main and well-established approaches to identify a virus in the human (for overview, see Striebel et al., 2003): (i) Isolation and cultivation techniques These techniques have some disadvantages, in particular long cultivation times from 3 up to 30 days, lack of sensitivity, and the rather limited number of viruses that can be identified in comparison to the broad spectrum of possible viruses that cause infection and/or disease. Identification of isolates from cell cultures occurs by titration with different serum pools. This is additionally hampered by constant evolving of new viral subtypes. Alternatively, type specific detection may directly be performed using monoclonal antibodies, immunofluorescence, or enzyme immunoassay (EIA). (ii) Indirect virus detection by virus-specific antibodies Indirect virus detection methods include serological determination of specific antibodies by immunofluorescence assay, enzyme immunoassay, or by titration. After infection, only a few viruses release antigens in amounts sufficient to be detectable in body fluids by antibody assays (ELISAs). The presence of antibodies indicates immunity of the human organism in most virus infections, whereas re-infections do not show any symptoms. Antibody generation is detectable shortly after disease breaks out (e.g., in patients with viral hepatitis), but not during the highly infectious incubation time. Diagnosis of infections by cytopathogenic viruses, such as influenza or respiratory infections, may be easily performed in late states of these diseases, but are hardly possible during early disease states. While serotyping of viruses is mostly done by conventional immunological methods, many clinical isolates remain unclassifiable due to the limited number of antibodies against virus surface proteins, for example, of enteroviruses. Array-based assays are able to detect several
540
CHAPTER 48
TABLE 48.2
■
Viral Chip Technology in Genomic Medicine
Some common viruses that cause diseases in humansa
Virus genome
Virus family
Virus Genus with type species
Human disease
DNA(ds)
Herpesvirida
Simplex virus with human herpesvirus 1 and 2 or herpes simplex type 1and 2 (HSV-1, HSV-2)
Cold sores, genital herpes, encephalitis
Varicellovirus with human herpesvirus 3 or varicella-zoster virus (VZV)
Chickenpox, shingles
Cytomegalovirus with human herpesvirus 5 (CMV)
Mononucleosis
Lymphocryptovirus with human herpesvirus 4 or Epstein-Bar virus (EBV)
Mononucleosis, Burkitt’s lymphoma
Roseolovirus with human herpesvirus 6 (HHV-6)
Erythema subitum, roseola infantum
(ds)
Adenoviridae
Mastadenovirus with human adenoviruses types 3 (HAdV-A), 4 (HAdV-E), and 7 (HAdV-B)
Acute respiratory disease (ARD)
(ds)
Poxviridae
Orthopoxvirus with vaccinia virus
Cowpox, vaccinia, smallpox
(ss)
Parvoviridae
Erythrovirus with human parvovirus B-19
Erythema infectiosium, seronegative arthritis, hydrops fetalis
(ds)
Papillomaviridae
Human papilloma virus with human papilloma virus types (strains) 16, 18, and 31
Cervical cancer
(ds)
Hepadnaviridae
Orthohepadnavirus with human hepatitis B virus
Hepatitis B
RNA(ss)
Orthomyxoviridae
Influenza virus A with types H1N1, H2N2, H3N2, and H5N1
Spanish flu, Asian flu, Hong Kong flu, or pandemic threat in 2006-7 flu season
(ss)
Paramyxoviridae
Morbillivirus with measles virus
Measles
Rubulavirus with Mumps virus
Mumps
(ss)
Coronaviridae
Coronavirus with SARS coronavirus
Severe acute respiratory syndrome (SARS)
(ss)
Picornaviridae
Hepatovirus with hepatitis A virus
Hepatitis A
Enterovirus with polioviruses and coxsackieviruses
Poliomyelitis and infectious myocarditis by coxsackievirus B
Rhinovirus with human rhinovirus A
Common cold (acute viral nasopharyngitis)
Rotavirus with rotavirus A, and B
Severe diarrhea among infants and children, or adult diarrhea by rotavirus B
Coltivius with Colorado tick fever virus
Colorado tick fever
(ds)
Reoviridae
(ss)
Togaviridae
Rubivirus with rubella virus
Rubella
(ss)
Flaviviridae
Flavivirus with yellow fever virus, or Dengue virus
Yellow fever, or Dengue hemorrhagic fever found in the tropics
Hepacivirus with hepatitis C virus
Hepatitis C
Arenavirus with lymphocytic choriomenigitis virus
Lymphocytic choriominigitis
Arenavirus with Lassa virus
Lassa fever
(ss)
Arenaviridae
(ss)
Rhabdoviridae
Lyssavirus with rabies virus
Rabies
(ss)
Retroviridae
Deltaretrovirus with human T-lymphotropic virus
T-cell leukaemia and T-cell lymphoma in adults
Lentivirus with human immunodeficiency virus 1 and 2 (HIV-1 and HIV-2)
Acquired immunodeficiency syndrome (AIDS)
a
For general information, see: Rich, R.R., Fleisher, T.A., Shearer, W.T., Schroeder, H.W., Frew, A.J. and Weyand, C.M. (eds.) (2008). Clinical Immunology: Principles and Practice, 3rd edition. Mosby Ltd., New York.
Microfabrication
serotypes with high accuracy. Highly sensitive array-based assay may become a useful alternative in clinical diagnostics. (iii) Direct detection of viral antigen or genomic features Nucleic acid detection methods in virology may be separated into two basic groups: in vitro nucleic acid-amplifying techniques such as PCR including reverse transcriptase (RT)-PCR and non-amplifying techniques. Analysis of virus specific nucleic acids without in vitro amplification may be performed best by hybridization of gene specific probes, followed by radioactive or immunological detection methods (i.e., ELISA, or chemiluminescence). Practical applications of these direct methods are rare, as sensitivities often are too low in comparison to quantitative real-time PCR or RT-PCR by LightCycler or TaqMan. However, PCR and RT-PCR bear increased risks of contamination by non-specific amplification products but the main limitation resides in the difficulty of designing compatible multiplex primer sets for large numbers of viruses to be screened in a single assay, being restricted to a limited number of viral families. By contrast, serial analysis of gene expression and differential display is a powerful high-throughput methodology (ConejeroGoldberg et al., 2005).
■
541
insert. Of course, positive selection vectors facilitated screening, but still we had to do the hard stuff by hand. Today’s DNA chip/ array technology reverses that early approach. Instead of screening an array of unknowns with a defined probe like a synthetic oligonucleotide, PCR product or cloned gene, now a defined probe occupies the chip and then the chip is probed with the unknown sample. Today, identification of disease-relevant genes or proteins often starts with transcription profiling experiments on DNA chips and/or proteomics approaches. The interpretation of the measured data is largely based upon the complete sequences of the human genome and the most important model organisms, respectively. The overall objective of this chapter is twofold. First, the objective is to increase awareness in clinical medicine of the challenges and opportunities presented by DNA microarray technology and the emerging and rapidly changing field of genomic medicine. Second, the aim is to publicize to a broader research community additional challenges associated with the use of microarray technology in medically relevant genome research.
