Nature Biotechnology 06 2010 (magazine journal; June 2010)

Author: Andrew Marshall

92 downloads 1720 Views 14MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

volume 28 number 6 june 2010

e d i tor i a l 529

An empiric victory

© 2010 Nature America, Inc. All rights reserved.

news Artist’s impression of digital ELISA. Rissin et al. determine protein abundance by capturing beads bearing single analyte molecules in arrayed femtoliter-volume reaction chambers (p 595). Credit: Ken Eward © BioGrafx.

531 Landmark approval for Dendreon’s cancer vaccine 533 Firms chase diabetic inflammation with anti-IL-1β antibodies 534 African GM safety drill 535 Burgeoning stem cell product market lures major suppliers 536 GMP cell lines to order 536 Open-access fermenter 537 Glyphosate resistance threatens Roundup hegemony 538 SBIR grants wax 538 Relief over stem cell lines 539 Obama appoints bioethics panel to offer practical advice 539 GSK’s RNA splash 539 Germany caps drug prices 540 News feature: Biotech breeding goes bovine 544 News feature: Up for grabs

B i oe n trepre n eur B u i l d i n g a bus i n ess 547

Beyond venture capital John Hollway

op i n i o n a n d c omme n t

Marker-assisted breeding, p 540

C O R R E S P O ND E NC E 551 1 out of 27—European politicians score poorly in agbiotech 552 Split approvals and hot potatoes 553 Why drought tolerance is not the new Bt 554 Health impact in China of folate-biofortified rice 556 Alive and kicking

Nature Biotechnology (ISSN 1087-0156) is published monthly by Nature Publishing Group, a trading name of Nature America Inc. located at 75 Varick Street, Fl 9, New York, NY 10013-1917. Periodicals postage paid at New York, NY and additional mailing post offices. Editorial Office: 75 Varick Street, Fl 9, New York, NY 10013-1917. Tel: (212) 726 9335, Fax: (212) 696 9753. Annual subscription rates: USA/Canada: US$250 (personal), US$3,520 (institution), US$4,050 (corporate institution). Canada add 5% GST #104911595RT001; Euro-zone: €202 (personal), €2,795 (institution), €3,488 (corporate institution); Rest of world (excluding China, Japan, Korea): £130 (personal), £1,806 (institution), £2,250 (corporate institution); Japan: Contact NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 (03) 3267 8751, Fax: 81 (03) 3267 8746. POSTMASTER: Send address changes to Nature Biotechnology, Subscriptions Department, 342 Broadway, PMB 301, New York, NY 10013-3910. Authorization to photocopy material for internal or personal use, or internal or personal use of specific clients, is granted by Nature Publishing Group to libraries and others registered with the Copyright Clearance Center (CCC) Transactional Reporting Service, provided the relevant copyright fee is paid direct to CCC, 222 Rosewood Drive, Danvers, MA 01923, USA. Identification code for Nature Biotechnology: 1087-0156/04. Back issues: US$45, Canada add 7% for GST. CPC PUB AGREEMENT #40032744. Printed by Publishers Press, Inc., Lebanon Junction, KY, USA. Copyright © 2010 Nature America, Inc. All rights reserved. Printed in USA.

i

volume 28 number 6 june 2010 pate n ts KRAS G12D

557 560

Pluripotent patents make prime time: an analysis of the emerging landscape Brenda M Simon, Charles E Murdoch & Christopher T Scott Recent patent applications in epigenetics

N E W S A ND V I E W S

Evaluating cancer models, p 561

561

Raising the bar for cancer therapy models see also p 585 Giulio Francia & Robert S Kerbel

562

Scalable pluripotent stem cell culture see also p 581, 606, 611 Larry Couture

564 Complex molecular dynamics in the spotlight Lois Pollack & Watt W Webb

© 2010 Nature America, Inc. All rights reserved.

565 Detecting methylated bases in real time Markus Elsner 566

Research highlights

c omputat i o n a l b i o l og y a n a ly s i s 567 Comparative assessment of methods for aligning multiple genome sequences Xiaoyu Chen & Martin Tompa

resear c h R ev i e w 573 Synthetic hESC culture matrix, p 581

Rationalizing the development of live attenuated virus vaccines A S Lauring, J O Jones & R Andino B R I E F C O M M U NIC AT I O N

581

Synthetic polymer coatings for long-term growth of human embryonic stem cells L G Villa-Diaz, H Nandivada, J Ding, N C Nogueira-de-Souza, P H Krebsbach, see also p 562 K S O’Shea, J Lahann & G D Smith A R T ICL E

585

Assessing therapeutic responses in Kras mutant cancers using genetically engineered mouse models M Singh, A Lima, R Molina, P Hamilton, A C Clermont, V Devasthali, J D Thompson, J H Cheng, H B Reslan, C C K Ho, T C Cao, C V Lee, M A Nannini, G Fuh, R A D Carano, H Koeppen, R X Yu, W F Forrest, G D Plowman & L Johnson see also p 561

Digital protein assay, p 595

nature biotechnology

iii

volume 28 number 6 june 2010 l etters 595

Single-molecule enzyme-linked immunosorbent assay detects serum proteins at subfemtomolar concentrations D M Rissin, C W Kan, T G Campbell, S C Howes, D R Fournier, L Song, T Piech, P P Patel, L Chang, A J Rivnak, E P Ferrell, J D Randall, G K Provuncher, D R Walt & D C Duffy

600 Identification of influenza A nucleoprotein as an antiviral target R Y Kao, D Yang, L-S Lau, W H W Tsui, L Hu, J Dai, M-P Chan, C-M Chan, P Wang, B-J Zheng, J Sun, J-D Huang, J Madar, G Chen, H Chen, Y Guan & K-Y Yuen 606

© 2010 Nature America, Inc. All rights reserved.

Druggable influenza protein, p 600

Synthetic peptide-acrylate surfaces for long-term self-renewal and cardiomyocyte differentiation of human embryonic stem cells Z Melkoumian, J L Weber, D M Weber, A G Fadeev, Y Zhou, P Dolley-Sonneville, J Yang, L Qiu, C A Priest, C Shogbon, A W Martin, J Nelson, P West, J P Beltzer, see also p 562 S Pal & R Brandenberger

611 Long-term self-renewal of human pluripotent stem cells on human recombinant laminin-511 S Rodin, A Domogatskaya, S Ström, E M Hansson, K R Chien, J Inzunza, O Hovatta & K Tryggvason see also p 562 R esour c e

Recombinant hESC culture matrix, p 611

617

Analysis of a genome-wide set of gene deletions in the fission yeast Schizosaccharomyces pombe D-U Kim, J Hayles, D Kim, V Wood, H-O Park, M Won, H-S Yoo, T Duhig, M Nam, G Palmer, S Han, L Jeffery, S-T Baek, H Lee, Y S Shim, M Lee, L Kim, K-S Heo, E J Noh, A-R Lee, Y-J Jang, K-S Chung, S-J Choi, J-Y Park, Y Park, H M Kim, S-K Park, H-J Park, E-J Kang, H B Kim, H-S Kang, H-M Park, K Kim, K Song, K B Song, P Nurse & K-L Hoe

624

ERRATA

c areers a n d re c ru i tme n t

nature biotechnology

625

The ABC’s of industry: a postdoc program provides a sneak peek into industry careers Adnan O Abu-Yousif, Erik C Hett, Ann M Skoczenski & Tayyaba Hasan

628

people

v

in this issue

© 2010 Nature America, Inc. All rights reserved.

Defining hESC culture Human embryonic stem cells (hESCs) in culture—at least in the culture conditions we have today—tend to lose their defining characteristic of pluripotency. To maintain them in an undifferentiated state, researchers often rely on complex culture components such as mouse ‘feeder’ cells and Matrigel, an extracellular matrix–like substance derived from mouse sarcoma cells. But these materials are xenogeneic, chemically undefined and variable from lot to lot, compromising the reproducibility of experimental results and raising safety concerns over possible contaminants. Feeder cells can already be replaced with fully defined media, but a satisfactory alternative to Matrigel remains to be found. Three groups now report fully defined substrates that support long-term culture of hESCs. Melkoumian and colleagues show that peptide-acrylate surfaces bearing synthetic peptides derived from bone sialoprotein and vitronectin are suitable both for maintaining hESCs and for differentiating them to cardiomyocytes. Tryggvason and colleagues use a recombinant form of laminin-511 to culture hESCs and human induced pluripotent stem cells. Finally, Smith and colleagues culture hESCs on a surface made of the synthetic polymer poly[2(methacryloyloxy)ethyl dimethyl-(3-sulfopropyl)ammonium hydroxide]. In combination with fully defined media, fully defined substrates should improve reproducibility and safety as well as facilitate scale-up of hESC production. [Letters, p. 606, 611; Brief Communications, p. 581; News and Views, p. 562] KA

Reinventing live attenuated vaccines Many early viral vaccines were produced by attenuating viral activity through repeated passage in culture. However, because of unpredictability of the process and safety concerns, vaccinologists turned to other methods, such as inactivation of viruses by mutation or making vaccines using viral membrane protein subunits as antigens. But now, as virologists have learned more about the genes that are required for virulence, and developed tools that can modulate their activity, vaccinologists are once again seeing a role for live-attenuated vaccines. Andino and colleagues review some of the more promising methods that employ molecular tools for preparing attenuated viruses and point out both the progress and pitfalls of each approach in the production of vaccines. [Review, p. 573] LD Written by Kathy Aschheim, Laura DeFrancesco, Markus Elsner, Michael Francisco, Peter Hare, Craig Mak & Lisa Melton

nature biotechnology volume 28 number 6 JUNE 2010

Fission yeast knockout library Genome-wide gene deletion libraries are powerful tools to elucidate gene functions, in the investigation of molecular mechanisms and in the identification of potential drug or molecular engineering targets. Hoe and colleagues present a collection of heterozygous knockout fission yeast strains that covers 98.4% of the Schizosaccharomyces pombe genes. An analysis of gene dispensability in haploid cells reveals that 1,260 of the 4,836 genes are essential under the growth conditions used. Comparison with the budding yeast Saccharomyces cerevisiae— the only other eukaryotic organism for which such a comprehensive library exists—shows that 83% of the genes present in both yeast species have the same dispensability. Differences are mainly found in genes involved in mitochondrial function, DNA replication and intracellular transport. Haploinsufficient and haploproficient genes were identified in growth profiling experiments. Again, comparison with budding yeast showed marked differences and similarities. The authors speculate that some of the genes regulating cell expansion in both yeasts might be more generally important regulators throughout the eukaryotic lineage. [Resource, p. 617] ME

KRAS cancer model road test The power of many mouse models to predict the success of therapies in patients remains disappointingly low. Johnson and colleagues compare data obtained from mice genetically engineered to develop KRAS-driven pancreatic or non-small cell lung tumors with results from human clinical trials. For both cancers, they optimize the therapeutic regimen in mice to mimic the human protocols as closely as possible. Using different combinations of standard-ofcare chemotherapeutics and experimental targeted therapies like vascular endothelial growth factor and epidermal growth factor receptor inhibitors, the authors show that the genetically engineered mouse models accurately reproduce many features of the human responses to the different treatments. In contrast to many mouse studies that use parameters like tumor volume or growth rate that are easy to measure but difficult to compare with clinical trial results, Johnson and colleagues monitor the more commonly used clinical end points, such as overall survival and progression-free survival, in their mouse studies. [Articles, p. 585; News and Views, p. 561] ME

Digital ELISA Duffy and colleagues extend the capacity of the standard sandwich ELISA to enable single-molecule sensitivity. They accomplish this by capturing target analytes on microbeads (no more than one target analyte per bead), which are then segregated in arrays of vii

in this iss u e femtoliter-volume reaction chambers. The small size of the wells, each of which cannot accommodate more than a single bead, ensures that a sufficiently high local concentration of fluorescent product accumulates to permit detection of a single immunoconjugate. In a standard ELISA, the enzymatically generated fluorescent product diffuses into too large a volume to permit detection of a single labeling event. Rather than quantifying mean fluorescence, the authors quantify the percentage of wells with fluorescent product relative to the total number of wells containing beads. The ability to detect fluorescence on a bead-by-bead basis enables the authors to detect subfemtomolar concentrations of spiked protein standards in diluted bovine serum and permits unprecedented sensitivity in detection of prostate-specific antigen from the serum of patients who have undergone radical prostatectomies. This approach may facilitate earlier diagnosis of disease and the discovery of low-abundance biomarkers. [Letters, p. 595] PH

© 2010 Nature America, Inc. All rights reserved.