MICROFABRICATION Chip Technology Modern DNA microarray technology started in the mid-1990’s when Pat Brown and his colleagues at Stanford University first reported a high-capacity system in which DNA sequences specific to individual transcripts were immobilized at defined locations in an array format on a solid surface (Brown et al., 1999; DeRisi and Iyer, 1999; DeRisi et al., 1996; Schena et al., 1995), either through mechanical spotting (Lipshutz et al., 1995) or some implementation of in situ synthesis (Fodor et al., 1991). With the original DNA chip technology, microarrays were prepared by high-speed robotic printing of complementary DNAs on glass. Because of the small format and high density of the arrays, hybridization volumes of 2 L could be used that enabled detection of rare transcripts in probe mixtures derived from 2 g of total cellular messenger RNA (Schena et al., 1995). Differential expression measurements were performed by means of simultaneous, two-color fluorescence hybridization in parallel. This microarray-based genome-wide expression analysis of 45 genes first provided an unbiased view of the transcriptional state of a cell population (from Arabidopsis). DNA chips/microarrays are the biotechnology tool of our decade. Pathogen detection, resequencing, comparative genome analysis, and gene expression – they do it all. The amount of data they are generating is overwhelming. The expression levels of thousands of viral genes as well as many mutant variants are documented for numerous states of growth and disease. It is inevitable that DNA chips/microarrays will drive genomic medicine to new horizons. In the early 1990’s, we purified DNA, ligated it, transformed it, and the cloned gene had to be somewhere. We screened thousands of clones by hand with an oligonucleotide just to find one
Probe selection and microarray/chip design are central to the reliability, sensitivity, specificity, and robustness of viral arrays. Because common viral microarray/chip manufacturing technology synthesizes probes with defined sequences, positions, and lengths, array performance can be optimized using data collected from multiple databases, bioinformatics tools, and computer models. Viral oligonucleotide sequences can be designed according to open reading frames sequences obtained from Entrez Genome and Nucleotide Databases (http://www.ncbi.nlm.nih.gov/). The probe selection should be based on the criteria for specificity, sensitivity, and uniformity (Hughes et al., 2001;Wright and Church, 2002). Arrays can be designed with Array Designer 2.02 software (http://www.premierbiosoft.com). Recently, a bioinformatic platform and database were developed with specific probes of all known viral genome sequences to facilitate the design of diagnostic chips (Lin et al., 2006: http://www.bioinfo.csie.ncu.edu.tw/). Some of the key elements of viral chip microfabrication are shown in Figure 48.1 and are common to the production of all microarrays/chips. Probe concentration, probe DNA length, and printing buffer need to be optimized for high quality chip performance depending on the surface chemistry. Slide autofluorescence, spot morphology, reproducibility of arraying, binding efficiency and purity of probes are crucial for the accuracy and reliability of chip data analysis. The manufacturing process ends with a comprehensive series of quality control tests (Földes-Papp et al., 2004). Additionally, a sampling of arrays from every wafer is used to test the batch by running control hybridizations. A quantitative test of hybridization is performed using standardized control probes. For example, the results of 12,900 hybridization reactions on about 150 configured human herpes virus microarrays for the parallel detection of HSV-1 and HSV-2, VZV, EBV,
542
CHAPTER 48
■
Viral Chip Technology in Genomic Medicine
CMV, and HHV-6 showed that the established microarray/chipbased typing procedure was reproducible, virus-specific and sufficiently sensitive with a lower limit of 100 viral copies per mL sample (Földes-Papp et al., 2004, 2005b). A core element of array design, the perfect match/mismatch probe strategy, is also universally applied to the production of viral chips. For each probe designed to be perfectly complementary to a target sequence, a partner probe is generated that is identical except for a single base mismatch in its center. These probe pairs, which are called the perfect match probe and the mismatch probe, allow the quantization and subtraction of signals caused by nonspecific cross-hybridization. The difference in hybridization
signals between the partners, and their intensity ratios, serve as indicators of specific target abundance. The efficiency with which RNA target molecules are captured by, or hybridized to, surface-immobilized oligonucleotides depends upon secondary and tertiary structure of the RNA target strand. To overcome this limitation, RNA is often fragmented to reduce structural effects. The hybridization efficiency of the resulting fragments was determined as a function of fragment length and the amount of RNA captured was evaluated qualitatively by fluorescence intensity normalized to an internal standard (Mehlmann et al., 2005). Optimized conditions for influenza RNA were determined to include a fragmentation time of
Sample preparation
DNA/RNA extraction
AATAGCCTGG Oligo labelling
Primer selection
PCR NH2 NH2 NH2 NH2 NH2
Microarray design Hybridization
Detection
Interpretation
Figure 48.1
Some key elements of viral chip microfabrication (see Striebel et al., 2003).
Immobilization chemistry
Microfabrication
20–30 min at 75°C. These conditions resulted in a maximum concentration of fragments between 38 and 150 nucleotides in length and a maximum in the capture and label efficiency. In the case of viral chips, for example, an online platform and database that provide users with specific probe sequences of all known viral genome sequences was established to facilitate the design of diagnostic chips (Lin et al., 2006). A user can select any number of different viruses and set the experimental conditions such as melting temperature and length of probe. The system then returns the optimal sequences from the database. The experimental design of a microarray determines the confidence that can be assigned to the data. Specificity, sensitivity, reproducibility, and robustness of the experimental data are essential to a successful use of a viral DNA chip (Földes-Papp et al., 2005b). The microarray-chip platform to be chosen has high value for “multiplex” array diagnostics, that is the parallel determination of several parameters from one sample. There are several platform categories available, such as amplified cDNAs, oligonucleotides chemically synthesized from known sequences, and one-color and two-color samples labeled and hybridized to a chip. The selected chip platform has an effect on the complexity and flexibility of the generated data. For example, the analysis of one-color chips is more straightforward than the analysis of two-color chips. Data formats like CEL, CHP, MAS 5, RMA GenePix, or ImaGene affect the signal intensity and quality of data for downstream analysis. In differential gene expression studies, thousands of genes are simultaneously interrogated, and this in turn determines the statistics to be applied (Draghici, 2002) and the use of real and simulated chip data for testing multiple hypotheses (Reiner et al., 2003). The application of z-score statistics to gene ontology terms is described in Doniger et al. (2003). Because microarray experiments produce tremendous amounts of data, data management by databases efficiently organizes and retrieves the raw data. Sharing the data both within the lab and with collaborators requires web-based system access. Several aspects of probe selection and array design are dictated by an array’s intended use. To select probes for viral microarrays/chips, sequence and annotation data from multiple databases are integrated. Custom arrays can be designed for subsets of known genes. Structured Substrates Traditional hybridization assays used nitrocellulose and nylon membranes, and measurements were carried out by autoradiography. Microarrays and other chip assays are based on solid supports, usually glass. Measurements are commonly performed by fluorescence labeling and detection. In contrast to membranebased macroscopic arrays, microarray/chip supports (substrates) are non-porous, and therefore prevent absorption of reagents and samples on the substrate matrix. Non-porosity permits fast removal of organic and fluorescence substances during chip microfabrication and application. Microarray/chip supports allow highly parallel reactions, as well as significant increase in accuracy
■
543
of measured data (Schena, 2001). These advantages maintain high sample concentrations and fast hybridization kinetics. In general, glass substrates are characterized by high reproducibility of the fluorescence signal during hybridization of immobilized oligonucleotides with fluorescence-labeled samples due to low autofluorescence and thus high signal-to-noise ratio. They also show high reaction stability over extended time periods. Borofluorate glass substrates are particularly suitable because of their low alkali content and therefore low autofluorescence. Spotters, Growers, and Grabbers Spotting Probe immobilization can be achieved using spotters for creating the arrays. The spatial resolution of the robotics determines the density of the array. Mechanical pins (contact printing) and ink-jetting usually gives spots of 100 m in diameter. The pin diameter and shape, solution viscosity, and substrate characteristics lead to some variation in spot size, shape, and concentration of solution transferred. A number of companies offer robotic systems for spotting arrays, but only commercially available DNA chips will pack as many spots as possible onto a chip. Modern inkjet-like printers reduce the spot volume into the picoliter range. Efficient covalent immobilization of biological molecules on solid supports is a crucial step during microarray production (Pirrung, 2002). Biomolecules need to be immobilized firmly enough to prevent their replacement during reaction and hybridization steps, but flexibly enough to allow conformational changes and binding of target molecules. An advantage of spotting of prefabricated oligonucleotides onto solid supports (substrates) followed by chemical immobilization is the independence from the length of oligonucleotides to be immobilized. Oligonucleotides of 20–70 nucleotides in lengths are usually aminomodified and covalently bound onto the support. For glutaraldehyde immobilization, aminated slides and 5-aminated oligonucleotides may be used (Pease et al., 1994). Glutaraldehyde-mediated coupling of biomolecules to solid supports is the classical chemical immobilization method. In order to covalently attach oligonucleotides to the chip, preactivated or surface-modified solid supports, homo- or hetero-bifunctional crosslinkers, and modified oligonucleotides are needed. Table 48.3 surveys common immobilization techniques (see also Wittmann 2005). To immobilize modified DNA, glass slides with reactive aldehyde, amino, mercapto, or epoxy groups are commercially available. Another covalent and direct immobilization of DNA on glass with oxidic surface (Jung et al., 2001) requires positioning of DNA droplets on a heated surface with an efficiency of 150–300 fmol/mm2 during the coupling reaction. Lindroos et al. (2001) compared six different commercial slides with respect to fluorescence background, turnover efficiency and signal-to-noise ratio. They found that attachment chemistry affects genotyping accuracy when mini-sequencing is used for genotyping on microarrays. Genotyping results were best when mercaptosilane-coated slides were used to attach disulfide-modified oligonucleotides.
544
CHAPTER 48
TABLE 48.3
■
Viral Chip Technology in Genomic Medicine
A selection of DNA immobilization techniques on microarrays/chips (see Striebel et al., 2003)
Immobilization
Examples
Non-covalent complexation
Biotin–avidine/streptavidine
Self assembled monolayer
Alkanethioles, mercapto- or epoxysilane-compounds
Inclusion/ crosslinking
NH2-dextrane matrix, aminoethylcellulose, polyacrylamide
Polymerization
Co-polymerization of pyrrole and oligo-nucleotides with pyrrole residue at 5-ends, pyrimidinepyrimidine-dimers, vinylsubstituted nucleotides
Silanization by bifunctional reagents
Aminopropyltriethoxysilane (aminofunction with succinic anhydride convertible into carboxic acid function), 3-mercaptopropyl-triethoxysilane (disulfide bridge/ thioether), glycidoxypropyltriethoxysilane, p-amino-phenyltrimethoxysilane (ATMS)/diazotation chemistry and unmodified oligonucleotides
Covalently by linkerphotoactivated crosslinking
3-Hydroxyfunction, 5phosphate residue, bromcyan, cyanurchloride, carboxylfunction, aldehyde, primary amine, DNA coupling by thymidines, DNA coupling via alkylamines, photolinkage (i.e., psoralen, p-nitrophenyl-3-diazopyruvate)
Diazotized chip surfaces are described for immobilization of unmodified oligonucleotides (Dolan et al., 2001). Using paminophenyltrimethoxysilane (ATMS), diazotization chemistry was developed, and microarrays were fabricated and analyzed. The method produced uniform spots containing equivalent or even larger amounts of DNA than those obtained by commercially available immobilization techniques Oligonucleotides with hairpin stem-loop structure and multiple phosphorothioate moieties in the loop were used to anchor the oligonucleotide to glass slides that are pre-activated with bromoacetamidopropylsilane (Zhao et al., 2001). Consolandi et al. (2002) describe two robust procedures for oligonucleotide microarray preparation based on polymeric coating. Chemical procedures include a glass functionalization step with -aminopropyltriethoxy-silaneAPTES, or poly-L-lysine, or polyacrylic acid-polyacrylamide co-polymer, which is covalently bound to the modified glass. A surface activation step allows attachment of amino-modified oligonucleotides. Results show high loading capacity, good uniformity, ready availability of immobilized DNA to hybridization targets, and stability to thermal cycles. The AR Chip Epoxy obtained from two competitors (3D-LinkTM and Easy Spot) was compared with respect to slide autofluorescence, spacer length and signal-to-noise ratio (Preininger and Sauer, 2003). The two chip surfaces were assayed by hybridizations with the same targets, and under the same hybridization and scanning conditions. When polyamide (PAMAM) dendrimers containing 64 primary amino groups were used as linkers, signal intensities could be increased significantly compared with amino and epoxy silanized surfaces (Benters et al., 2002). It was shown that dendritic PAMAM linker systems reveal high immobilization efficiencies for amino-modified DNAoligomers. This was used to assay the performance of dendrimerbased DNA microarrays for discrimination of SNPs. Aminated glass slides are the most common functionalized supports available. The slides may be used for biomolecule immobilization by reaction of the amino groups with glutaraldehyde,
which in turn is coupled to amino groups of target molecules in a sandwich-like manner. Resulting Schiff bases need to be reduced to secondary amines. This may be performed by sodium cyanoborohydride (Birch-Hirschfeld et al., 2002). The method has disadvantages such as complex storage conditions for chemically fully active glutaraldehyde and low biomolecule binding capacity. In addition, glutaraldehyde coatings often do not meet the criteria for low fluorescence background due to impurities and artifacts acquired during production. Another immobilization chemistry is based on epoxycoated slides. For preparation, activated glass slides are coated with 3-glycido-oxypropyltrimethoxysilane (GOPS), dissolved in toluene. When low-fluorescence glass substrates are used, the resulting microarray supports show greatly reduced fluorescence background, combined with superior DNA immobilization properties (Földes-Papp et al., 2004; Striebel et al., 2003, 2004). A hetero-bifunctional photoreactive cross-linking reagent, 4-nitrophenyl 3-diazopyruvate (DAPpNP), reacts with glass slides bearing amino groups (Földes-Papp et al., 2004; Striebel et al., 2003). After design and spotting of NH2-modified probe oligonucleotides, the DAPpNP-coated glass slides are irradiated at 360 nm.This UV-radiation leads to conversion of diazogroups into reactive ketene groups, which react with amino groups of the DNA (Goodfellow et al., 1989; Harrison et al., 1989; Kalachikov et al., 1992). In aqueous solutions, ketene groups formed by UV-irradiation of diazopyruvic acid are transformed into carboxyl groups, which then may be used to react with 5amino-modified 2-deoxynucleotide oligomers in the presence of carbodiimide (Penchovsky et al., 2000). A further light-driven oligonucleotide immobilization technique applies psoralen-mediated covalent coupling of complementary DNA strands (Kittler and Löber, 1995; Pieles and Englisch, 1989). In a first step, an oligonucleotide of 15 alternating adenine and thymidine nucleotides is synthesized directly on a glass support.The sequence carries an amino group at its 3-end and a psoralen group at its 5-end. Psoralen may be added using
Microfabrication
2-[4-(hydroxymethyl)-4,5,8-trimethylpsoralen]-hexyl-1-O[(2-cyanoethyl)-(N,N-diiso-propyl)]phosphoramidite. A strand, which is complementary to the synthesized support oligonucleotide carrying a stem of five thymidines plus the favored cDNA sequence at its 3-end, is then hybridized to the support. In an irradiation step (2 min at 280 nm), the psoralen function of the support oligonucleotide is bound covalently to the complementary sequence. Microarrays generated that way may be hybridized to single- or double-stranded, Cy3-labeled probe DNAs. Advantages of the psoralen-based technique are: oligonucleotides may be synthesized completely in the standard 3 to 5 direction; densities of immobilized oligonucleotides are optimally suited for subsequent hybridizations, and the light-driven coupling process may be performed in an aqueous environment without impairing bases or other parts of the DNA carrying the cDNA sequence (Földes-Papp et al., 2004). Growing In addition to spotting, oligonucleotide/DNA sequences can be grown directly on the microarray/chip. Photolithography opened the way for in situ fabrication of defined oligonucleotide arrays (Fodor et al., 1991). The oligonucleotide sequence grows on the surface of a glass wafer in a manner similar to conventional solid-phase oligonucleotide synthesis but modified to include a light-sensitive deprotection step. Masks are used to add photospecific bases to selected points on the chip to create a series of oligo-nucleotides with a variety of different sequences. The process, which comprises solid-phase synthesis, photolithography, and affinity labeling, allows the synthesis of up to 250,000 different oligonucleotides (or oligopeptides) per square centimeter on a glass slide by light-directed, spatially addressable chemical synthesis with spot sizes of about 20 m in diameter. Optimal spacer lengths are crucial for hybridization behavior and fluorescence yields. Different spacer compositions to reduce steric interference of immobilized oligonucleotides with the support were studied in order to improve hybridization behavior (Shchepinov et al., 1997). Optimal spacer length was determined to be at least 40 atoms in length, yielding up to 150-fold increase in hybridization. These spacers are composed of a variety of monomeric units that are synthesized using phosphoramidite chemistry, by condensation onto an amino-functionalized polypropylene support (Matson et al., 1994). Steric hindrance during hybridization may also become problematic, if the immobilized oligonucleotides are too close to each other. Surface coverage was varied using a combination of cleavable and stable linkers. Results showed that highest hybridization yields were obtained from surfaces containing approximately 50% of the maximally possible oligonucleotide concentration. High-density oligonucleotide arrays are one of these tools for large-scale hybridization. With their ability to produce global views of genome sequences and activity, they have emerged as a key analytical research tool. A vast amount of data must be collected and analyzed to understand biological functions. Affymetrix Inc. creates breakthrough tools for genomic applications. The unique combination of photolithography and
■
545