Genome aligners compared More genomes are better than one, especially when it comes to identifying conserved sequences that may be functional or to investigating mechanisms of evolution. But the typical method of comparing multiple genomes—large-scale multiple sequence alignment—is a computationally difficult problem, with many tools in existence and a lack of clear strategies for comparing performance among them. Chen and Tompa assess four commonly used multiple sequence alignment tools. The authors analyze the coverage, accuracy and level of agreement of

Patent roundup The field of induced pluripotent stem (iPS) cells has moved at a blistering pace, and this has been reflected in the international patent landscape. While Michael Eisenstein explores the emerging intellectual property, Simon and colleagues examine of the claims of three iPS patents to help determine their ultimate value. [News Feature, p. 544; Patent Article, p. 557] LD & MF Patents for an anti-interleukin-1β humanized monoclonal antibody have been awarded to Xoma of Berkeley, California. The new awards for XOMA 052 encompass claims for treating inflammatory conditions including type 2 diabetes. These could challenge Novartis and Eli Lilly, who are also pursuing disease-modifying breakthrough therapies to restore insulin sensitivity. [News, p. 533] LM Recent patent applications in epigenetics. [New Patents, p. 560]

viii

MF

the tools across 1% of the human genome previously aligned to 27 vertebrate genomes as part of the ENCODE project. They also apply a statistical method to identify suspiciously aligned genomic regions that could provide feedback to improve alignment tools. By analyzing their results with respect to different species and categories of genomic regions, Chen and Tompa report a surprising lack of agreement among the alignments and suggest that one alignment method, Pecan, performs well overall. Their work provides a blueprint for assessing the next generation of tools and alignments that are sure to ride the tidal wave of genomes being sequenced. [Analysis, p. 567] CM

Influenza drug target The emergence of influenza virus strains resistant to the current generation of antiviral drugs necessitates the development of new medicines and the identification of additional targets. Using a chemical genetics screen, Kao and colleagues now show that the influenza nucleoprotein A can be inhibited by small-molecule compounds. Treatment of infected cells with their lead compound, nucleozin, led to a relocalization of nucleoprotein A from the nucleus to the cytoplasm, most likely by inducing the formation of large aggregates that cannot be transported to the nucleus. Virus replication is inhibited with a nanomolar median effective concentration. Nucleozin protects mice from lethal challenge with the highly pathogenic H5N1 strain. A mutation that can confer resistance to nucleozin is identified in in vitro experiments. Although uncommon in most viral strains, this mutation is prevalent in the latest H1N1 viruses. [Letters, p. 600] ME

Next month in • Engineering activatable kinases • Inhibitors for ubiquitin E3 ligases • Mouse knockout library • Synthetic live attenuated influenza vaccine • Modeling signal integration by platelets • Genome of a model mushroom

volume 28 number 6 JUNE 2010 nature biotechnology

www.nature.com/naturebiotechnology

EDITORIAL OFFICE [email protected] 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9200, Fax: (212) 696 9635 Chief Editor: Andrew Marshall Senior Editors: Laura DeFrancesco (News & Features), Kathy Aschheim (Research), Peter Hare (Research), Michael Francisco (Resources and Special Projects) Business Editor: Brady Huggett Associate Business Editor: Victor Bethencourt News Editor: Lisa Melton Associate Editors: Markus Elsner (Research), Craig Mak (Research) Editor-at-Large: John Hodgson Contributing Editors: Mark Ratner, Chris Scott Contributing Writer: Jeffrey L. Fox Senior Copy Editor: Teresa Moogan Managing Production Editor: Ingrid McNamara Senior Production Editor: Brandy Cafarella Production Editor: Amanda Crawford Senior Illustrator: Katie Vicari Illustrator/Cover Design: Kimberly Caesar Senior Editorial Assistant: Ania Levinson

© 2010 Nature America, Inc. All rights reserved.

MANAGEMENT OFFICES NPG New York 75 Varick Street, Fl 9, New York, NY 10013-1917 Tel: (212) 726 9200, Fax: (212) 696 9006 Publisher: Melanie Brazil Executive Editor: Linda Miller Chief Technology Officer: Howard Ratner Head of Nature Research & Reviews Marketing: Sara Girard Circulation Manager: Stacey Nelson Production Coordinator: Diane Temprano Head of Web Services: Anthony Barrera Senior Web Production Editor: Laura Goggin NPG London The Macmillan Building, 4 Crinan Street, London N1 9XW Tel: 44 207 833 4000, Fax: 44 207 843 4996 Managing Director: Steven Inchcoombe Publishing Director: Peter Collins Editor-in-Chief, Nature Publications: Philip Campbell Marketing Director: Della Sar Director of Web Publishing: Timo Hannay NPG Nature Asia-Pacific Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843 Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 Publishing Director — Asia-Pacific: David Swinbanks Associate Director: Antoine E. Bocquet Manager: Koichi Nakamura Operations Director: Hiroshi Minemura Marketing Manager: Masahiro Yamashita Asia-Pacific Sales Director: Kate Yoneyama Asia-Pacific Sales Manager: Ken Mikami DISPLAY ADVERTISING [email protected] (US/Canada) [email protected] (Europe) [email protected] (Asia) Global Head of Advertising and Sponsorship: Dean Sanderson, Tel: (212) 726 9350, Fax: (212) 696 9482 Global Head of Display Advertising and Sponsorship: Andrew Douglas, Tel: 44 207 843 4975, Fax: 44 207 843 4996 Asia-Pacific Sales Director: Kate Yoneyama, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746 Display Account Managers: New England: Sheila Reardon, Tel: (617) 399 4098, Fax: (617) 426 3717 New York/Mid-Atlantic/Southeast: Jim Breault, Tel: (212) 726 9334, Fax: (212) 696 9481 Midwest: Mike Rossi, Tel: (212) 726 9255, Fax: (212) 696 9481 West Coast: George Lui, Tel: (415) 781 3804, Fax: (415) 781 3805 Germany/Switzerland/Austria: Sabine Hugi-Fürst, Tel: 41 52761 3386, Fax: 41 52761 3419 UK/Ireland/Scandinavia/Spain/Portugal: Evelina Rubio-Hakansson, Tel: 44 207 014 4079, Fax: 44 207 843 4749 UK/Germany/Switzerland/Austria: Nancy Luksch, Tel: 44 207 843 4968, Fax: 44 207 843 4749 France/Belgium/The Netherlands/Luxembourg/Italy/Israel/Other Europe: Nicola Wright, Tel: 44 207 843 4959, Fax: 44 207 843 4749 Asia-Pacific Sales Manager: Ken Mikami, Tel: 81 3 3267 8765, Fax: 81 3 3267 8746 Greater China/Singapore: Gloria To, Tel: 852 2811 7191, Fax: 852 2811 0743 NATUREJOBS [email protected] (US/Canada) [email protected] (Europe) [email protected] (Asia) US Sales Manager: Ken Finnegan, Tel: (212) 726 9248, Fax: (212) 696 9482 European Sales Manager: Dan Churchward, Tel: 44 207 843 4966, Fax: 44 207 843 4596 Asia-Pacific Sales & Business Development Manager: Yuki Fujiwara, Tel: 81 3 3267 8765, Fax: 81 3 3267 8752 SPONSORSHIP [email protected] Global Head of Sponsorship: Gerard Preston, Tel: 44 207 843 4965, Fax: 44 207 843 4749 Business Development Executive: David Bagshaw, Tel: (212) 726 9215, Fax: (212) 696 9591 Business Development Executive: Graham Combe, Tel: 44 207 843 4914, Fax: 44 207 843 4749 Business Development Executive: Reya Silao, Tel: 44 207 843 4977, Fax: 44 207 843 4996 SITE LICENSE BUSINESS UNIT Americas: Tel: (888) 331 6288 Asia/Pacific: Tel: 81 3 3267 8751 Australia/New Zealand: Tel: 61 3 9825 1160 India: Tel: 91 124 2881054/55 ROW: Tel: 44 207 843 4759

[email protected] [email protected] [email protected] [email protected] [email protected]

CUSTOMER SERVICE www.nature.com/help Senior Global Customer Service Manager: Gerald Coppin For all print and online assistance, please visit www.nature.com/help Purchase subscriptions: Americas: Nature Biotechnology, Subscription Dept., 342 Broadway, PMB 301, New York, NY 100133910, USA. Tel: (866) 363 7860, Fax: (212) 334 0879 Europe/ROW: Nature Biotechnology, Subscription Dept., Macmillan Magazines Ltd., Brunel Road, Houndmills, Basingstoke RG21 6XS, United Kingdom. Tel: 44 1256 329 242, Fax: 44 1256 812 358 Asia-Pacific: Nature Biotechnology, NPG Nature Asia-Pacific, Chiyoda Building, 2-37 Ichigayatamachi, Shinjuku-ku, Tokyo 162-0843. Tel: 81 3 3267 8751, Fax: 81 3 3267 8746 India: Nature Biotechnology, NPG India, 3A, 4th Floor, DLF Corporate Park, Gurgaon 122002, India. Tel: 91 124 2881054/55, Tel/Fax: 91 124 2881052 REPRINTS [email protected] Nature Biotechnology, Reprint Department, Nature Publishing Group, 75 Varick Street, Fl 9, New York, NY 10013-1917, USA. For commercial reprint orders of 600 or more, please contact: UK Reprints: Tel: 44 1256 302 923, Fax: 44 1256 321 531 US Reprints: Tel: (617) 494 4900, Fax: (617) 494 4960

Editorial

An empiric victory Provenge already looks like the product of a bygone era.

© 2010 Nature America, Inc. All rights reserved.

T

he approval of Provenge is remarkable. It represents a triumph of esotericism over the scientific method; a victory for doggedness over diligence; a success for clinical and manufacturing brawn over molecular precision. As the first approved cellular vaccine against cancer, it vindicates the persistence of those who have labored for decades to obtain clinical validation of the approach. More specifically, the product proves that the T-cell arm of the adaptive response can be harnessed to fight advanced cancers. These are good things. But there are also reasons why Provenge is likely to achieve only limited commercial success. The Provenge story starts with studies at the Stanford University School of Medicine published in 1997 (J. Immunol. 159, 3113–3117) that showed that a cytotoxic T-lymphocyte (CTL) response to prostatic acid phosphatase (PAP) antigen could result in the destruction of PAP-bearing tissues. In other words, eliciting T-cell immunity, and not solely an antibody response, might be effective as a cancer immunotherapy. But as tumors are so adept at cloaking themselves from immune surveillance, the question was how to elicit an appropriate cellular response to the tumor antigen of interest. Dendreon’s solution was to induce cellular immunity ex vivo by removing and semipurifying (by centrifugation) patients’ antigen-presenting cells (APCs), or dendritic cells, and exposing them to tumor antigen supercharged with an immunomodulator. The specific trick, in essence, is to present dendritic cells with a growth-promoting cytokine (granulocytemacrophage colony stimulating factor) hooked up to an antigen that is enriched in prostate cancer (PAP). Thus, from a scientific perspective, Provenge provides evidence of the clinical relevance of anti-tumor T cell–mediated immunity. And it shows that if you can prime APCs correctly and infuse sufficient numbers of them back into the circulation, life-prolonging immune defenses can be invoked. Provenge validates the immunologists’ original vision by adding over 4 months (on average) to the lives of very sick patients. Unfortunately, between the immunological vision and the clinical validation lay a drug developers’ nightmare. Provenge stuttered and stumbled through the regulatory pipe. Yes, it was relatively quickly shown to be safe enough (phase 1 studies were reported first in 2000). But the dose escalation and efficacy work took another decade. This is partly because, when it comes to human cell products, there are no relevant animal studies to guide dose-ranging, product formulation and administration protocols. Humans are the animal model and this, necessarily, slows down development. Furthermore, the autologous nature of the product means that the sources of cells are highly variable. Each patient presents an individual challenge, varying in age, disease severity, prior treatment, tumor microenvironment and immune status. In addition, despite ‘enrichment’ using the marker CD52, the Provenge preparation remains a complex mixture of lymphocyte and myeloid cell types and their macromolecular products. Thus, every Provenge treatment is slightly different from the next. It is personalized medicine in its worst sense. Little more is known now about nature biotechnology volume 28 number 6 JUNE 2010

what constitutes an effective cellular compote in Provenge than was known at the beginning of the century. Similarly, little is known about what differentiates patients who have positive clinical responses to Provenge from those who don’t. And any knowledge that has been acquired is likely to be of limited use to developers of other cancer vaccines. Another problem from the drug developer’s standpoint is that Provenge is less a product and more a service—and a logistically awkward, multistep, difficult-to-control service at that. Dendreon raised over $600 million in 2009, much of which will fund a production facility for Provenge. Unsurprisingly, the cost of a Provenge treatment is >$90,000 per patient. Any of the slew of other autologous cancer vaccine candidates making their way through the clinic (Nat. Biotechnol. 27, 129–139, 2009) is likely to face a similarly adventurous route to market: difficult regulatory birthing and awkward, expensive, undrug-like products. Label expansion is likely to be equally as painful as the biologic license application process. As a fourth-line treatment (after surgery, radiation and chemotherapy) in advanced prostate cancer, Provenge’s path to commercial nirvana also looks less than straightforward, beset by manufacturing and scale-up issues. While there remains an absence of other treatment options, it has a chance of market success. But it looks very vulnerable to competition from more tractable and patient-friendly immunotherapies, such as the next generation of off-the-shelf cancer vaccines or antibodies that direct prostate cancer antigens to dendritic cells. As with other new drug modalities, market registration of the first cancer T-cell vaccine product will enable clinicians to start to systematically gather patient data to better characterize the immunotherapy itself as well as the immune responses it elicits in patients. This can only boost a field that has struggled to translate findings gleaned from animal models into human subjects. With rapid recent progress in our understanding of tumor immunology, emphasis should now shift to assessing the quality and composition of the types of dendritic cells involved in eliciting CD4 and CD8 T cells with the highest avidity for tumor antigens. At the same time, it will be important to understand those cell types that thwart vaccine strategies through promoting the expansion of regulatory T cells or the recruitment of immature myeloid suppressive cells to tumors from the bone marrow. To date, most cancer vaccines in the clinic have focused on factors that promote expansion of CTLs rather limiting immune suppression in the tumor microenvironment. Several immunosuppressive targets are now starting to be explored, including CTL antigen 4 (CTLA-4), TPD-I receptor (CD279) and PD-I (glucocorticoid-induced tumor necrosis factor receptor–related protein ligand). The market authorization of Provenge marks the end of the beginning for cellular immunotherapy in cancer. The field can now move ahead, with a proof of concept in hand. But if Provenge signifies anything for cancer vaccines, it is that the path forward lies less in empiricism and more in scientific rigor. 529

news in this section IL-1 blockers treat diabetic inflammation p533

Big suppliers sell stem cells as screening tools

Warning on weed resistance to glyphosate p537

p535

The April 29 approval of Seattle-based Dendreon’s prostate cancer vaccine, Provenge (sipuleucel-T), is being hailed as a victory for cancer immunotherapy. For Dendreon, the US Food and Drug Administration’s (FDA) go-ahead marks the end of a tortuous regulatory path, marked not only by missteps by the company but also by controversy at the FDA, not least the decision in 2007 by the Center for Biologics Evaluation and Research (CBER) to act against its advisory panel’s positive recommendations. After the turmoil of ad campaigns critical of the agency, picketing and lobbying by patient groups, death threats, lawsuits and even calls for a Congressional investigation (Nat. Biotechnol. 26, 1, 2008), the FDA issued a complete response letter on the earlier trials and requested further clinical evidence of efficacy. Dendreon then soldiered on with a phase 3 placebo-controlled trial (Immunotherapy for Prostate Adenocarcinoma Treatment; IMPACT), the results of which were submitted to FDA last November. On the basis of these data, which have yet to be published in a peer-reviewed journal, the agency finally gave Provenge its imprimatur, approving the first therapeutic vaccine for use in individuals with asymptomatic, or minimally symptomatic hormone refractory metastatic prostate cancer. The approval has received an ecstatic reception from patient groups, oncologists and cancer vaccinologists, who view Provenge and potentially other cell vaccines as a valuable and complementary adjunct to the growing list of different cytotoxic and cytostatic therapies used in the fight against cancer. Provenge is being touted as a nontoxic cancer treatment for an underserved population, which in the US alone surpasses 76,000 patients. “It’s a hugely exciting time for our field,” says Bernard Fox, president of the International Society for Biological Therapy of Cancer. “We’ve been doing it [immunotherapy] for 25 years, through times that have been very bad for the community. We didn’t have much to point to except clinicians that had seen their patients respond.”