combinatorial chemistry eliminates the need for individual laboratories to produce and test their own chips. Using technologies adapted from the semiconductor industry, GeneChip™ manufacturing begins with a 5-inch square quartz wafer. Initially the quartz is washed to ensure uniform hydroxylation across the surface. Because quartz glass is naturally hydroxylated, glass provides a very good substrate for the attachment of chemicals such as linker molecules that are later used to position the probes on the chips. The wafer is placed in a bath of silane that reacts with the hydroxyl groups of the quartz glass surface, and forms a matrix of covalently linked molecules. The distance between these silane molecules determines the probes’ packing density, allowing arrays to hold over 500,000 probe locations, or features, within a mere 1.28 cm2. Linker molecules attached to the silane matrix provide a surface that may be spatially activated by UV light. Probe synthesis occurs in parallel and results in the addition of an A, C, T, or G nucleotide to multiple growing chains simultaneously. To define which oligonucleotide chain will receive a nucleotide in each coupling (growing, propagation step), photolithographic masks, which carry 18–20 square micron windows corresponding to the dimensions of individual features, are placed over the coated wafer. The windows are distributed over the mask based on the desired sequence of each probe. When UV light is shone over the mask in the initial step of oligonucleotide synthesis, the exposed linkers become deprotected and are then available for nucleotide coupling. Critical to this step is the precise alignment of the mask with the wafer. To ensure that this critical step is accurately completed, chrome marks on the wafer and on the mask are perfectly aligned. Once the desired features have been activated, a solution containing a single type of deoxynucleotide with a removable protection groups is flushed over the wafer’s surface. The nucleotide attaches to the activated linkers, initiating the synthesis process. Although the synthesis process is highly efficient, some activated molecules failed to attach the new nucleotide. To prevent these “outliers” from becoming probes with missing nucleotides, a capping step truncates them in each round of oligonucleotide synthesis. In addition, the side chains of the nucleotides are protected to prevent the formation of branched oligonucleotides. In the next round of synthesis, another mask is placed over the wafer to allow the next cycle of deprotection, coupling and capping. The cyclic process is repeated until the probes reach their full length, usually 25 nucleotides. Although each position in the sequence of an oligonucleotide can be occupied by one of four nucleotides resulting in an apparent need for 25 4 100 different masks per wafer, situations are identified when the same mask can be used repeatedly. Once the synthesis is complete, the wafers are deprotected and diced. Disadvantages of this method are oligonucleotide lengths that are limited to approximately 30 nucleotides, as well as an unfavorable oligonucleotide orientation, where the 3-end is bound to the solid support. This is important since hybridization efficiencies depend on orientation and accessibility of the immobilized probe (Southern et al., 1999). A further disadvantage of photolithographic array generation is the accumulation of truncated sequences that
546
CHAPTER 48
■
Viral Chip Technology in Genomic Medicine
cannot be removed (fidelity of the synthesis process: FöldesPapp et al., 1994). For optimal performance, for example, of a photolithographically generated microarray, requirements are therefore (Földes-Papp et al., 1994): (i) Minimization of failure sequences during chemical synthesis of oligonucleotides. This may be achieved by keeping side reactions of chemical synthesis low. Side reactions comprise clustering, depurination in the detritylation reaction, substitution of nucleotides at nucleobases, or breaking of internucleotide bonds. (ii) Minimization of homologous sequences. Oligonucleotides become increasingly more similar in their physico-chemical properties with increasing chain lengths. Therefore, sequences of the type N-1, N-2, N-3, and so forth (N stands for the number of nucleotides in the projected probe oligonucleotide/DNA sequence) are to be minimized by optimization of the coupling and capping reactions. (iii) Suppression of incomplete reactions during chemical cycles. Incomplete reactions are due to masking of non-converted ends by degradation of protection groups. (iv) High reproducibility of chemical syntheses within single steps. The chemical criteria “yield” (e.g., trityl yield) and “coupling and capping efficiency” used in practice do not meet the demands of the quantitative characterization of chemical oligonucleotide/DNA/RNA/peptide syntheses. Yields, coupling and capping efficiencies only relate to single-reaction steps, but do not consider the length N of projected probe sequences; therefore, these parameters do not say anything about the frequency distribution of failure (truncated, false, erroneous) sequences, and they do not express the dynamics with multi-cyclic syntheses. If syntheses of different nucleotide lengths are compared regarding the optimization of external and internal synthesis conditions, then a lower yield of achieved full length sequences does not mean poorer synthesis conditions. Synthesis conditions leading to more probe sequences of lengths N are better optimized if fewer false sequences are synthesized (Földes-Papp et al., 1995a, b, 1996, 1997a, b, 1998). From theoretical considerations, exact measures of cyclic multistep syntheses, such as chemical oligonucleotide/DNA/RNA/peptide syntheses, were derived. The studies showed that fractal dimensions are such measures for oligonucleotide/DNA/RNA/peptide syntheses. They are superior to commonly used parameters like trityl yield and coupling efficiencies, and they can be applied to the experimental separation of crude synthesis products by high-resolution ion-exchange HPLC, capillary electrophoresis, and gel electrophoresis. Based on experiments, the dynamics of the syntheses on solid support was modeled by stochastic processes with nucleotide chain length N, propagation/elongation (coupling, oxidation), and termination (capping) probabilities as the basic parameters. The following Eqn. 48.1 provides an exact measure for quantitative assessment and comparison of multi-cyclic syntheses of different target length, such as chemical oligonucleotide/DNA/RNA/ peptide syntheses:
D( N , d0 ) 2 N
d0N 1 1 ln 1 N 1 d0 d0
(48.1)
Here, N is the projected length of the probe sequence, and d0 the average (constant) coupling (propagation/elongation) probability that can be expressed, for example, as averaged trityl yield in the case of chemical oligonucleotide/DNA syntheses (Földes-Papp et al., 1995a, b). Oligonucleotide syntheses of different projected probe lengths are equally well optimized regarding their synthesis conditions, if the fractal dimensions D D(N, d0) are equal; syntheses showed less propagation and accumulation of erroneous sequences with respect to their projected target lengths N, which is formulated by Eqn. 48.1 and experimentally tested and validated (Földes-Papp et al., 1995a, b, 1996, 1997a, b, 1998). Grabbing A grabber is an electronic addressing technology. A chip is flooded with a solution of a probe, and a charge is introduced onto the surface by electrical activation of a row or spot on the chip. The oligonucleotide/DNA probe concentrates close to the charge and is then chemically bonded in place. The chip is washed and another probe solution is introduced. One mm2 of the microarray/chip contains 25 up to 400 wire-bonded electrodes, or even 10,000 sites.
Labeling Strategies and Fluidics Workstations Once the DNA microarray/chip is fabricated, some form of chemistry has to occur between the sample and the array. The simplest approach is direct labeling of the viral target DNA with fluorophores. Common methods employ enzymatic incorporation of fluorescent nucleotides, or PCR amplification with fluorescent primer pairs. RNA probes are usually prepared from cloned DNA by incorporation of fluorescent nucleotides via RNA polymerase. Selection of fluorescence dyes for primer labeling depends on the filter system of the microarray reader and scanner, respectively. As fluorescence dyes are large hydrophobic molecules, care has to be taken to avoid intercalation of these dyes during PCR (Földes-Papp et al., 2001a, b). Suitable dyes that provide a good long-time stability, such as cyanide dyes Cy3 or Cy 5, are commercially available. Cyanide dyes show an intensive fluorescence and are all water-soluble. This property, combined with their good quantum yield, makes them superior to many other fluorescence dyes (Wessendorf and Brelje, 1992). Usually, the dyes are attached to the respective DNA sequences in the form of their N-hydroxy-succinimide (NHS) esters. The attachment requires an amino group at the 3 or 5ends of the DNA, which may be introduced via an amino link. Additionally, the method allows introduction of spacers of different lengths (Földes-Papp et al., 2001a, b). Optimal spacer lengths were obtained using C6-amino link reagents. If solvents and pH values are appropriately adjusted (reactions of some fluorescence dyes require heterogeneous phases), covalent fluorescence dye coupling may be achieved within 1 h. Subsequently, an HPLC purification step is required.
Microfabrication
An alternative to NHS ester coupling is coupling of some fluorescence dyes in the form of their phosphoramidite derivatives. Dye-phosphoramidites are easily attached covalently to DNA 5-ends in computer-assisted automatic oligonucleotide synthesis without any difficulties. After cleavage from the support and chemical removal of protection groups, purification is carried out by preparative Rp18-HPLC. Incorporation of fluorescent dyes, such as Cy3, Cy5, or fluorescein into nucleic acids, can also be performed with their appropriate 2-desoxyuridine-5-triphosphates. These commercially available compounds carry the respective dyes that are attached to spacers at nucleotide 5-positions. With this internal incorporation of the dye-labeled nucleotides, false base pairing cannot be excluded during PCR reactions. In particular, this holds true for multiplex PCR reactions. Additionally, these dyetagged nucleotides need to be added in excess, since they are less well inserted by DNA-polymerases than naturally occurring nucleotides (Földes-Papp et al., 2001a, b). A serious problem of viral diagnostics is the variable, often small quantity of viral particles in patient samples. The amount may vary between one and several thousands of virus copies per milliliter. Signal intensities may be amplified by adding dendrimer building blocks at 5-ends of PCR primers, allowing multiple coupling of fluorescence dyes per primer (Striebel et al., 2004). The review article of Caminade et al. (2006) focuses on the improvements that hyperbranched and perfectly defined dendrimers also called nanomolecules can provide to DNA microarrays. Two main uses of dendrimers for such purpose have been described up to now (Caminade et al., 2006). Either the dendrimer is used as linker between the solid surface and the probe oligonucleotide, or the dendrimer is used as a multilabeled entity linked to the target oligonucleotide. In the first case the dendrimer generally induces a higher loading of probes and an easier hybridization, due to moving away the solid phase. In the second case, the high number of localized labels (generally fluorescent) induces an increased sensitivity, allowing the detection of small quantities of biological entities. Using dendrimer technology for multilabeling of viral DNA, Striebel et al. (2004) incorporated dendrimer building blocks at DNA ends during oligonucleotide synthesis, and then fluorescence dyes coupled to amino groups at dendrimers’ branches via NHS ester chemistry. In this way, two, four, or eight fluorescence dye molecules may be attached covalently to a primer, using “doubler” dendrimer building blocks. By using “trebler” dendrimer building blocks, three or nine fluorescence dyes may be linked to a primer. In oligonucleotide chemistry, branched oligonucleotides are already being used for signal amplification in hybridization assays. Branched oligonucleotides allow detection of a concentration below 100 human CMV copies per milliliter sample (template) in the case of 5-(Cy3)3-dendrimer-labeled CMV DNAs (Striebel et al., 2004). The fluorescence signal was enhanced via the dendrimers up to 30 times compared with the quenched single Cy3-labeled human HSV-1 DNA. The on-chip signal-amplifying effect depended upon the number of branches and the concentration of fluorophore-labeled pathogenic DNAs.
■
547
Treblers were superior to doublers, as trebler-labeled nucleic acids had fluorescence-signal-enhancing effects over a broad range of labeled DNA concentrations exemplified for the quenched single Cy3-labeled HSV-1 and non-quenched single Cy3-labeled CMV DNAs. Fluorescence dye labeling invariably involves hybridization of the sample to the oligonucleotides/DNA present on the array. Because chip-bound DNA is often quite short and varied in sequence, a single stringent hybridization condition that is optimal for every spot on the chip is impossible. However, it is possible to find temperature and salt conditions that gives acceptably strong signals for the desired hybridization products and much weaker signals for mismatches. Unbound viral target DNA is then washed away, and the chip is ready to be scanned. Fluidics workstations are available to automate some or all of the wet working steps. The traditional array detection strategy was autoradiography of radiolabeled samples, but other options are available including electronic signal transduction. With bioelectronic detection, electron transfer reactions from the DNA to the substrate allow fast and probe-cell specific detection systems. The ability of mass spectroscopy to identify and, increasingly, to precisely quantify genes and genome sequences (as well as proteins) from complex samples can be expected to impact broadly on biology and medicine. The systematic analysis of the much larger number of genes expressed in a cell, an explicit goal of functional genomics, is now also rapidly advancing, due mainly to the development of new experimental approaches. Using MALDI-TOF, DNA fragments are analyzed with the SpectroChip™ from Sequenom Inc. A piezoelectric gene sensor microarray for HBV quantification in clinical samples was constructed using crystal units that oscillate independently (Chen et al., 2005a, b). Imaging and Scanning, Data Processing Scanning a fluorescence-labeled DNA array is conceptually quite simple. A light source excites the labeled samples and a detector system measures and records the emitted fluorescence. However, the instrumentation requirements differ according to the precise nature of the array. Most image capture instruments and microarray readers, respectively, use a scanning detector similar to line-scanning detector systems for DNA sequencing instruments. The number of discrete points that the detector can sample across the array and the row-to-row step interval determine the size of the features that the detector can image. It is important that the detector pixels are sufficiently small to gather data from enough significant pixels in each probe cell. Confocal laser-scanning systems for fluorescent microarray biochips should have a spatial resolution of a detector on the order of 1/8 to 1/10 of the diameter of the smallest element in the microarray. While the absolute number of DNA molecules that can be crammed into a small space will probably prevent the field from reaching the 0.2 m features found in semiconductors, 5 m probe cells appear attainable. This would put about four million features on a 1 cm2 chip.
548
CHAPTER 48
■
Viral Chip Technology in Genomic Medicine
Image analysis is performed by placing a grid over the array to allow integration of the signal from each probe cell. If the signals from the arrays are strong and fairly consistent, a threshold algorithm is usually adequate, but weak signals demand a computationally intensive approach. The background signal or baseline is the fluorescence from the support matrix. Raw fluorescence intensity data are background subtracted. Hybridization outcomes can be represented as log2 fluorescence intensity ratio of test versus reference samples. To virtually enrich viral specific concordant probes, data filtration can be further filtered by elimination of spots below signal intensity and a predefined absolute log2 ratio; this eliminates outlier data from further analysis due to labeling bias. The processing improves data correlation. Hierarchical clustering and visualization of data can be performed using Gene Cluster and Tree View software (Stanford University, USA). High-density DNA chips with thousands of probe cells need special managing software for microarray probe tracking and data analysis like Affymetrix’s LIMS workstation, BioDiscovery’s Clone Tracker software or GeneSpring, a software package offered by Silicon Genetics. Based on observed microarray hybridization patterns an algorithm called E-Predict is reported for microarray-based viral species identification. E-Predict compares observed hybridization patterns with theoretical energy profiles representing different viral species in a set of clinical samples but its relevance to other metagenomic applications is discussed (Urisman et al., 2005). Viral Genotyping Regardless of how the microarrays and chips, respectively, are made, viral chip technology is superior to conventional PCR methods for the fast, sensitive, specific, and parallelized diagnostics of heterogeneous populations of different viral species and strains (Li et al., 2001; Striebel et al., 2003). Viral genotyping is an approach made possible by the availability of viral gene and genome sequence databases and technical and conceptual advances in chip technology. Interest exists in providing highthroughput platforms and readout technologies that can selectively examine changes in the transcript levels of specific genes in response to multiple treatments. Although conventional microarrays (DeRisi et al., 1996; Schena et al., 1995; Wang et al., 2002) have, in principle, the necessary throughput, they were limited with respect to cost and dynamic range and therefore are not well suited for the task of producing genotyping data for multiple viruses from which accurate reliable information can be calculated. For routine clinical testing, it is not possible to test thousands of genes on a microarray. Therefore, some researchers have called for “less” genes to be put on the array, because most of the other data points are inconsequential (Wooster, 2000) or even detrimental to the usefulness of a set of classifier genes (Draghici et al., 2003). Can array data be used to identify a handful of critical genes that will lead to a more-detailed taxonomy and can this or similar array data be used to predict clinical outcome (Wooster, 2000)? In genotyping applications, one looks for sequence variation that has previously been characterized. Oligonucleotides of known sequence variants of a viral gene or collections of viral