It remains unclear, however, whether Dendreon’s decade-long struggle to pass regulatory muster has clarified the path of oversight for other cancer vaccines or even whether autologous cellular vaccines will rival the success of ‘off-the-shelf ’ vaccines or other types of adjunct therapies, such as antibodies or small molecules. Therapeutic cancer vaccines are a diverse group of products; they can be cellular or acellular (peptides, proteins, DNA), be targeted against a single antigen or groups of antigens, use viruses or other scaffolds to present antigens or use patient cells or cell lines (Nat. Biotechnol. 27, 129–139, 2009). Provenge, unique among cancer vaccines in late-stage clinical trials (Table 1), is an autologous, cell-based therapy created by incubating (activating) the patient’s own antigen-presenting cells ex vivo with a fusion of prostatic acid phosphatase (an antigen specific to prostate tissue) and granulocyte macrophage colony-stimulating factor, which act to stimulate immune cell responses. This is a first-generation product, but it is both simpler (uses a single antigen) and more complex (works with a mélange of cells) than some of the other products under development.

Recombinant prostatic acid phosphatase (PAP) antigen combines with resting antigen presenting cel (APC)

APC takes up the antigen

Dendreon’s clinical trial design and analysis of the human data have been dogged by controversy. Two early trials of Provenge showed a benefit in overall survival (OS) but not progression-free survival (PFS), which is unusual according to Don Berry, chairman of biostatistics at MD Anderson Hospital in Houston. “If something is effective in cancer, it inhibits or slows growth and this apparently does not,” he says. Unfortunately for Dendreon, PFS was the primary endpoint in these early trials. The FDA refused to move the goalposts, and sent Dendreon back to gather more data, this time using OS as an endpoint in a large (512-patient) phase 3 trial, which was already underway. Last October, Dendreon announced interim results, essentially priming the pump for investors, if not regulators. (The company raised $409.5 million in a stock offering the following month.) The release of interim results to the company by the data monitoring group was unusual, according to Susan Ellenberg of the University of Pennsylvania in Philadelphia, who led the team that wrote the guidance for placebo-controlled trials when she was at CBER. Apparently in this case it was done with the consent of the FDA. The interim data had not achieved statistical

Antigen is processed and presented on surface of the APC Active T-cell

T-cells proliferate and attack cancer cells

Fully activated, the APC is now Provenge INFUSE PATIENT Inactive T-cell

Provenge activates T-cells in the body

Dendreon

© 2010 Nature America, Inc. All rights reserved.

Landmark approval for Dendreon’s cancer vaccine

The making of a cancer vaccine. The precise mechanism of Provenge in prostate cancer has not been established.

nature biotechnology volume 28 number 6 JUNE 2010

531

NEWS Table 1 Selected cancer vaccines in phase 3 clinical trials

© 2010 Nature America, Inc. All rights reserved.

Company (location)

Product description

Indication

Antigenics HSPPC-96 Oncophage: heat-shock protein vaccine (Lexington, Massachusetts) isolated from patient tumor cells

Melanoma Glioma Renal cell carcinoma

BioVest International (Tampa, Florida)

Biovaxid: patient-specific immunoglobulin idiotype vaccine conjugated to the immunogenic protein KLH

Non-Hodgkin’s lymphoma

Genitope (Fremont, California)

Patient-specific immunoglobulin idiotype-KLH conjugate

Non-Hodgkin’s lymphoma

GlaxoSmithKline (Brentford, UK)

MAGE: liposomally packaged tumor-specific antigen

Melanoma Lung cancer

Northwest Biotherapeutics (Bethesda, Maryland)

DCVax: patient-derived dendritic cells loaded with cancer proteins or lysates

Prostate cancer Brain cancer

NovaRX (San Diego)

Lucanix: four cell lines carrying antisense oligos against transforming growth factor

Lung cancer

Oncothyreon (Seattle)

Stimuvax: liposomal vaccine with a synthetic peptide derived from tumor-specific antigen MUC-1

Lung cancer

Oxford Biomedica (Oxford, UK)

TroVax: pox viral vector carrying tumor-associated antigen 5T4

Renal cell carcinoma

significance (Dendreon needed to achieve 22.5% improvement in OS but at that time, they were only at 20%). Dendreon researchers were confident, based on their prior experience from randomized trials, however, that once all the results had been collected, the data would meet the mark, which was, in fact, the case. As reported at the American Society of Clinical Oncology (ASCO) 2010 Genitourinary Cancers Symposium, held March 5–7 in San Francisco, three-year OS rates were 38% higher among men who received the drug than those who received placebo. Provenge showed a median OS benefit of 4.1 months compared with the placebo (P = 0.032). On the basis of these results, the FDA declined to convene an advisory panel, although rumors circulated in March that one might take place. Dendreon’s stock price took a hit, as investors tried to second-guess which way the winds were blowing at the agency. But as the previous panel had voted for approval with clinical data that fell short of statistical significance, it seemed unlikely that agency officials would convene a panel again. Indeed, at the end of the next month, FDA finally gave the formal green light, announcing marketing authorization for Provenge. One further complication with the IMPACT data has had statisticians scratching their heads. This is the use of previously frozen Provenge—which some are calling Frovenge—as the salvage protocol for patients who progressed on the placebo arm. Those on the experimental arm whose disease progressed received chemotherapy with docetaxel. Offering progressors alternative therapies is common, but giving an unproven therapy, which on top of being unproven, is different from the product

532

given to the experimental arm, introduces an uncontrolled variable and confounds analysis when the endpoint had yet to be met (death). Mark Frohlich, Dendreon’s chief medical officer, explains that using Frovenge was preferable to creating the vaccine anew from trial participants. Each patient, regardless of which arm they were on, had to undergo three leukophoreses, an invasive procedure, to isolate the cells necessary for the therapy or the placebo. According to Frohlich, Frovenge met the same specifications as Provenge, “Scientifically there is really no biological or scientific rationale as to why a product that meets the same release specs would be deleterious to the patient,” he says. Furthermore, when the trial started, there was no therapy available for progressors; doxetaxol was approved only later for use in this patient population. Another factor tempering enthusiasm in some quarters is the fact that, at least for now, the data have only been reviewed by the FDA and Dendreon, which presented a summary of the data at ASCO. (Frohlich says Dendreon intends to publish the data in a peer-reviewed journal, but has not indicated when.) Steven Rosenberg, chief of surgery at the National Cancer Institute, who has been working on immunotherapies for over 20 years, finds it strange that the data have not been released, given the newness of the approach that Provenge represents. “Particularly for a field that has had a rash of negative results, it’s important for the scientific community to see the data. That’s how science works,” he says. From the viewpoint of vaccine developers, the ‘rocky’ ride that Provenge received during FDA review also poses some questions. For example, Dendreon’s Frohlich challenges

the conventional wisdom that assessments of efficacy should be the same for vaccines as for more conventional oncology treatments, such as chemotherapy. Until now, the gold standard used by regulators has been the shrinkage of tumors or the downregulation of tumor markers. In such a system, even if a cancer vaccine has a positive effect on OS, “you don’t necessarily expect” to see an effect on tumor shrinkage/burden, Frohlich says. Howard Scher of Sloan Kettering Cancer Institute in New York concurs. He encourages sponsors “not to mandate stopping therapy at the first sign that [the signs are seemingly going in the wrong direction. We just have to be smarter on how we measure [response].” Scher was one of four dissenting votes on the 2007 advisory panel that gave Provenge the green light. After the slog over the regulatory finish line, Dendreon is now faced with the Herculean effort of producing an autologous cell therapy on a large scale. The company started to ramp up its manufacturing capacity before approval, with $630 million raised in two follow-on stock offerings last year. Even so, the firm plans to commence commercialization with dosing of only 2,000 patients—a fraction of the population indicated on the label. The $93,000 price tag for three infusions may also dictate who gets the treatment. According to Frohlich, Dendreon is in discussions with the Center for Medicare and Medicaid Services, as the majority of Provenge’s target patient population is over 65. Eric Schmidt, an analyst with Cowen and Company in Boston, predicted that the price tag would be high, but feels it’s appropriate for a product with proven efficacy and a great safety profile, with no added expense from supportive care. “Price is not a factor,” he believes. Whether Provenge’s approval heralds a new era for cellular cancer vaccines remains to be seen. It seems likely that off-the-shelf products that are simpler to produce, such as tumor antigens targeted to dendritic cells by way of antibody moieties, are likely to supersede more complex cellular products that often suffer from batch-to-batch variability. MD Anderson’s Berry remains guarded about the field’s prospects. “It will still be a hard road to approval for companies with vaccines because the vaccine batting average is still very low. But one hit is better than none,” he says. Fox is more sanguine. “There is a lesson here for us to look at what some might think are crazy ideas. I don’t think a lot of people would have thought it was going to work.” But although it is an important step, he adds, “People are still dying on [Provenge].” Laura DeFrancesco Pasadena, California

volume 28 number 6 JUNE 2010 nature biotechnology

news

A recent US patent award to Xoma for XOMA 052, an anti-interleukin-1β(IL-1β) IgG2 humanized monoclonal antibody (mAb), has spiced up what was already an intriguing contest to uncover disease-modifying therapies for type 2 diabetes. The Berkeley, California–based antibody developer is aggressively staking out territory in a rapidly emerging—and potentially lucrative—field, which has also captured the attention of Novartis and Eli Lilly. Each is pursuing the goal of developing a breakthrough therapy for type 2 diabetes to help restore insulin production and insulin sensitivity. Xoma’s new patents, US 7,695,717 and US 7,695,718, awarded on April 13, encompass claims for treating inflammatory conditions with high-affinity anti-IL-1β mAbs or antibody fragments and for using similar molecules for treating type 2 diabetes and associated complications. They specify mAbs with affinities in the femtomolar range. “Xoma’s patents are very interesting in the sense that they’re quite broad,” says Christopher James, analyst at McNicoll, Lewis, & Vlak, a New York-based investment bank and institutional broker dealer. “Potentially you have a situation where they could challenge both of those companies and be the winner.” Although about a year behind its big pharma rivals, he says, Xoma has already obtained promising earlystage data from a phase 1a trial with XOMA 052, which demonstrated that modulating IL-1β levels improved pancreatic beta cell function and insulin sensitivity. An influential review published earlier this year lent theoretical gravitas to the mechanism (Science, 327, 296–300, 2010). Juerg Tschopp, of the University of Lausanne, Switzerland, and his colleagues propose a central role for the NLRP3

(NOD-like receptor family, pryin domain containing 3) inflammasome—a multiprotein sensor for metabolic danger. This ‘sensing’ complex, they contend, initiates the inflammatory response by promoting the processing of pro-IL-1β into its active extracellular form, in the pathophysiology of type 2 diabetes. “A lot that was known about the IL-1 pathway has been now shown to involve the inflammasome,” says Marc Donath, of University Hospital, Zurich. The first therapies that target inflammasome-associated conditions have already appeared. The anti-IL-1β mAb, Ilaris (canakinumab; human IgG), developed by Basel -based Novartis, and the IL-1 trap Arcalyst (rilonacept), developed by Regeneron Pharmaceuticals, of Tarrytown, New York, have both gained approval for familial Cold Autoinflammatory Syndrome and for Muckle-Wells syndrome, two of the three genetic disorders of the innate immune system that are collectively known as cryopyrin-associated periodic syndromes. These rare diseases result from mutations in NLRP3, which encodes the inflammasome component cryopyrin, and all are characterized by a harmful overproduction of IL-1β. A rapidly growing body of basic and clinical research implicates the inflammasome—and, by extension, IL-1β—in a host of other conditions, including gout, multiple myeloma, central nervous system disease, type 1 diabetes and cardiovascular disease. One recent study, for example, provides evidence that tiny crystals of cholesterol are responsible for the initiation and progression of atherosclerosis. It suggests a role for therapies that block the inflammasome pathway in tackling the problem (Nature 464, 1357–1362, 2010). “That’s the real opportunity here. This is a

charitymeyers/istockphoto

© 2010 Nature America, Inc. All rights reserved.