genes are represented in the chip. Genotyping arrays that are designed to examine viral DNA sequences are listed in Table 48.4. The viral chip has the enormous strength that only one chip is needed to analyze a mixture of different viral DNA sequences. In practice, there may be some pitfalls, as exemplified by long-term variability studies of the human herpes simplex virus type 1 (HSV-1) genome. The HSV-1 genome consists of iterative DNA sequences of variable amounts. The HSV-1 genome from a HSV-1 DNA pool has an isomolar genome organization with L and S components. Genome complexity is due to tetra-isomers derived from the rolling circle-like intracellular replication process. In the L-S region, transient variabilities are formed by insertions or deletions of the repetitive a-sequences. The genomic changes are more or less strain-specific. Outside the L-S junction the virus progeny under study showed no or negligible deviations from the parent strain. The viral DNA material of three HSV-1 strains F (Roizman), Mst (Ulm) and AK (Ulm) was sampled after intracellular self-assembled virus particles were formed (Földes-Papp et al., 1997a; Klauck et al., 1995). The DNA replication was carried out in Vero cell infections and the purified viral DNA was digested with BamHI restriction endonuclease (Földes-Papp et al., 1997a). The resulting BamHI-K2-fragments of HSV-1 were extracted from agarose gels by electroelution, ligated into plasmid pBR322 or pAT153 and cloned in E.coli K 12 C600 cells for hybridization in Southern blots. The restriction patterns obtained by endonucleases KpnI, and SalI were analyzed by hybridization with K2fragments of HSV-1 (F) labeled by the random priming method with [32P] -dCTP. Blots were autoradiographed and the bands quantified by scanning densiometry. We found that the nucleotide sequence pattern of the HSV-1 genome in the L-S region can be approximated by distribution profiles of digested components expressed in powers of nucleotide length N of the largest component in the region. The largest components of interest studied had nucleotide length N of 13,500 base pairs obtained with KpnI of HSV-1 strains F, Mst, AK and nucleotide length N of 9730 base pairs obtained with SalI of HSV-1 strains F, Mst, AK. We obtained the fractal dimensions D 1.63 (strain F), D 1.66 (strain Mst), and D 1.66 (strain AK) with restriction endonuclease KpnI. With restriction endonuclease SalI, we found the fractal dimensions D 1.36 (strain F), D 1.36 (strain Mst), and D 1.39 (strain AK). The fractal dimension D is here a measure for producing internal repeats in the L-S region of strains F, Mst, AK during replication at the viral genome level (Földes-Papp et al., 1997a). Here, a lower value of the fractal dimension D indicates a distribution of sequence variability with lower amounts of internal repeats. The fractal dimension D was obtained from the scaling to the largest nucleotide length of interest, N, and is interpreted as typical longterm DNA sequence variability. Viral Sequence Determination At its most complex, viral genotyping using a DNA chip involves complete determination of the nucleotide order via resequencing by hybridization. Viral resequencing discovers
Microfabrication
TABLE 48.4
■
549
Representative examples of viral genotyping studies with microarrays/chips
Viral species and strains
Examples
Herpes viruses including Epstein-Barr virus
Mezzasoma et al. (2002); Duburcq et al. (2004); Conejero-Goldberg et al. (2005)
HSV-1, HSV-2, VZV, EBV, CMV, HHV-6
Földes-Papp et al. (2004, 2005b); Striebel et al. (2004)
VZV
Sergeev et al. (2006)
Influenza viruses
Chen and Evans, (2001); Sengupta et al. (2003); Ivshina et al. (2004); Kessler et al. (2004); Mehlmann et al. (2006); Townsend et al. (2006); Dawson et al. (2007); Dankbar et al. (2007)
Adenovirus
Lin et al. (2004)
SARS
Wang et al. (2003); Wong et al. (2004)
Noro- and astroviruses
Jääskeläinen and Maunula, 2006
Hepatitis B and C viruses
Zhao et al. (2003); Perrin et al. (2003); Duburcq et al. (2004)
Hepatitis B virus
Song et al. (2006)
Drug resistant HBV
Tran et al. (2006); Chen et al. (2005a, b)
Hepatitis C virus
Simmonds (2001); Martell et al. (2004); Daiba et al. (2004)
Human papilloma viruses
An et al. (2003); Cho et al. (2003); Klaassen et al. (2004); Oh et al. (2004); Hoffmann et al. (2004); Park et al. (2004); Delrio-Lafreniere et al. (2004); Wallace et al. (2005); Albrecht et al. (2006); Gheit et al. (2006); Son et al. (2006); Min et al. (2006); Kim et al. (2006); Seo et al. (2006); Lin et al. (2007); Luo et al. (2007)
Polioviruses
Cherkasova et al. (2003)
Orthopoxviruses including variola, monkeypox, cowpox, vaccinia
Laassari et al. (2003); Ryabinin et al. (2006)
Measles virus
Neverov et al. (2006)
Human group A rotaviruses
Chizhikov et al. (2002); Lovmar et al. (2003)
Hantaviruses
Nordstrom et al. (2004)
Flaviviruses
Nordstrom et al. (2005)
HIV
Günthard et al. (1998); Bean and Wilson, (2000); Ghedin et al. (2004); Boriskin et al. (2004)
Drug resistant HIV-1
Cherkasova et al. (2003); Duburcq et al. (2004); Gonzalez et al. (2004)
Baculoviruses
Yamagishi et al. (2003)
novel sequence variations (see Table 48.5). The longer the target DNA, the longer the oligonucleotides must be to eliminate ambiguities. For example, a viral chip was designed after dividing each viral genome into overlapping 70-mer oligonucleotides using sequencing analysis tools BLASTn (Wang et al., 2003; Wong et al., 2004). The 70-mers were ranked by the number of viral genomes to which significant homology were found by alignments. Virus resequencing of the genomes in infected humans was performed with the Affymetrix™ SARS resequencing GeneChips (Sulaiman et al., 2006, 2007). High-density resequencing GeneChips have potential biodefense applications and may be used as an alternate tool for rapid identification of smallpox virus in the future. For sequencing of any viral DNA/RNA, a gene of unknown sequence can be rapidly screened. In this approach, oligonucleotide probes represent all possible combinations of
sequence in a given length. The number of probe “cells” can be calculated by four to the power of the oligonucleotide length, that is 65,536 8-mers, 262,144 9-mers, and so forth. The viral target sequence to be sequenced is broken into small pieces, fluorescently labeled, and hybridized with the immobilized oligonucleotides on the chip. However, the random sequencing procedure is weak for non-random sequences such as direct and inverted repeats. Viral Gene Expression There is a diverse range of experimental objectives and uses for viral gene expression microarray data, which makes the areas of experimental design and data analysis quite broad in scope (see Table 48.6). As such, there are many ways to design expression profiling experiments, as well as many ways to analyze and
550
CHAPTER 48
■
Viral Chip Technology in Genomic Medicine
TABLE 48.5 Representative examples of viral resequencing studies with microarrays/chips Viral species and strains
Examples
SARS coronavirus
Rota et al. (2003); Ksiazek et al. (2003); Wong et al. (2004); Sulaiman et al. (2006)
Smallpox viruses (variola virus)
Sulaiman et al. (2007)
Coccolithovirus
Allen et al. (2007)
Human influenza virus
Wang et al. (2006)
TABLE 48.6 Representative examples of viral gene expression studies with microarrays/chips Viral species and strains
Examples
Pan-viral
Wang et al. (2002, 2003)
Human and mammalian retroviruses
Seifarth et al. (2003)
Choristoneura fumiferana nucleopolyhedrovirus
Yang et al. (2007)
Human immunodeficiency virus type 1, Human T cell leukemia virus types 1 and 2, Hepatitis C virus, EpsteinBarr virus, Human herpesvirus 6A and 6B, and Kaposi’s sarcomaassociated herpesvirus
Ghedin et al. (2004)
HSV-1 and HSV-2
Aguilar et al. (2006)
HPV18 (E2 mutants)
Thierry et al. (2004)
mine data functional genomics expression profiling experiments, including transcriptional analysis of normal biological processes of viral and host replication, discovery and validation of viral drug targets, and studies into the mechanism of action and toxicity of pharmaceutical compounds. Chambers et al. (1999) described the first viral DNA microarray for the temporal profiling of viral gene expression human cytomegalovirus, CMV. Basically, viral DNA microarrays/chips that apply a genomic approach from known open reading frames allow analysis of many virally encoded genes at the steady-state level of mRNA in the cell for a single (Kellam, 2001) or multiple pathogen detection (Wang et al., 2002). Single Pathogen Detection Viral RNA expression monitoring studied for a single pathogen identified patterns of immediate early, early, and late gene expression for some herpes viruses (Aguilar et al., 2004; Chambers et al., 1999; Stingley et al., 2000). Lytic expression profiles of viral herpes genes are established following
reactivation from latency (Jenner et al., 2001; Leenman et al., 2004; Paulose-Murphy et al., 2001). The predicted open reading frames of the human CMV and an established protocol for simultaneously measuring the expression of all CMV genes have been studied by Yang et al. (2006). In order to study the global pattern of VZV gene transcription, VZV microarrays using 75base oligomers to 71 VZV ORFs were designed and validated (Kennedy et al., 2005). A human herpes virus chip for six latent EBV genes (EBNA1, EBNA2, EBNA3A, EBNA3C, LMP1, LMP2), and four lytic EBV genes (BZLF1, BXLF2, BKRF2, BZLF2) allowed Bernasconi et al., 2006 to accurately measure EBV gene transcription changes triggered by treatment interventions. The EBV infection rate and the gene expression profile of EBV in tumor biopsies were determined using EBV genome chips (Li et al., 2006). By using Affymetrix™ human gene chip and NetAffx analysis through the Affymetrix™ website, the gene expression profiles of variant core proteins were implicated in HCV replication, pathogenesis, or oncogenesis in the Huh-7 cell line, which is useful for our understanding of HCV variant core protein biological function and its pathogenic mechanism (Dou et al., 2005). Multiple Pathogen Detection A viral detection DNA microarray composed of oligonucleotides corresponding to the most conserved sequences of all known viruses identified the presence of gammaretroviral sequences in cDNA samples that impair function of RNase L, particularly R462Q, from seven of 11 R462Q-homozygous cases, and in one of eight heterozygous and homozygous wild-type cases in human prostata tumors (Urisman et al., 2006). The novel virus, named XMRV, is closely related to xenotropic murine leukemia viruses (MuLVs), but its sequence is clearly distinct from all known members of this group. Attempts to identify differentially expressed transcripts in complex neuropsychiatric disorders, such as schizophrenia, have shown the necessity of screening a large sample set in order to rule out medication effects and normal individual variation when attempting to determine disease association (Conejero-Goldberg et al., 2005;Wang et al., 2002). The genome wide gene expression profile aids in the understanding of genes that may be regulated in a particular pathological condition-like cardiovascular disease and how it pertains to viral and parasitic infections of the heart (Mukherjee et al., 2006). Viral Chips for Several Diagnostic Purposes Recent successes with viral chips illustrate their role as an indispensable tool for molecular and cellular biology and for the emerging field of systems biology. Studies focusing on the analysis of viral populations from cells or tissues typically pose challenges owing to the high degree of complexity of a mixture of numerous viral species and the low abundance of many of the viruses, which necessitates highly parallelized analysis. The ability of viral chips to identify thousands of viruses from complex samples can be expected to impact broadly on biology and medicine (Wang et al., 2002). However, the use of viral chips
Microfabrication
for a diagnostic purpose is still limited because there is little agreement as to which portion of the viral genome is specific for diagnostics and allows establishment of intra- and inter-species relationships (Földes-Papp et al., 1997a; Striebel et al., 2003). One application is the “forecast” of the efficacy of drugs for therapy (Chambers et al., 1999; Murphy, 2002). Identification of mutations responsible for resistances of bacterial and viral particles (i.e., mycobacteria, HIV), including the proof of special resistance genes, plays an important role (Günthard et al., 1998; Hamels et al., 2001; Ramaswamy et al., 2003; Rogers and Barker, 2002; Schembri et al., 2002). Studying pathogen–host relations contributes to the understanding of disease pathogenesis and helps develop therapeutic strategies for new drugs (Cummings and Relman, 2000; Haselbeck et al., 2000; Kato-Maeda et al., 2001; Mahony, 2002; Peek et al., 2001). Since phenotype-based testing of resistance with virus isolation and growth on cell cultures is very time and labor consuming, genotype-based testing of resistance represents a meaningful alternative. In hepatitis C, duration and success of an interferon therapy mainly depends upon the HCV genotype (Zein, 2000), apart from other factors such as age, duration of infection and height of the HCV RNA titer before specific therapy starts. So far, there have been very few applications of microarrays/chips for quantitative estimations of viral loads. Preliminary studies are reported for quantification of hepatitis B virus (HBV) DNA in serum (Kawaguchi et al., 2003). Because of the high clinical importance of virus quantification, in particular for estimation of virus replication and therapy monitoring, there is a great potential for chip application. For example, a quarter of the chronically infected people suffer from SNPs in the precore/basal core promoter region of the HBV genome, causing the loss of HBeAg expression. These HBeAg-negative patients often remain asymptomatic and viremic until the alanine aminotransferase level in serum is persistently elevated (Lai et al., 2003). It would therefore be important to develop methods to detect the precore/basal core promoter mutations of the HBV genome at an earlier stage of infection for large-scale clinical and diagnostic analysis. The conventional and common methods for detecting mutations such as single-strand conformation polymorphisms, denaturing gradient gel electrophoresis, direct sequencing and chemical cleavage are not practical for detecting multiple SNPs in high-throughput format. The arrayed primer extension method (APEX), which is based on a hybridization of short single-stranded DNA templates with an array of immobilized oligonucleotide probes on a glass surface followed by a polymerase-assisted incorporation of different fluorescently tagged dideoxynucleotides (Tönisson et al., 2000), enables large-scale SNP detection. All methods of mutant detection so far described can only identify mutants in the serum, and cannot determine the proportion of those mutants. Li et al. (2005) reported the development of a novel technique that can quantify the relative proportion of mutants in serum utilizing gene microarray technology. Clinical applications of microarray technology predominantly refer to collections of virus variants and to differentiation of subtypes. A pan-virus DNA microarray (Virochip) was used to detect
■
551
a human metapneumovirus (hMPV) strain associated with a critical respiratory tract infection in an elderly adult with chronic lymphocytic leukemia (Chiu et al., 2007). Using the Virochip, human parainfluenzavirus 4 (HPIV-4) infection was detected in an immunocompetent adult presenting with a life-threatening acute respiratory illness (Chiu et al., 2006). So far, viral chip technology has no practical applications in clinical diagnostics due to its high costs and insufficient standardization; it so cannot be compared to biological chip experiments. This technology thus only can be a useful supplement to other diagnostic methods (Wang et al., 2002). The main focus of clinical diagnostics with viral DNA chips will be to find answers to a handful of questions about viruses safely, quickly and inexpensively. Clinical diagnostics approaches using viral DNA chip technology simply ask the question whether particular genes, which are indicative of particular viruses, are present in a sample or not. For example, the knowledge-based, low-density “focused” microarray is based on the molecular diagnostics of six human herpesvirus types: HSV-1, HSV-2, VZV, EBV, CMV, HHV-6 (Földes-Papp et al., 2004, 2005b; Striebel et al., 2004). These studies attempted to optimize parameters of chip design, surface chemistry, oligonucleotide probe spotting, sample labeling, and DNA hybridization. The well-designed and tailored chip platform developed utilizes low-fluorescence background coverslips, epoxy surface chemistry, standardized oligonucleotide probe spotting, PCR-labeling with Cy3 of isolated viral DNA, array hybridization, and detection of specific spot fluorescence by an automatic microarray reader. Korimbocus et al. (2005) reported on the simultaneous detection of major CNS pathogens HSV-1, HSV-2, CMV; all serotypes of human enteroviruses, and five flaviviruses (West Nile virus, dengue viruses, and Langat virus) using amplification by PCR and detection of amplified products by a DNA microarray. Consensus primers for the amplification of all members of each genus and sequences that are specific for the identification of each virus species were selected from the sequence alignments of each target gene. A multiplex PCR-based DNA array for EBV, CMV, and Kaposi’s sarcoma-associated herpesvirus (KSHV) may serve as a rapid and reliable diagnostic tool for clinical applications (Fujimuro et al., 2006). The macroarray test of Fitzgibbon and Sagripanti (2006) for capturing generic members of the orthopox- or alphavirus families and a collection of additional oligonucleotides to bind specifically nucleic acids from five individual alphaviruses, including Venezuelan equine encephalitis, or DNA from each of four orthopoxviruses, including variola virus (VAR) is easy to perform, inexpensive, relatively fast, uncomplicated to interpret, and its end point is read visually without the need of additional equipment. This nucleic acid hybridization assay onto nylon membranes in macroarray format can help in detecting or excluding the presence of threat viruses in environmental samples and appears promising for a variety of biodefense applications. A microarray-in-a-tube system, for example, in an eppendorf-tube format, was first suggested by Földes-Papp et al. (2005a, b). It is a solution to complicated handling and expensive microarry equipment. Recently, a microarray-in-a-tube
552
CHAPTER 48
■
Viral Chip Technology in Genomic Medicine
system with a 5 5 oligonucleotide “microarray” was developed by Liu et al. (2007), for detecting four respiratory tract viruses (severe acute respiratory syndrome-associated coronavirus, influenza A virus, influenza B virus, and enterovirus) in a specially designed Eppendorf cap with a flat, optically transparent window.