Firms chase diabetic inflammation with anti-IL-1β antibodies

Most people with type 2 diabetes eventually need insulin. XOMA 052 and other interleukin-1 blockers could prevent beta cell deterioration.

nature biotechnology volume 28 number 6 JUNE 2010

533

NEWS

in brief African GM safety drill

© 2010 Nature America, Inc. All rights reserved.

AP Photo/Karel Prinsloo

The African Union has set up a school to educate and train future regulators in genetically modified (GM) crop biosafety. The African Biosafety Network of Expertise (ABNE) was officially launched in April in Ouagadougou, Africa shores up GM Burkina Faso, with crop regulators. a five-year, $10.4 million grant from the Bill & Melinda Gates Foundation. This continent-wide initiative, administered by the African Union’s New Partnership for Africa’s Development (NEPAD), aims “to build functional biosafety systems,” says program director Diran Makinde, who notes that of the 12 African countries that have biotech crop research programs, only 3 have reached the stage of commercialization. A tour of Africa taken in 2008 by Makinde and his staff to assess the nations’ different needs highlighted the lack of regulatory expertise. The visiting team concluded that any pan-African solution would need to provide online information resources, training workshops, technical support and partnerships. Today, ABNE’s website offers environmental, socioeconomic and food safety advice and information on issues related to GM crops through a live chat function handled by staff. In late March, before the official launch, ABNE hosted a workshop for about 40 regulators in Accra, Ghana, to discuss locally developed, insectresistant transgenic crops. ABNE’s staff also took part in a training course last fall at Michigan State University in East Lansing to ramp up their own expertise. These newly minted ABNE trainers are equipped to guide regulators in risk assessment and management issues to enable GM crop adoption. But they will need to learn quickly if they are to succeed in training regulators and consultants across Africa’s major languages, according to Theresa Sengooba, a researcher with the International Food Policy Research Institute in Kampala, Uganda. Given the varied state of African biosafety infrastructure, another of ABNE’s challenges will be to determine “how best to help countries which are already advanced as well as those which are behind,” Sengooba adds. Makinde points out, however, that ABNE enjoys an “added value” from NEPAD’s status as a technical arm of the politically well-connected African Union. This supplies the network with the necessary kudos to approach national ministers responsible for agricultural planning and biotech research in African countries. Makinde intends to help two or three additional African countries reach the commercialization stage, and improve regulatory decision-making in the rest within the program’s initial budget. “Our main objective,” Makinde stresses, “is to contribute to food security in Africa.” Lucas Laursen

534

multi-indication situation we’re talking about,” says Xoma CEO Steven Engle. Donath was the first to demonstrate that type 2 diabetes has an inflammatory dimension that is mediated by IL-1β (J. Clin. Invest. 110, 851–860, 2002). That work was motivated by an effort to understand how prolonged, excessive glucose levels led to the destruction of insulin-producing pancreatic beta cells. “We observed that everything comes down to an induction of an inflammatory response, which is driven by IL-1β,” says Donath. He then demonstrated clinical proof of concept, using Kineret (anakinra), an IL-1β inhibitor that is a recombinant, 152–amino acid, nonglycosylated, human N2-L-methionyl IL-1 receptor antagonist (IL-1RA), originally isolated from monocytes (N. Engl. J. Med. 356, 1517–1526, 2007). Kineret competes with IL-1 for binding to the IL-1 receptor type 1 (IL-1R1) on target cells. “That was an earthquake-like shift in thinking,” says Engle. A follow-up study showed the effect was sustained. Patients exhibited improved insulin production nine months after receiving the drug (Diabetes Care 32, 1663–1668, 2009). Although originally approved for treating rheumatoid arthritis, Kineret is rarely used in that indication because of the high doses required owing to its short half-life and to the presence of high levels of IL-1R1 in the joints of rheumatoid arthritis patients. In type 2 diabetes, its long-term use is even more problematic, as pancreatic beta cells express very high concentrations of IL-1R1. “If you want to target the receptor you need a lot of drug,” says Donath. IL-1β, in contrast, exerts its effects at extremely low concentrations, and it is therefore more amenable to modulation. Xoma and each of its big pharma rivals already have efficacy testing trials of anti-IL-1β mAbs in type 2 diabetes underway. Interim, three-month data from a six-month phase 2a trial of XOMA 052 in 80 patients is due in the fourth quarter, and a phase 2b trial, involving 325 patients, will report in full in the first quarter of 2011. Novartis developed Ilaris at the Novartis Institutes for Biomedical Research, in Cambridge, Massachusetts; its phase 2/3 dose-ranging study in diabetes was initiated in April 2009. Indianapolis-based Lilly’s LY2189102, which was developed by scientists now based at the recently opened Lilly Biotechnology Center in San Diego, entered phase 2 testing for diabetes in June 2009. Neither company was willing to comment on their respective programs, however. Xoma is setting XOMA 052 apart on the basis of its mechanism of action, which is to attenuate rather than eliminate the IL-1β signal (J. Biol. Chem., published online, doi/10.1074/ jbc.M110.115790, 21 April, 2010). Although this mAb binds the cytokine with ultra-high affinity, it does not block the action of the cytokine completely. “That antibody and IL-1β [complex] can still bind to the receptor but it doesn’t give a good

signal,” says Alan Solinger, Xoma’s vice president of clinical immunology. That residual signal could be important, as low levels of the cytokine are required to maintain beta cell activity and proliferation. “If you block IL-1β too much—drop IL-1β to zero—the beta cells cannot function normally,” Solinger says. IL-1β is not the only potential anti-inflammatory target in type 2 diabetes. Startup firm Catabasis Pharmaceuticals last month raised $39.6 million in series A funding to develop conjugates of salicylate and omega-3 fatty acids. Both these agents act upstream from the inflammasome, through nuclear factor kappa B (NF-κB), a transcription factor that stimulates production of proIL-1β in response to inflammatory stimuli. “The inflammasome could be an important component of the underlying pathway. I think there’s probably more biology left to work out as well,” says CEO Jill Milne, who cofounded the Cambridge, Massachusetts–based firm with chief scientific officer Mike Jirousek and Steve Shoelson, of Harvard Medical School and the Joslin Diabetes Center in Boston. They aim to move their first compound, CAT-1904, into clinical studies next year. The Catabasis approach builds on Shoelson’s clinical work, which has already shown that highdose salsalate—a dimeric pro-drug of salicylate— resulted in a modest improvement in glycemic control in type 2 diabetes patients (Ann. Intern. Med. 152, 346–357, 2010). By conjugating salicylate with an omega-3 fatty acid, such as eicosapentaenoic acid or docosahexaenoic acid, Catabasis aims to exert a broad effect on the inflammation associated with type 2 diabetes. “All of the approaches to treating inflammation up to this point have focused on trying to inhibit pro-inflammatory pathways,” says Jirousek. Catabasis is attempting to stimulate anti-inflammatory signals as well, by exploiting the conversion of omega-3 fatty acids to antiinflammatory eicosanoids. It plans to publish further details on the mechanism it is pursuing. But the basic rationale is to restore normal functioning by means of a synergistic effect of the two compounds. “It’s only when the cell is in a stressed state that these molecules have an effect,” Jirousek says. In the meantime, all eyes are on Xoma, which is open to partnering XOMA 052 at any stage. “We haven’t restricted ourselves,” says Engle. Its patent awards could result in additional licensing income, says Liana Moussatos, analyst at Wedbush Securities, in San Francisco, although she does not expect the company to engage in any major litigation. “They’ll go through the motions, but in the end it’ll be some kind of cross-licensing deal. That’s usually how these things work out. Or maybe Lilly or Novartis will become their partner.” Cormac Sheridan Dublin

volume 28 number 6 JUNE 2010 nature biotechnology

news

Life sciences supplier Lonza has struck a deal with Axiogenesis of Cologne, Germany, to offer mouse embryonic stem cell–derived cardiomyocytes in its product catalog. The agreement, signed in March, is the latest move of several large reagent and material suppliers to grab a slice of the rapidly expanding market for stem cell products for use in in vitro assays and testing kits for predictive toxicology. Life Technologies, which was formed from the merger of Carlsbad, California–based Invitrogen and Foster City, California–based Applied Biosciences, has been aggressively marketing its range of embryonic stem cell (ESC) and induced-pluripotent stem cell (iPSC) reagents of animal-free origin, and GE Healthcare, of Chalfont St. Giles, UK, has forged a two-year partnership with Geron in Menlo Park, California, to scale up production of differentiated cells from human ESCs. Merck’s acquisition of US reagent and materials supplier Millipore also signals the growing interest of big pharma in iPSC expertise and capacity—a signal of increasing receptiveness within the pharmaceutical industry to embrace stem cell technology. Indeed, with the political climate in the US now favorable, big pharma is openly pursuing the use of such cells in its preclinical research programs, opening up a sizeable market. The biopharma industry spends over $4.98 billion a year on R&D for new therapies. The cost of taking a compound through to late-stage development continues to escalate, yet, at the same time, up to 30% of leads fail because of an unacceptable safety profile. Stem cell–derived products are potentially a useful resource for toxicity screens that could identify leads with unacceptable safety profiles. Until now, the scarcity, expense and batch-to-batch variability of differentiated cells derived from donor tissues have hampered the use of such primary cells in preclinical research. With the advent of stem cell–derived products that can potentially create differentiated cells of all the different lineages—endoderm, mesoderm and ectoderm—a plentiful, consistent and competitive source of cells is becoming available for drug screening. Industry is increasingly recognizing the value of such products for two reasons. First, stem cell products provide a reliable source of primary cells, avoiding the expense, ethical issues and quality control problems associated with deriving such cells from human donor and cadaver tissue. Over the past 20 years, Basel-based Lonza has supplied the research community with primary cells from human donor tissue, says Alex Batchelor, the company’s head of marketing-drug discovery. “Unfortunately, some of the more difficult cell types [to obtain from donors] are the

most interesting ones for researchers: neural cells, cardiomyocytes, hepatocytes and possibly pancreatic cells.” Products differentiated from pluripotent cells can meet that demand for material for testing. A second impetus for the increased interest in stem cell products is their ability to reduce dependency on human tissue and the number of animals used in drug testing—an issue that is particularly troublesome for the public perception of pharma companies in Europe, according to Mahendra Rao, vice president for research in stem cells & regenerative medicine of Life Technologies in Carlsbad, California. Rao views the recent boost in stem cell interest from pharma companies as a pleasant surprise. The company has not disclosed figures relating to these deals, but Rao points to the flurry of deals including the recent $7.6 billion paid by Darmstadt, Germany–based Merck for Millipore (page 536). Last year, another pharma giant, Paris-based Sanofi-aventis, formed a partnership with the Salk Institute by which the company agreed to provide, among other things, funding to support the Institute’s stem cell facility. He attributes this progress—particularly in the US—to both advances in iPSC research and the Obama administration’s more receptive policies toward human ESCs. The move into the marketplace for differentiated products from human pluripotent cells has now become a steady flow (Table 1). In late 2009, Cellular Dynamics International (CDI) in Madison, Wisconsin, began selling iCell Cardiomyocytes (cardiomyocytes derived from human iPSCs), the first such iPSC product to be commercialized. Though CDI originally looked at preclinical toxicology and safety testing as the primary application of cardiomyocytes, the availability of the cells has stimulated new ideas for their use, says chief commercialization officer, Chris Kendrick-Parker. “Already our customers have been able to understand how they can induce a disease state in these cells, which has therefore now moved them into a discovery model,” he says. CDI produces and ships billions of cardiomyocytes per day, according to Kendrick-Parker. The company has delivered cardiomyocytes to more than half the top 20 pharma companies, he adds. GE Healthcare is close behind. In a partnership with Geron, the UK company is scaling up production of differentiated cells from human hESCs at its Cardiff research center; cardiomyocytes will be launched as a commercial product for toxicology testing and drug discovery later this year. According to Stephen Minger, R&D director for cell technologies at GE Healthcare, the initial interest level is huge. After seeing the cells’ attributes, potential partners have said, “If

nature biotechnology volume 28 number 6 JUNE 2010

R. BICK, B. POINDEXTER, UT MEDICAL SCHOOL / SCIENCE PHOTO LIBRARY

© 2010 Nature America, Inc. All rights reserved.

Burgeoning stem cell product market lures major suppliers

Beating heart cells. Firms are using industrialized quantities of stem cell–derived human cardiomyocytes to predict toxicity and screen for efficacy in a dish.

you can supply the cells that you just showed us on a routine basis, we will buy a lot of them,” Minger adds. Cardiomyocytes are the initial target of many programs. The differentiation protocols for these cells are robust, and they have a clear visual readout: the cells contract or ‘beat’ in vitro. Hepatocytes, for example, require trickier protocols and several biochemical readouts to determine whether they have differentiated appropriately, Minger says. Both CDI and GE Healthcare are working towards large-scale production of hepatocytes and other differentiated cell types for use in toxicology and drug discovery screening. One drug company that is embracing the use of such cells in preclinical research is Roche of Basel. The company began a collaboration with CDI in March 2008 to test drug development candidates for their potential to cause toxicity on cardiomyocytes derived from hESCs and iPSCderived cells. With the Institute for Stem Cell Therapy and Exploration of Monogenic Diseases (I-STEM), for instance, an academic center near Paris, Roche is matching its high-throughput screening expertise with I-STEM’s hESC-derived neuronal cells to search for new drug candidates for neurodegenerative and psychiatric disorders. Earlier this year, Roche began working with stem cell researchers at Massachusetts General Hospital in Boston and Harvard University in Cambridge, Massaschusetts, to develop cellular models for metabolic and cardiovascular diseases using

535

NEWS

in brief

© 2010 Nature America, Inc. All rights reserved.

GMP cell lines to order Eden Biodesign and Millipore have struck a deal to offer a service of mammalian cell lines on demand for companies developing antibodies and protein therapeutics. The collaboration marries Millipore’s Ubiquitous Chromatin Opening Element (UCOE) expression technology with Eden’s cGMP production. Eden, a contract manufacturing organization based in Liverpool, England, has had a long relationship with Millipore, the life science research and biomanufacturing products supplier located in Billerica, Massachusetts. The new partnership is a “natural fit” says Roger Lias, president of Eden Biodesign’s US office and group commercial director. The UCOE vector yields cell lines with a high level of gene expression that are both productive and easy to scale up for clinical trials and commercial supply. According to independent consultant Linda Somerville, based in Peebles, Scotland, the UCOE system also has the potential to shorten production time considerably compared with traditional transfection methods. Eden scientist David Simpson originally developed the UCOE technology before it was acquired by Millipore in 2005. On 1 March, Millipore was involved in a $7.6 billion transaction in which it became wholly owned by Merck KGaA of Darmstadt, Germany. The purchase has expanded Merck’s remit, traditionally focused on chemicals, into life sciences and biomanufacturing. It is also a bonus for Eden, says Lias: “The deal with Eden will help drive more Merck customers to the UCOE technology.” Susan Aldridge

Open-access fermenter The UK’s first open-access facility will soon be available for firms wanting to ramp up biotech processes. The UK’s Centre for Process Innovation (CPI) is expanding the capacity of its National Industrial Biotechnology Facility (NIBF) in Wilton from 1 to 10 tons to provide startups and established businesses with equipment and expertise for proof-of-concept development. Companies will be able to use the facility—in which projects may be backed by governmental funding or by private contracts—to make pilot batches of molecules, to de-risk their technology or to figure out how to scale up production processes. “They might want to rent some space, they might want to use the equipment in collaboration with my team, or they might want us to develop a process package for implementation in a manufacturing plant,” says Chris Dowle, director of sustainable processing at CPI. “We’re very flexible.” Similar sorts of services have been around for some time, he says, but the improved NIBF site will be a first in terms of the large scale and the versatility of the equipment. For instance, a bespoke continuous fermentation system will be on offer as well as ‘plug and play’ machinery that can purify biofuels and other potentially marketable biochemicals. The plant will not produce biotherapeutics. A similar project is being developed in Leuna, Germany by the Munichbased Fraunhofer Institute and is scheduled to open next year. Asher Mullard

536

Table 1 Selected companies and stem cell products Company

Product

Pluripotent cell source

Availability

Cellular Dynamics International

iCell Cardiomyocytes

Human iPSCs

Dec 2009

GE Healthcare

Cardiomyocytes

Human ESCs

Later in 2010

Lonza

Cor.At Cardiomyocytes

Mouse ESCs

Jan 2010

iPSC-derived cells. The emphasis is on exchanging ideas with partners with particular cellular expertise and finding ways to make drug discovery more productive, says Matthias Steger, Roche’s global alliance director for stem cell research. The cells bridge the gap between preclinical and clinical research, he adds, and different stem-cell platforms are likely to become widely adopted as the tool of choice for finding new drugs “in two to three years.” Three large European drug makers have also come together with the UK government to form the Stem Cells for Safer Medicines consortium. This nonprofit company, which was launched in 2007 and includes Roche and London-based GlaxoSmithKline and AstraZeneca, was founded with both public and private funds to develop hESCs for early safety testing of new medicines and to establish a set of best practices. The first phase has been to optimize differentiation protocols for generating hepatocytes and cardiomyocytes, says Julie Holder, preclinical director of the stem cell performance unit at GlaxoSmithKline. Both cell types have been produced from optimized differentiation protocols and researchers are now testing the cells using assays with small molecules to ensure they are fit for purpose. Even as stem cells are moving toward mainstream use within pharma, much of the scientific expertise in stem cells remains outside large companies. GE Healthcare’s Minger expects that the large-scale production, particularly of ESCderived products, will remain with biotech companies with specialized expertise. Human ESCs are very sensitive to changes in their environment, such as cell density, the matrix on which they’re grown and the concentration of growth factors. “It’s a lot of work,” Minger says. “The expertise that’s required is not readily available.” For routine screening, cell lines derived from ESCs are still considered the first choice, say Steger and Rao. Although the iPSC field is moving away from using viruses to incorporate the reprogramming factors, questions linger over whether iPSCs are completely reprogrammed and the equivalence of iPSCs to their ESC counterparts. But even as comparisons of iPSC- and ESC-derived cells continue, iPSC technology offers the unique opportunity to develop diseasespecific cell models, such as motor neurons with disease phenotypes, Rao says (Nat. Biotechnol. 27, 977–979, 2009). Pharma companies also have libraries stuffed with compounds that have yet to be tested on

stem cells. These small molecules could have a variety of unique activities, such as directing cellular reprogramming or differentiation. On April 15, Pfizer of New York and stem cell reagent company Stemgent announced a partnership that will allow stem cell researchers to run assays with some of Pfizer’s proprietary compounds. “Pfizer has a lot of interesting small molecules and would like to find out more about what they can do, particularly in the regenerative medicine space,” says Ian Ratcliffe, president and CEO of Stemgent in Cambridge, Massachusetts and San Diego. “They want to put those into the scientific community in a controlled fashion.” At the same time, Pfizer’s regenerative medicine chief scientific officer Ruth McKernan has joined Stemgent’s scientific advisory board. “It’s nice to keep a pulse on what people are thinking and what’s important in stem cell research and what’s important in controlling stem cells,” McKernan says. Even with the current emphasis on toxicology and drug discovery tools and disease models, pharma companies are also eyeing a future landscape that includes regenerative medicine. “Right now it seems like almost every pharma company has some kind of investment in some regenerative medicine company,” Rao says. “They’ve all made some kind of bet that primary cells or stem cells are going to be useful for the next generation of drugs.” Pfizer founded its regenerative medicine division in December 2008. Roche has a center in Pittsburgh that focuses on cellular therapeutics. Some companies are focusing on financial investments in companies with stem cell and related technologies. For example, New Brunswick, New Jersey–based Johnson & Johnson’s venture capital group has invested in San Diego’s Novocell. On February 28, Merck announced that it would buy Bedford, Massachusetts–based Millipore, a supply company which already sells several stem cell–based products, including a differentiation kit for mouse iPSCs. According to Robert Shaw, Millipore’s commercial director for the project, Millipore started an internal initiative last year aiming to scale up the production of iPSC-derived human hepatocytes and neuronal cells for drug discovery and eventually clinical applications. The company is currently validating these products to make sure that they have the expected phenotypes and metabolic function, Shaw says. The hope is to be selling cells, reagents and other tools for large-scale cell production “very soon.” Sarah Webb Brooklyn, New York

volume 28 number 6 JUNE 2010 nature biotechnology

news

Weeds are becoming increasingly resistant to glyphosate, a report from the US National Academy of Sciences (NAS) released in April has found. The driving force, according to the report, is farmers’ dependence on the weed killer accompanied by the widespread adoption of genetically modified (GM) herbicide-tolerant crops. Seed makers are hoping to forestall the problem by developing GM crops with ‘stacked’ traits that tolerate multiple herbicides. But weed scientists warn that if farmers manage these new crops in the same way as they managed their glyphosate-tolerant predecessors, weeds will simply become resistant to the new technologies. “The number of weed species evolving resistance to glyphosate is growing,” the report says. At least eight weed species in the US have become resistant to glyphosate, and the trait is prevalent in areas where farmers grow crops that have been genetically engineered to resist the weed killer. The authors are calling for “national attention” to the weed problem. Glyphosate was first commercialized in 1974 by St. Louis -based Monsanto under the brand name Roundup. It became a key weed control tool for farmers in 1996 when Monsanto developed a GM soybean variety called Roundup Ready that expresses a gene encoding enol pyruvate shikimate-3-phosphate synthase from the microbe Agrobacterium tumefaciens conferring tolerance to the herbicide. Growers loved it. Roundup killed a broad array of weeds without killing their Roundup Ready soybeans. Seed makers went on to commercialize glyphosatetolerant corn, cotton, canola and sugar beets. After Monsanto’s patent on Roundup expired in 2000, companies commercialized dozens of glyphosate formulations. “We have found something that really works, and we have really used it,” says William Johnson, a weed scientist at Purdue University in West Lafayette, Indiana. The system was so convenient that growers began relying exclusively on glyphosate for weed control—a recipe for resistance. “Glyphosate had been used forever, so people were not as cautious,” says Yves Carrière, an author of the report and an entomologist at the University of Arizona in Tucson. The first report of glyphosate resistance associated with a GM glyphosate-tolerant crop came in 2000 in Delaware in a species of horseweed. Since then, a new resistant weed species has been reported nearly every year in the US and South America. Glyphosate resistance has also been reported in Australia, South Africa, Europe, China, Malaysia and Canada. Glyphosate isn’t the only herbicide facing resistance from weeds. At least 195 weed species have evolved resistance to 19 herbicides, accord-

ing to the NAS report. Glyphosate, however, is one of the most economically important. In the US alone, over 90 million pounds are used annually, and it is the number-one-selling herbicide worldwide. Biotech seed makers are responding to the problem by developing new herbicide-tolerant crops (Table 1). “The strategy is that the crops would be tolerant to two or more herbicides,” says John Soteres, global weed resistance management lead at Monsanto. The hope is that the crops will allow growers to use a wider variety of herbicides and weed control practices. For example, if a crop is tolerant to both glyphosate and glufosinate, a farmer could alternate the herbicides, or use them in combination. The most likely candidates for development, according to the NAS report, are genes that confer resistance to herbicides such as dicamba, glufosinate, hydroxyphenylpyruvate dioxygenase (HPPD) inhibitors, 2,4-D and other synthetic auxins, acetolactate synthase (ALS) inhibitors and some acetyl-CoA carboxylase (ACCase) inhibitors. Most companies are focused on combining one or two of these traits with glyphosate tolerance. “You don’t want to throw out glyphosate completely,” says Nicholas Storer, a global science policy leader for biotech at Dow AgroSciences in Washington, DC. But some scientists say the next generation of GM crops will only buy growers more time until another group of weeds evolves resistance. “This is an incredible technology that is being compromised because of farm management decisions and there is nothing in the pipeline that is going to fix that,” says Michael Owen, an author of the NAS report and a weed scientist with Iowa State University in Ames. Purdue’s Johnson predicts that it will take 7 to 9 years for some weeds to evolve resistance to the next generation of herbicide-tolerant crops, if the new technologies are adopted with the same celerity as Roundup Ready. Because technology alone won’t solve the weed problem, companies in recent years have “taken up the banner for stewardship,” says Owen. “As scientists, we can’t prevent resistance to any herbicide, and that being the case, it comes down to basic farm management,” adds Monsanto’s Soteres. The company provides educational materials to growers through sales meetings, technical use guides and websites (e.g., http://www.weedresistancemanagement.com/). The materials encourage crop rotation and the addition of a nonglyphosate herbicide to their weed control program—recommendations that fall in line with messages from university extension scientists and the NAS report. “What

nature biotechnology volume 28 number 6 JUNE 2010

BILL BARKSDALE / AGSTOCKUSA / SCIENCE PHOTO LIBRARY

© 2010 Nature America, Inc. All rights reserved.

Glyphosate resistance threatens Roundup hegemony

These marestail plants infesting a crop of wheat in Tennessee are resistant to glyphosate herbicide Roundup.

we’re preaching is that a diversified program is needed,” says Soteres. “To rely on a single herbicide exclusively can lead to resistance.” Monsanto hasn’t always taken that stance, however. “Monsanto had a very aggressive marketing campaign in the late 1990s saying that weeds would not develop resistance if [growers] used herbicides at the right time and at the right rate,” says Johnson. “They told users that the resistance mechanisms were very complex, not commonly found in plants grown in the wild, and difficult to insert into the crop,” adds Carol Mallory-Smith, a weed scientist at Oregon State University in Corvallis, “Monsanto discouraged the use of other herbicides in the system.” For example, to qualify for Monsanto’s ‘Roundup Rewards’ program―a kind of warrantee for crops―growers in the late 1990s could only use Roundup and certain approved herbicides on Roundup Ready crops. The company has since changed the qualifications for its rewards program. A spokesperson for Monsanto says the company has “never restricted the use of nonglyphosate herbicides in Roundup Ready crops” but that the company also did not initially promote weed control programs that included other herbicides. “Based on what we knew in the early years, we believed our recommendations were appropriate,” says Eric Sachs, director of global scientific affairs at Monsanto.

537

NEWS

in brief

© 2010 Nature America, Inc. All rights reserved.