NANOFABRICATION Nanofabrication combines different metals, semiconductors, or carbon in spherical particles as quantum dots and in linear particles as nanowires, nanotubes, or nanorods. Nanofabrication reduces the amount of biological samples and reagents for the detection of viruses. A number of assays of metal nanoparticles for DNA detection and analysis and their basic principle, advantages, disadvantages and detection limits are very recently reviewed (Möller and Fritzsche, 2007). The detection schemes can aptly be divided into optical category such as quantum dots, stripped nanovirus and electrical/electrochemical category such as semiconducting nanowires and carbon nanotubes (Rosi and Mirkin, 2005). However, I distil the salient features of nanofabrication for viral detection. Quantum Dots Quantum dots are also known as nanocrystals or artificial atoms. They consist of clusters of a few hundred to a few thousand atoms. Their diameter is in the nanometer range. Quantum dots (QD) are made from metals such as gold, silver, cobalt, and semiconductor materials such as cadmium sulfide, cadmium selenide, or cadmium telluride (Gerion et al., 2003). QDs are characterized by broad absorption spectra from the ultraviolet to the far infrared, good photostability, and a narrow emission spectrum. These properties could improve the sensitivity of biological detection and imaging by 10- to 100-fold. Conjugated gold nanoparticles with nucleic acids were used as labels for DNA microarrays (Taton et al., 2000). With this method, target nucleic acids at concentrations of 500 femtomolar were detected (Bao et al., 2005; Liang et al., 2005; Park et al., 2002). QDs have also been applied to the detection of viral species (Perez et al., 2003). To detect specific DNA sequences without first performing an amplification step, Wang and colleagues developed an ultrasensitive nanosensor that uses QDs linked to DNA probes (Zhang et al., 2005). The nanosensor can detect less than 50 copies of DNA and has much less background fluorescence than conventional fluorescence resonance energy transfer (FRET) systems. The nanosensor includes two oligonucleotide probes that bind to separate regions of an assayed strand of DNA (target DNA). The reporter probe is labeled with a Cy5 fluorophore, and the capture probe is labeled with biotin. The probes bind the target DNA, and a streptavidin-coated QD binds to the capture probe. This process brings the Cy5 molecule close to the QD for FRET to occur. Because QDs have broad absorption and narrow emission spectra, the QD-based nanosensor system
can be finely tuned so that background fluorescence is negligible. QDs also concentrate the FRET signal by binding many target–probe complexes. To detect FRET signals, the researchers developed a novel confocal fluorescence microscopy platform. The QD–target–probe complexes were continuously flowed through a microcapillary past two detectors that were specific for signals from either FRET donors or acceptors. When the system was tested with a single-stranded target DNA, the fluorescence was observed with both detectors. However, when the target was absent or when a non-complementary DNA strand was used, only donor fluorescence was observed. Wang and colleagues also compared the performance of the QD nanosensor with that of a molecular beacon, which is typically used in FRET assays. The QD nanosensor was about 100 times more sensitive. Finally, the researchers combined the nanosensor with an oligonucleotide ligation assay. This method allowed the discrimination of point mutations in DNA samples from ovarian cancer patients. Stripped Nanowires These can be made of metals such as Au, Pt, Ni, Co, or Cu by chemical vapor deposition or more commonly by electrodeposition. The length of the stripped nanowires depends upon the electrochemical deposition time for each metal band (Nicewarner-Pena et al., 2001). Nanowires consisting of two metals and 13 strips are metallic barcodes in biological assays with up to 4160 distinguishable patterns. Stripped nanowires of three metals yield 8.0 105 different tags (Keating, 2003). On the surface of nanorods, the barcoded nanowire molecular beacon approach with a quenched fluorophore was used to detect DNA binding of human immunodeficiency virus (HIV), hepatitis B (HBV) and hepatitis C (HCV) virus sequences. Semiconducting Nanowires They are building blocks for nanoscale electronics. For example, they are set up for field-effect transistors for the real-time detection of DNA. The hybridized nucleic acid is used as substrate for the deposition of metal and construction of a nanowire. There are several nucleic acid metallization chemistries available. Hybridization between target DNA and the probe bound to the nanowire or a support followed by metallization of nucleic acid changes conductance. The hybridized and metallized nucleic acid is electrocatalytically detected by applying voltage to one of the two electrodes in each test structure and measuring the increased current of a redox indicator or the changes in conductivity or capacity (Jianrong et al., 2004). Semiconductorbased oligonucleotide microarrays were used for rapidly identifying influenza A virus hemagglutinin subtypes 1 through 15 and neuraminidase subtypes 1 through 9 (Lodes et al., 2006). Silicon Nanowires Silicon semiconducting nanowires transport electrons and holes (Cui and Lieber, 2001; Cui et al., 2001). They can be synthesized by chemical vapor deposition, electrochemical size reduction, and
Are There Additional Alternatives to Diagnostic Microarrays?
pulsed laser vaporation. Slicon nanowires with 20 nm gold clusters as catalysts, silane as reactant, and diboran as p-type dopant are label-free, electrochemical biosensors that were used for the detection of herpesvirus type 1 and 2 DNA (Patolsky et al., 2004). With an antibody functionalized nanowire, single influenza A viruses could also be detected. Carbon Nanowire The graphitic surface chemistry and electronic properties depend upon the tube diameter and chirality. Carbon nanowires can be metallic or semiconducting and yield semiconductor– semiconductor and semiconductor–metal junctions. They can be functionalized with nucleic acids, proteins and antibodies (Cai et al., 2003).
■
553
ARE THERE ADDITIONAL ALTERNATIVES TO DIAGNOSTIC MICROARRAYS? For phenotype screening in functional genomics, live-cell chips are powerful new tools. For example, a cell-based microarray on primary and cancer cells based on the localized reverse infection by retroviruses (Carbone et al., 2007). Viral vectors are immobilized on a nanostructured titanium dioxide (ns-TiO2) film obtained by depositing a supersonic beam of titania clusters on a glass substrate. We validated the retroviral cell array by overexpression of Green Fluorescence Protein reporter genes in primary and cancer cells, and by RNA interference of p53 in primary cells by analyzing effects in cell growth.
Stochastic SPSM-FCS experiment
ln {N P(X 1, T C)} ln {N P1} ln C C ln {P(X 2, T C)} ln {P2} 2 ln C ln 2 C (Criterion 3) N : Absolute number of molecules measured in the detection volume with N 1 NA:Avogadro‘s number
C : Average frequency number of molecules in the detection volume
of a single molecule
(Criterion 2) : Analytical sensitivity to detect
j2 N . P(L1 ∩ L2) cxp 2p 4 D t
a single molecule
: Arrival and departure probability of the same single molecule
Measurable map (q, t)
* * * *
(Criterion 1) : Arrival probability
Real function x(q, t )
*
I
: Detection probability per time unit T
0.3 f 0.2 D : Diffusion coefficient of a single I 0.1 molecule measured 0 j v : Half axis of the detection volume measured x,y 4 cm: Molar concentration of t : Measurement time
other molecules of the same kind in the bulk
τdiff : Diffusion time
2
Single-Phase Single-Molecule Fluorescence autoand two-color cross-Correlation Spectroscopy
1 0.8 0.6 0.4 t 0 q
0.2 2 4
0
Tm : Meaningful time in which one is able to study just one single molecule in the probe region V
(Criterion 4 ) t Tm
τdiff cmNA V exp {cm NA V}
:
The repertoir of single-molecule spectroscopy in solution is extended to continuous observation of an individual (“selfsame” single) molecule within the measurement time from millisecond range up to hours
Figure 48.2 Synopsis of a new physically grounded technology of fluorescence fluctuation spectroscopy for observing single molecules at longer time scales than currently available. (See also Földes-Papp, 2001c, 2002, 2005, 2006, 2007a, b, c; Földes-Papp et al., 2001, 2005a) In many respects, such methods are an alternative to DNA chip technology (Földes-Papp et al., 2005a, b). They detect the molecular Brownian movement of fluorescent particles in a very tiny volume of the laser focus. Quantitative understanding of molecular interactions at the level of single molecules within single cells is the next step in basic and applied biomedical research for the analysis of the dynamics and localization of molecules in a variety of physiological and pathophysiological processes.
554
CHAPTER 48
■
Viral Chip Technology in Genomic Medicine
The technical solutions for high mutation and recombination rate studies are power-l