SBIR grants wax Awards under the Small Business Innovation Research (SBIR) program have just been given a boost. As of March 30, the cap for SBIR phase I awards has risen from $100,000 to $150,000, and for phase II awards from $750,000 to $1,000,000. The increases are intended to take account of inflation since 1992 when the threshold amounts were last set by Congress. “This will have an important positive impact at a critical [juncture] in the aftermath of the nation’s great recession,” says Simcha Jong, university lecturer in management science and innovation at University College London. Jong says that, historically, the SBIR program helped forge links between university science and industry and, at this pivotal time, could help kick-start the US job engine. The Senate has passed a bill to extend the SBIR and related Small Business Technology Transfer through July 31 (Nat. Biotechnol. 27, 1065–1066, 2009). Even more generous than SBIR grants are the new Small Business Helping Investigators to Fuel the Translation of Scientific Discoveries (SHIFT) awards launched on March 5 by the US Department of Health and Human Services. These awards, aimed at fostering translational research, offer companies up to $2.65 million over five years. “The main point is to encourage current academic researchers to apply, and use it to move to biotech,” says Jiwu Wang, president and CEO of Allele, a San Diego-based company that has taken products to market with SBIR support. “It is a great idea.” Emma Dorey

Relief over stem cell lines The US National Institutes of Health (NIH) announced the addition of 13 lines to its Stem Cell Registry. The news was cheered by the research community, as the two most widely studied lines— H7 (WA07) and H9 (WA09) owned by the WiCell Research Institute of Madison—were included in the batch approved by NIH director Francis Collins. The total number of NIH-approved human embryonic cell lines in the registry, and thus eligible for federal funding, has risen to 64 as of April 29. These recent approvals ease frustrations among scientists who watched President Obama’s March 9, 2009 Executive Order—welcomed at the time and intended to remove barriers for such research—later backfire when NIH insisted that cell lines used during the George W. Bush presidency be reevaluated under revised ethical guidelines that NIH began following in July 2009 (Nat. Biotechnol. 27, 681, 2009). Playing down the vociferous complaints since then, Collins says the approvals this April should enable researchers to “continue their studies without interruption, and we can all be assured that valuable work will not be lost.” Even though Collins seems to discount projects that were disrupted during that interval, NIH-supported human embryonic stem cells research now is poised to get back on track. The H7, H9 and other recent approvals are indeed a “huge relief,” says bioethicist Christopher Scott, who directs the Program on Stem Cells and Society at Stanford University. Jeffrey L Fox

538

Table 1 Selected crops in development tolerant to two or more herbicides Company (location)

Crop

Bayer CropScience (Monheim am Rhein, Germany)

Soybean

HPPD inhibitors, glufosinate, glyphosate

Cotton

Glufosinate, glyphosate

Corn

Phenoxy auxins (e.g., 2,4-D), aryloxyphenoxypropionate ACCase inhibitors (e.g., quizalofop-p-ethyl), glyphosate

Dow Agrosciences

Monsanto

Cotton, soybean

2,4-D, glyphosate

Corn, cotton

Dicamba, glufosinate, glyphosate

Soybean

Dicamba, glyphosate

Corn

Dicamba, glufosinate, glyphosate

Pioneer Hi-Bred (Johnston, Iowa) Corn, soybean Syngenta (Basel)

Herbicides tolerated

Soybean

ALS inhibitors, glyphosate HPPD inhibitors (e.g., mesotrione), glufosinate, glyphosate

HPPD, hydroxyphenylpyruvate dioxygenase; 2,4-D, 2,4-dichlorophenoxyacetic acid; ACCase, acetyl coenzyme A carboxylase; ALS, acetolactate synthase.

Despite the recent efforts by companies and continued efforts by university scientists, the message to ‘diversify’ doesn’t always stick with growers. According to the NAS report, growers are reluctant to stop using glyphosate even when facing signs of resistance in their fields. “For controlling problematic weeds, [growers] prefer increasing the magnitude and frequency of glyphosate applications, using other herbicides in addition to glyphosate, or increasing their use of tillage,” the authors of the report wrote. A 2009 survey sponsored by Monsanto found that >75% of farmers were aware of the potential for weeds to develop resistance to glyphosate. But less than half of those farmers said they believed that rotating crops and alternating herbicides would be effective practices for minimizing weed resistance.

Growers can’t be legally forced to reduce their glyphosate use. Unlike pesticide use, herbicide use is not regulated by the US federal government. Regulations wouldn’t be practical anyway, says Owen. “It can’t be done in a way that would keep resistance from evolving,” he says. “The impossible part would be enforcing the regulations.” Weed scientists say they hope that the NAS report will at least raise awareness among the general public about the weed resistance problem. The 253-page report also emphasized that insectresistant crops help farmers reduce pesticide use, and found that overall, “planting of [genetically engineered; GE] crops has largely resulted in less adverse or equivalent effects on the farm environment compared with the conventional non-GE systems that GE crops replaced.” Emily Waltz Nashville, Tennessee

in their words

BIO’s beastly bugs When was the last time someone called E. coli cute? Many did at this year’s Biotechnology Industry Organization (BIO) Annual Meeting in Chicago where conference goers were treated to a collection of giant fuzzy microbes courtesy of biomanufacturers SynCo Bio Partners. MRSA or HIV anyone?

“Science is not a 100-yard dash. It is a marathon—a marathon run by a relay team that includes researchers, patients, industry experts, lawmakers and the public.” While testifying to a congressional subcommittee NIH Director Francis Collins stresses the long timelines involved in translating $32.2 billion of proposed funding into products. (GenomeWeb News, 28 April 2010) “We’ve been selling it since 1998, probably 200 million pounds from Honolulu, and not a single bad case of anything going wrong.” Papaya farmer Ken Kamiya makes the case for transgenic papaya recently approved in Japan, where a single papaya can fetch $10. (Honolulu Advertiser, April 25 2010) “The worst case of corporate bullying I’ve ever seen.” Attorney Ray Chester on Botox producer Allergan’s (Irvine, CA) campaign to recover $460,000 in legal costs from Dee Spears, who unsuccessfully sued the drugmaker over the death of her 7-year old daughter with cerebral palsy who had received the treatment. (Orange County Register, 20 April)

volume 28 number 6 JUNE 2010 nature biotechnology

news

© 2010 Nature America, Inc. All rights reserved.

Obama appoints bioethics panel to offer practical advice In April, President Barack Obama named 11 more members to the Presidential Commission for the Study of Bioethical Issues. They will join commission chair Amy Gutmann, president of the University of Pennsylvania, and commission vice-chair James Wagner, Emory University president, who were appointed last year. The new commission replaces the controversial President’s Council on Bioethics set up by President George W. Bush in 2001, which contained several members of the pro-life lobby. The 13 newly appointed commissioners are not preponderantly “professional bioethicists” but rather come from other fields “at the intersection of science, technology and ethics” (Table 1). Part of the idea with this departure from mainstream bioethicists is to reach beyond biology and medicine to involve those working with “hardware, software and related technologies such as robotics,” officials say. The president’s new panel is expected to react rapidly and provide practical guidance, a radical departure from the Bush-era commission, which favored discussion and was often accused of producing reports with ideological leanings. Obama’s commission differs from the Bush council in several ways. First, the new commission is lean, with only 13 members to “keep the group nimble and facilitate discussion and consensus building.” Second, its members are asked not to engage in “arcane philosophical discussions” but to provide the president and administration with practical, policy-oriented, ethics-based recommendations. For that reason, the commission includes several insiders who work for federal agencies, a shift that is meant to keep discussions and advice from straying outside “the complex framework of federal policymaking processes and procedures.” Jeffrey L Fox

Table 1 Who’s who—new bioethics appointees Panel member

Group

Post

Lonnie Ali

Advocate

Wife of former heavyweight boxing champion Muhammad Ali is an advocate for research on Parkinson’s disease

Anita L. Allen

Academic

Professor of law and philosophy and also deputy dean at the University of Pennsylvania Law School, and senior fellow in the Bioethics Department, School of Medicine; served in the 1990s on the National Advisory Council for Human Genome Research

John D. Arras

Academic

Porterfield professor of biomedical ethics and professor of philosophy at the University of Vierginia; longtime fellow of the Hastings Center

Barbara Atkinson

Academic

Executive vice chancellor of the University of Kansas Medical Center and executive dean of the University of Kansas School of Medicine

Nita A. Farahany

Academic

Associate professor of law and philosophy at Vanderbilt University; focuses on legal, philosophical and social issues arising from developments in behavioral genetics and neuroscience

Alexander Garza

Government Assistant secretary for health affairs and chief medical officer for the Department of Homeland Security; specialized in emergency medicine

Christine Grady

Government Acting chief of the Department of Bioethics at the National Institutes of Health Clinical Center; focuses on research subject recruitment, incentives and vulnerability

Stephen L. Hauser

Academic

Professor and chair of the Department of Neurology at the University of California, San Francisco; focuses on the genetic and immune basis of multiple sclerosis

Raju Kucherlapati

Academic

Professor in the Harvard Medical School Department of Genetics and the Department of Medicine at Brigham and Women’s Hospital; was the first scientific director of the Harvard Medical School-Partners Healthcare Center for Genetics and Genomics

Nelson Michael

Academic

Director of the Division of Retrovirology at the Walter Reed Army Institute of Research; directs the US Military HIV Research program

Daniel Sulmasy

Academic

Franciscan Friar and chair in medicine and ethics in the Department of Medicine and Divinity School, and associate director of the MacLean Center for Clinical Medical Ethics at the University of Chicago

nature biotechnology volume 28 number 6 JUNE 2010

in brief GSK’s RNA splash Antisense-drug developer Isis Pharmaceuticals and GlaxoSmithKline (GSK) have forged a collaboration to develop drugs for rare diseases that could earn the Carlsbad, California–based biotech up to $1.5 billion dollars in licensing fees and milestones. London-based GSK will pay Isis $35 million upfront and up to $20 million for each of the six programs, which Isis has agreed to develop to proof of concept. Isis will apply its antisense-drug discovery platform to work on novel targets in new therapeutic areas, including infectious diseases and some conditions causing blindness. “The deal is emblematic of deals that financially sound companies can enter into,” says Isis CFO and COO Lynne Parshall. “We can rely on longerterm royalties over the course of a drug’s lifetime rather than bigger upfront licensing fees,” says Parshall. Isis’ pact with the pharma comes on the heels of encouraging phase 3 trial data for the cholesterolreducing antisense drug mipomersen that Isis is developing with its partner Genzyme (Nat. Biotechnol. 28, 295–297, 2010). The latest agreement underscores GSK’s interest in nucleic acid–based therapeutics notes Lindsay Meyer, a senior consultant with Deloitte Recap, San Francisco. She highlights GSK’s recent alliances with Regulus of Carlsbad, California, and Prosensa, of Leiden, The Netherlands, and earlier with Santaris in Hoersholm, Denmark. Isis, however, stands to benefit more than other companies that have licensed their RNA-based therapeutics to GSK, Meyer believes. Janelle Weaver

Germany caps drug prices The German coalition government is putting into place new rules that will allow health insurers to influence the pricing of new medications. The changes are intended to save the healthcare system around €2 ($2.7) billion) a year. “This could impact innovation, because cutting the price of drugs will reduce the incentives for biopharma companies to invest in R&D,” says Marion Kronabel, managing director of the European Association of Pharma Biotechnology, part of the European Federation of Biotechnology. Germany allows firms to set prices for branded drugs, and prices are higher than in most countries. Under the new rules, the branded drug sector will be allowed to set prices for their products only in the first year after launch. After this, drug firms and insurers will enter negotiations, overturning the industry’s requests for two years’ price protection. The new law, which will be enacted by the end of this year, could potentially trigger patent law changes to extend a drug’s protection from generic competition, Kronabel believes. She also points at more pressing concerns, “Overall, this legal change and this approach for cost reduction will have less of a harmful influence on the biotechnology industry in Germany than the issue of taxes on R&D expenses and venture capital profits in Germany.” Suzanne Elvidge

539

N E W S f e at u r e

© 2010 Nature America, Inc. All rights reserved.

Dairy farmers are rapidly adopting molecular profiling to accelerate the process of siring cows. But this seismic shift in breeding practices is raising new questions and translating more slowly to the beef industry. Stephen Strauss reports. Following last year’s publication of the Bos taurus genome sequence1, the dairy industry has wasted little time in assimilating cattle genomics into its working practices. Only a few months ago, Illumina of San Diego announced the creation of a new bovine single-nucleotide polymorphism (SNP) chip with ten times the coverage of an earlier version. The chip’s predecessor had already been leapt upon by breeders keen to integrate the new genomic information into their siring practices. But although uptake of the technology has been rapid, questions remain concerning the ability of markerassisted breeding programs to ultimately predict complex traits, such as meat quality or even milk composition and yield, and the longterm effects of such tests on the meat and dairy industries remain unclear. The genetic cream In the summer of 2008, a group of senior managers at the Shawano, Wisconsin–based artificial insemination (AI) company Genex Cooperative huddled together to discuss the possible effects on their future business of the recently released 54,000-SNP cattle genome chip known commercially by the rather awkward name BovineSNP50 BeadChip. The question of the day was, would their dairy farmer customers buy semen from Genex’s bulls who hadn’t first been ‘proven’? Obtaining an accurate prediction of genetic value—what is called a sire proof—is a halfcentury-old procedure. In it, upwards of 100 randomly selected cows are artificially inseminated, give birth, and—when their calves grow old enough to produce milk—the offspring are tested to see if they and their milk exhibit desirable traits. A genetic prediction based on roughly 100 daughters will generally result in >90% predictive accuracy. The drawback is that the process is expensive and time-consuming, taking 5 or 6 years to complete and costing up to $50,000. Even more disconcerting is that only about one bull in ten that go through this process is eventually judged genetically superior enough to qualify as a high-quality stud. The Illumina BovineSNP50 BeadChip had only a ~65–70% accuracy rate in trait inheritance prediction. Even so, it allowed markers 540

associated with high-quality traits to be viewed at birth, providing substantial benefit over the traditional sire proof procedure. The Genex debate was about how many clients would value genomics’ greater speed over the greater accuracy of ‘being proven’. The consensus, says Roy Wilson, technology development manager for Genex, was that the intrinsic conservatism of farmers meant that, at best, in the short term, only 15% of the company’s business would switch over to unproven, but genomically highly promising bulls.

adoption of this technology has been breathtaking,” says Stephen Moore, a University of Alberta professor of agricultural genetics, who has been working with agricultural genomics companies to identify SNPs that contain genetic traits that are important for farmers. “No sooner was the chip designed than it was being used,” he adds. “We are in the front end of a major change in raising cattle. Some people are using words like ‘disruptive’ and ‘quantum leap’ to describe what is happening,” says Ronnie Green, senior director of global technical services, at Pfizer Animal Genetics in Kalamazoo, Michigan, which has been selling a DNA screening test for desirable traits in cattle (Table 1). “Whatever word you use, we are in a time of real upheaval.” The reasons the technology has proven revolutionary are varied. In part, the change has taken place in dairy farming, because the infrastructure was already in place to rapidly integrate genomic data. Because a single bull is tremendously valuable in terms of siring many cows—and therefore has a massive impact on the genetics of a dairy herd—for several decades farmers, breeding organizations and AI companies have been collecting as

They were not even close “In the year to date, around 40–45% of our sales are from sires with no milking daughters,” said Wilson in midDecember of 2009. What this number doesn’t capture is the dramatic reconfiguration of the breeding business that took place from being able to get semen to market in one-third the time. In 2008, Genex was progeny-testing bulls in 2,000 herds. By the end of last year, this had shrunk to 160 herds. Two years ago, 300 bulls were sent through progeny test- Genomic gold. Observer, deemed the top Holstein bull according to his ing; with the advent genetics, is the gold standard for bulls in the US. of the Illumina SNP profile predicting those sires with a greater many bull surface (phenological) traits as they likelihood of carrying desired traits, only 180 could (e.g., vigor and haunch size) that might to 200 bulls were judged candidates worthy of be linked to subsequent milk production and being further tested. And the one in ten bulls other useful traits in their daughters. According to Curt Van Tassell, a US who potentially might bear the title ‘superstud’ had through genetic prescreening grown to Department of Agriculture (USDA) research geneticist based in Beltsville, Maryland, who become one in five bulls. has actively worked with companies trying to make use of the new genomic information, A revolution waiting to happen The uptake and implementation of genetic pro- literally millions of data points have been filing by breeders is nothing short of astonish- gathered in the US and Canada since records ing, particularly to many working in or with were started 40 years ago. Data on 16 milthe dairy industry. “In dairy cattle, the rate of lion dairy cows are part of a collection that volume 28 number 6 JUNE 2010 nature biotechnology

Select Sires, Plain City, Ohio

Biotech breeding goes bovine

news f e at u r e

Table 1 Selected companies with a focus in agricultural genomics Company

Technology

Affymetrix

GeneChip Bovine Mapping 10K 10,000 SNPs from bovine genome sequencing project (92%) and Australia’s Commonwealth Scientific and Industrial Research Organisation (8%)

Product

On the market

Status

Igenity

Measures genetic component of 15 desirable traits

On the market

Ingenity Profile (dairy) Angus Genetics

54,001 highly informative SNPs uniformly distributed across the entire genomes of major cattle breed types

BovineSNP50 DNA Analysis BeadChip

On the market since 2007

500,000 to 800,000 SNPs with information gathered from >20 different breeds

High-density Bovine BeadChip

Announced in January of 2010, first shipments in second quarter of the year

Pfizer Animal Genetics

50,000 SNPs for 14 traits

HD 50K for Angus, GeneStar MVP, GeneStar Black, GeneStar Tender Elite, SireTRACE, SureTRAK and genetic defect testing

Launched January 2010

Metamorphix (Calverton, MD, USA)

Genomic services offered for DNA-based genetic parent verification, diagnostic testing

Tru-Marbling

On the market

© 2010 Nature America, Inc. All rights reserved.

Illumina

Tru-Tenderness DNA certified beef programs Horned polled diagnostics

Quantum Genetics (Saskatoon, Saskatchewan, Canada)

Genome manipulation to control obesity and fat deposition

Quantum Management Protocol

Under development

Genetic Visions (Middleton, Wisconsin)

Services to test for genes influencing coat color, animal health and viability, production traits

Genetic marker tests, various tests

Launched

Performagene Livestock

Launched

DNA Genotek Sample collection services (Kanata, Ontario, Canada)

has been in existence since the 1960s at the Animal Improvement Programs Laboratory in Beltsville. The data, which are available to breeders, researchers and AI companies alike, provides a pedigree proofing-based scale that shows how much more money offspring from one bull might earn than offspring from an inferior one based on its genetics (Box 1). Each year, more is collated into the collection from 40–50% of US dairy cattle. Thus, a huge database of 100 or so genetically linked traits has been amassed onto which the SNPs from genome sequencing efforts can be associated. Equally importantly, in an effort to link milk volume and quality with bull genetics, AI has become the method of choice for dairy farmers. Thus, it is employed by >80% of farmers breeding dairy herds, whereas only 7% of farmers use it for beef. What also drove the appeal of genomic testing for dairy farmers was the fact that the majority of cows in North America and in the developed world are Holsteins. In North America, Van Tassell says, Holsteins once accounted for >95% of milk cows; even today, its herd prevalence is still >90%. This means that any anomalies due to different SNP trait locations in different breeds are eliminated. The existing infrastructure within the dairy industry also made collaborative genomic research between companies, university scientists and the USDA easy to get off the

ground. Specifically, seven AI companies, two in Canada and five in the US, joined with the USDA, the University of Alberta, the University of Missouri and Illumina to correlate SNP locations to phenotypic data. In Canada, the University of Guelph and the Canadian Dairy Network, which is in charge of national evaluations for dairy cattle, also participated. In exchange for their financial participation and the providing of both DNA data and semen, the AI organizations were given a fiveyear exclusivity on the use of genomic evaluations for young bulls. Thus, another impetus for swiftly applying the genomic findings in dairy cattle is that the AI companies had an intrinsic stake in using the milk cattle genomic information quickly while their monopoly still could convey a business advantage. Finally, there was a Moore’s Law factor. As Jacques Chesnais, senior geneticist for Semex Alliance (a dairy breeding consortium owned by four AI cooperatives in Canada) points out, the Illumina chip contained twice as many traits and sold for half as much as its main competitor from Affymetrix of Santa Clara, California. Where’s the beef? While genomic information is transforming the dairy sector, the situation is very different for beef breeders. There are several reasons for the disparity.

nature biotechnology volume 28 number 6 JUNE 2010

One is that the diversity of beef cattle is greater than that of dairy. “There are lots and lots of variations between breeds,” says Moore, who is in the process of genotyping Angus and hopes to begin to do the same thing for a bull breed from the tropics. “An allele that might be a good predictor of trait in one breed might flip and actually become a negative indicator in another breed.” This difference is a big stumbling block in the North American beef industry. Despite the predominance of three breeds (Angus, Hereford and Simmental make up 60% of the US beef herd), a substantial minority (40%) of the beef herd is drawn from over 80 breeds. There is also a smaller amount of information available that associates meat product quality traits with SNP readings. And the list of qualities is much more diverse. Instead of milk quantity and fat composition, beef breeders must look at meat tenderness, fat thickness, ribeye area, marbling and yield grade among numbers of other things. There is also a price differential growing out of the fact that while dairy cows generate income over their milk-producing lifetimes, beef cattle’s value only occurs when they are slaughtered. Overall, meat breeders have paid much less attention to the genetic quality of bulls because the real money is made when animals are brought to feedlots and fattened up. This accounts for the low amount of AI usage in beef cattle and is a reason why beef cattle 541

N E W S f e at u r e

© 2010 Nature America, Inc. All rights reserved.

Box 1 How much is your cow worth? Researchers at the USDA’s Animal Improvement Programs Laboratory (AIPL), in collaboration with academia and industry, have been turning the art of calculating the value of cattle into a science long before SNP data from chips. A cow’s so-called net merit weighs traits that produce income, like milk fat and protein production, against those that cost the rancher, like the cost of feed consumed by a calf before she reaches milking age. The data come from performance records that have been collected on dairy farms for over 30 years, pedigree data provided by farmers and breed associations and, since 2009, SNP data from the Illumina BovineSNP50 BeadChip. The average cow born in 2005 is used as a reference point, which is called the base population, and has a net merit of $0. Cows with positive values will generate more profits relative to the base population, and those with negative, less. The top Holstein bull as of April 2010 is named Observer (see photo) and has a net merit of +$848, which means that his daughters will each earn $848 more during their lifetimes (on average) than daughters of an average bull. The US was the first to incorporate SNP chip data, according to John Cole, research geneticist at AIPL. They currently include data on 43,385 SNPs in analyzing the Brown Swiss (~1,500 genotypes), Holstein (~40,000 genotypes) and Jersey (~4,000 genotypes) breeds. The effects of each SNP are calculated for each trait, which number around 30, with some variation from breed to breed. The SNP genotypes were originally produced by the Bovine Functional Genomics Laboratory at USDA, but that service is now provided by commercial laboratories. Owners of the animals provide a source of DNA for genotyping, and pay to have a genotype produced. The data are entered into the national dairy database and the owners of the cows and bulls receive reports about their animals. SNP effects are recalculated for each trait at AIPL as more data become available, says Cole. Laura DeFrancesco

growers are not interested in a genetic test that can cost somewhere between $200 and $250 dollars per animal. Covering the bases The high price point is one of the factors that might be addressed by ongoing innovations and improvements to the genetic tests. The race is on both to exponentially expand the number of SNPs that can be measured and lower the price. In terms of SNP expansion, at the end of December, Van Tassell was wrestling with the problem of verifying 900,000 SNPs for Illumina in time for the company to launch a next version of their bovine chip in January. One hope is that an exponentially increased number of SNPs on a chip will allow AI and other companies to provide tracking for traits that have weaker genetic associations. “Something that has more markers has a greater statistical power in the association of traits with markers,” says Mike Thompson, global manager at Illumina’s animal division. Another hope is that the chip will contain enough information to allow the disconnect between breed difference and SNP trait readings to be resolved. In January, Illumina announced it was accepting orders for a >500K chip that contains genetic data from >20 breeds of cattle. At the same time, the San Diego–based com542

pany has let it be known that it is also going to be releasing a 3K bovine SNP chip, which is rumored to cost somewhere between $30 and $50. Here, the idea is that although there are not as many SNPs being tested, the ones that are will be of greatest interest to cattle growers and dairy farmers. University of Alberta’s Moore says he has done an as yet unpublished study using the 3K SNP chip and found “the results look a lot cleaner than the 50K one. All you do [at 50K] is increase the noise level.” But on the horizon is a holy grail of the intersection of Moore’s Law and bovine genomics—a beyond cheap test. Van Tassell says he has begun working with biotech companies Fluidigm and Sequenom in pursuit of a bovine DNA test that costs $10 or less. “That’s a number that resonates because it is analogous to the price of a pizza,” says Van Tassell, “that value seems to be a tipping point for very large-scale adoption.” It is also a price that is low enough to encourage every cattle farmer in North America— dairy and beef—to give all of their animals a genomics profile. Going global To facilitate the association and mapping of traits with the increasing numbers of SNPs that appear on Illumina BeadChips, as well as to understand breed differences, many in the

cattle industry believe that testing will need to expand beyond the narrowness of a locale or even a country. “One of the things that we are discovering with the application of this technology to real populations is that nobody has an adequate number of animals to characterize the sequences we are describing,” says John Pollak, a Cornell University professor who is director of the National Beef Cattle Evaluation Consortium. And driven by the need to get more information, countries that formally guarded their animals’ genomic qualities as a competitive advantage are coming together in the global marketplace. For example, last October a group of European livestock associations—UNCEIA (the French National Association of Livestock & Artificial Insemination Cooperatives), CRV (an international cattle improvement organization with headquarters in Arnhem, the Netherlands), DHV and vit (a German national umbrella organization of the Holstein breeding industry and German computing center of cattle data) and VikingGenetics (DanishSwedish cattle breeding association)—came together to form EuroGenomics. The organization is devoted to using their collective 16,000 proven Holstein bulls to increase the reliability of genomic testing. A similar collaboration pooling the genomic information of Brown Swiss cattle found in Italy, France, Austria, Switzerland, Slovenia, Germany, Canada and the United States has recently been initiated. A changing business Although getting more information is good, managing all the information is another matter. One issue that has begun looming in peoples’ minds is information overload. How exactly will a farmer deal with breeding and herd management decisions in a universe where complex traits are governed by hundreds of genes that may be found in numerous DNA locations? Already there are breeding calculators that try to make this easier, but in the short term, AI companies are beginning to change their business models when speaking with farmers. Lyle Kruse, vice president of US market development for Select Sires, a Plain City, Ohio–based federation of AI cooperatives, says that increasingly they find themselves having to act as sort of genetic consiglieres for their customers. “A lot of customers are really busy; genetics and the investment in reproduction take up a small part of day-to-day demands. They rely on us to focus on what to use and how to use it. We have a group of people who are mating evaluators. They go out and actually break down a cow into 16 traits. They will do a customized mating for a herd based on sire selection and

volume 28 number 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

news f e at u r e traits farmers want to focus on,” he says. This customized genetic counseling feeds into the question of exactly what will the future of all AI companies be when the five-year monopoly they have on bull semen genomics runs out. One model sees individual farmers discovering what they didn’t know before— that one of their bulls or cows by genetic chance carries a highly desirable mix of genes. The question is, would and could that farmer sell semen or flushed eggs directly to other farmers and circumvent the AI industry middle men entirely? Part of the lure of doing that is price. In the US, companies charge $13 or $14 for a ‘unit’— the amount of semen it takes to inseminate a cow. Kruse says that it only costs ~$2 or $3 to harvest that unit. Although cooling the semen with liquid nitrogen clearly raises this cost, Kruse suggests that after the semen genotype monopoly runs out, individual farmers or groups of farmers are likely to compete with existing AI corporations. “The bottom line is that a lot more private individuals will sell semen from specific bulls.” A somewhat similar challenge may arise for existing genomics companies, particularly Illumina, Affymetrix and other chip manufacturers. After one year, the SNP information that underlies the applications, which came through collaboration with the USDA, becomes a matter of public record. This means other companies could use the information to create SNP chips that undercuts their price or better their results. “They can do it in theory,” admits Rob Cohen, senior market manager for applied markets for Illumina. He says the specter of this is forcing his company to continue their innovation efforts at breakneck speed. And then there is the possibility that the simplicity of genetic testing might undermine the programs which today link phenotypes with SNPS associated with desirable traits. If genomics tells farmers with greater reliability what traits have been passed on, there may be less incentive to gather trait information. “We can lose the tests that actually help maintain the accuracy of the genomic data,” says Kruse about this paradox. A final problem is the issue of inbreeding. If the genes from prize bulls and prize cows get into herds in a third or half of the time it previously took, then bad gene combinations can enter into breeds much faster than before.

Even before wide implementation of SNP marker-assisted breeding strategies, scientists at the USDA and University of Guelph found that 30% of the Holstein’s genome has been shaped by human breeding. More troubling still has been their observation that many of the same SNPs that are associated with higher milk production also seemed tied to lower cow fertility. Although Green points out that with the new DNA tests “for the first time we actually have a way to measure inbreeding,” Chesnais and others argue that knowledge isn’t necessarily the same as the wisdom to do the right thing. “The competitive pressure in this industry is tremendous and farmers are used to wanting semen from the very best bulls. Unless enough caution is exercised, genomics could accelerate this trend and lead to a more rapid decline in the genetic diversity of the breeds we work with.” Whereas a restricted sire pool might in the long term decrease Holstein variability, it is difficult for any single company to simply start doing the right thing genomically speaking. “All the breeding companies are competing with one another and the way to compete is to breed the best of the best, even if it may not be the most desirable approach in the longer term,” Chesnais adds. Healthier prospects? Although marker-assisted breeding has been the emphasis until now, the great hope is that SNP information, integrated with other genetic information, will prove useful to animal husbandry more generally. “What we are looking for and what we think has a much greater application than just breeding is what I would call marker-assisted management,” says Stewart Bauck of Merial’s Igenity, in Duluth, Georgia, which produces a genomic profile of both beef and dairy animals. Here, the idea is not simply to select the best cows to breed, but to drill down further and see what an individual animal’s genetic make up tells you about how to treat them. What food would make a beef cow put on weight the fastest? Are there different strains of the same breed that would thrive better in Alberta than in Arizona? Would some animals respond better to a medication than others? Although their present lack of good predictive value means trait-specific tests haven’t experienced anything like the explosion of interest that followed the Illumina

nature biotechnology volume 28 number 6 JUNE 2010

BovineSNP50 BeadChip’s release, companies such as Pfizer, Igenity and others have started marketing tests looking at specific traits, mainly in beef cattle. In the meantime, marker-assisted technology is already starting to save dairy farmers money. Brad Sayles, vice president for global marketing at Semex in Madison, Wisconsin, says that semen from unproven but genomically validated bulls sells for anywhere from $15 to $30 less per dose in Canada than doses of proven bulls’ semen. As it takes an average of four doses to impregnate a cow, this means that for each 100 cows, Canadian farmers can now save between $6,000 and $12,000 yearly. It’s also starting to earn those animals with good genomic profiles more money. Kruse says that when pedigrees were all breeders had to go on, they paid $3,000–4,000 to buy a promising bull. Now that it is easier to separate future winners from losers on the basis of a genetic profile, the price has gone up to somewhere between $6,000 and $14,000. Even so, there is caution as people move ahead with a technology that is only just now beginning to bear fruit in terms of animals mature enough to produce milk. Carl Loewith, who with his brother and son, run a dairy farm with 330 milking animals and 700 cattle in toto in Ontario, has begun inseminating their cows with semen from unproven bulls. However, because the risk of a dud sire is higher than with proven semen, they have been following the cautions of the AI companies, who advise against taking all semen from the same bull, at least in the short term. “We are told because there is still a bit of unreliability you should pick groups of bulls, maybe five or so, because one or two might not live up to their genomic potential.” Nonetheless, with cheaper marker tests on the horizon, a wider piece of the genomics industry has started quite literally knocking at farmers’ doors. “Just last week a person came by test marketing a DNA kit that wasn’t yet on the market. You could just take a swab from the cow’s nose and put it into a solution or a test tube and get a reading,” says Loewith. Those knocking apparently got a positive reception; in January, DNA Genotek of Ottawa, Ontario, released a nasal swab DNA test for cattle, sheep and swine. Stephen Strauss, Toronto 1. Anonymous. Nat. Biotechnol. 27, 487 (2009).

543

N E W S f e atu r e

Up for grabs

As recently as three months ago, it still all seemed so simple. Shinya Yamanaka, whose team at the University of Kyoto in Japan is generally acknowledged by the research community as the first to successfully reprogram differentiated cells into iPS cells1, was also the sole patent holder for the technology. But as with any other patent land grab, iPS cell intellectual property (IP) is beginning to look less and less like a one-horse race. Two other recently issued patents in the United States and United Kingdom (Table 1), each awarded to a different inventor with a potentially strong claim to priority, now stand alongside Yamanaka’s patent, which was exclusively issued in Japan. With this newly tangled IP landscape, questions are arising about the possible emergence of a patent thicket. On the other hand, early signs suggest that the iPS cell marketplace may evolve to provide ample room for many different contenders. Whereas for now companies are focused primarily on iPS cell cultivation as a means for deriving clinically relevant mature cells, companies may take advantage of recent data on transdifferentiation that suggest that this pluripotent midpoint may even be dispensable in the future2.

(Table 1, see Supplementary Table for more information) and several other applications filed by Sakurada were acquired by iZumi Bio, a biotech startup which took on Sakurada as its CSO. (After a short stint as CSO, Sakurada left and now is at Sony Computer Sciences Laboratory in Tokyo.) This past July, iZumi merged with Boston-based Pierian to form iPierian, located in San Francisco, thereby gaining the scientific acumen of Pierian’s team of stem cell experts— including George Daley, Lee Rubin and Douglas Melton of Harvard University.

Who’s on first? Yamanaka’s 2006 Cell article1 was undeniably a landmark achievement. By using lentiviral or retroviral vectors to deliver known pluripotency genes Oct3/4, Sox2, c-Myc and Klf4 into mouse fibroblasts, the Yamanaka team was able to coax these fibroblasts into a pluripotent state, similar to that of embryonic stem (ES) cells minus the legal and ethical baggage1. Thanks to a Japan Patent Office (Tokyo) ‘fast track’ process, JP2008-131577 was issued on 12 September 2008, barely two years after publication of the original article. Outside of Japan, however, Yamanaka’s applications face stiff competition. “In this case, the patent filings do not correlate necessarily with discovery timelines,” says Stephen Chang, chief scientific officer (CSO) of San Diegobased Stemgent, “and there are earlier applications essentially predicting iPS cells before the Yamanaka publication.” Indeed, in January, the UK Intellectual Property Office, located in South Wales, issued its first iPS cell patent based on work from another Japanese scientist, former Yamanaka colleague Kazuhiro Sakurada. This patent

As of January, iPierian is now also the beneficiary of the potentially far-reaching claims of the Sakurada iPS cell patent, which covers reprogramming of neonatally derived cells via a combination of Oct3/4, Sox2 and Klf4—but not c-Myc, which represents a potential risk as an oncogene. “From our standpoint, it is the first patent outside of Japan that clearly covers the generation of human iPS cells from human postnatal somatic cells,” says iPierian CEO John Walker. “And it relates very specifically to ‘any combination of forced expression of genes’— that means whether done by viral vectors, small molecules, plasmids or proteins.” The Sakurada patent is still under review by the European Patent Office (EPO), headquartered in Munich, but filing in parallel with UK Intellectual Property Office may have given iPierian a valuable head-start in Europe in establishing their priority bona fides. “Many times it’s easier to get a patent out of the UK Patent Office,” says David Resnick, a patent attorney and partner at Nixon Peabody in Boston. “The examination is quicker, and many times you end up with broader claims than you would by going through the EPO.”

544

USPTO, Alexandria, Virginia.

© 2010 Nature America, Inc. All rights reserved.

As issued patents on induced pluripotent stem (iPS) cells stack up, the specter of a patent thicket looms. Michael Eisenstein investigates.

USPTO may hold the key to the future of iPS cell research unless investigators can find ways to work around patents.

Things could get complicated once the EPO begins to issue patents in this sector, however. “We have this dual patenting system—it could be granted in the UK but not in the EU [European Union], even though the UK is part of the EU,” explains Chris Mason, professor of regenerative medicine bioprocessing at University College London. “We’ve seen this sort of thing with ESC [embryonic stem cell] patents here.” Accordingly, the iPierian patent could end up with diminished impact on the continent, although the EU’s doctrine of subsidiarity, which enables member states to essentially ‘opt out’ of EPO-issued patents, should at least allow the company to protect its early dominance in the UK. The complexity doesn’t end here, however, as every issued European patent is immediately vulnerable to challenges through what is known as an ‘opposition proceeding’. “People can come in and say, I think those claims are too broad, there’s prior art or other issues, and attack the patent,” says Resnick, “and that will certainly happen as these commercial entities sort of line up.” With these challenges in mind, Walker indicates that iPierian’s primary focus is to stake out as much IP territory as possible. Establishing a ‘foundation’ in the US In the meantime, the United States Patent and Trademark Office (USPTO), headquartered in Alexandria, Virginia, has just issued its first iPS cell patent, drawing fresh attention to an application whose filing date precedes both Yamanaka’s and Sakurada’s by several years. Rudolf Jaenisch, a researcher at the Whitehead Institute in Cambridge, Massachusetts, filed his application for ‘Methods for reprogramming somatic cells’ in November 2003; this past February, he received a Notice of Allowance for his application (10/997,146), giving an important boost to the IP portfolio of San Diegobased Fate Therapeutics, a company for which Jaenisch is scientific cofounder. “We’ve exclusively licensed IP from the Whitehead Institute in connection with Dr. Jaenisch’s genetic-based reprogramming methods,” says Scott Wolchko, Fate’s chief financial officer. This early patent does not cover reprogramming methods in the same sense as the Yamanaka or Sakurada patents, but rather represents a means for identifying factors that can be used in iPS cell production co-expressing selectable markers linked to pluripotency genes. “It’s a tool for being able to screen for pluripotency factors, whether they’re chemicals, genes or whatever,” says Lisa Haile, a partner at DLA Piper’s San Diego office, who presently acts as outside counsel for Fate. As such, many still consider the landscape wide open for the issuance of so-called foundational patents. “We don’t antici-

volume 28 number 6 june 2010 nature biotechnology

news f e atu r e

Table 1 Some key early patent filings in iPSC generationa Patent

Jurisdiction

Owner

Core reprogramming factors involved

Yamanaka JP2008-131577

Japan

iPS Academia Japan

A combination of Oct3/4, Klf4 and c-Myc

Sakurada/Bayer GB2450603

UK

iPierian

Oct 3/4, Sox2 and Klf4, but not c-Myc; cultured in the presence of FGF-2

Jaenisch/Whitehead US2008/0280362 (10/997,146)

US

Whitehead Institute (exclusively licensed to Fate Therapeutics)

Pluripotency genes - including Oct4, Nanog and/or Sox2 – linked to selectable markers.

Yamanaka US2009/0068742

US

Kyoto University

US counterpart to issued Japanese patent (see above)

Sakurada/Bayer US2009/0191159

US

iPierian

US counterpart to issued British patent (see above)

Jaenisch/Whitehead PCT/US2008/004516

Multiple (EU, US and others)

Whitehead Institute (exclusively licensed to Fate Therapeutics)

One or more factors including Oct3/4, Sox2, Klf4, Lin28, Nanog or c-Myc

Thomson et al. US2008/0233610

US

Wisconsin Alumni Research Foundation

Some combination of Oct4, Sox2, Nanog and Lin28, but not including c-Myc or Klf4

Ding-Schultz US2007/0254884

US

The Scripps Research Institute

Covers method for identification of dedifferentiating chemical compounds

Mack & Thomson US2010/0003757

US

Stem Cell Products

Some combination of Sox2 and Oct4 with Nanog, Lin28, Klf4 or c-Myc

Issued

© 2010 Nature America, Inc. All rights reserved.

Filed

aMore

information is available in Supplementary Table 1.

pate ever needing the type of [screening] methods that were described in that issued patent,” says iPierian’s Walker. “I think the US is totally a blank slate as it relates to the creation of mouse or human iPS cells.” However, Wolchko points out that Fate is also awaiting a decision on a second Jaenisch patent that was filed in multiple jurisdictions, unlike the 2003 application, which was only filed in the US due to confidentiality issues. This application pertains more directly to conventional somatic cell reprogramming using one or more of the pluripotency genes. With a filing date of April 2007, it may also represent a strong early contender in both the EU and US. “This priority date is prior to the Sakurada application,” says Wolchko. “As such, the Sakurada application— or any patents issued thereunder—will not affect the prosecution, scope or patentability of the Jaenisch applications.” Nevertheless, discussions that focus on priority may overlook the bigger question of whether any truly foundational patent is likely to emerge. Many cite the ES cell patents issued to the Wisconsin Alumni Research Foundation (WARF), based on the groundbreaking work of University of Wisconsin at Madison researcher James Thomson. Three primary WARF patents for isolation and maintenance of mammalian ES cells raised a storm of controversy, owing to their breadth and what some considered to be onerous licensing requirements, and were subject to a court challenge and reexamination proceedings3. The results have been mixed—two were upheld, albeit with a narrowing of claims, one overturned—but the experience left a bad taste in the mouth of many in this field. “Several

years ago, a lot of people, ourselves included, were unhappy with WARF’s direction, and one of the things that the industry wants to do is make sure we don’t end up in quite that situation again,” says Brock Reeve, executive director of the Harvard Stem Cell Institute in Cambridge, Massachusetts. There is cause to believe that the USPTO may itself pursue a far more cautious course based on these prior lessons. “I think it’s difficult to get the patent office to issue these broad claims,” says Resnick. “They are very aware of the criticism they would get if they issue a broad patent that would preempt the whole field and create a problem with people doing any iPS cell research.” Accordingly, the broad scope of early patents may quickly be pared back considerably as the field continues to grow. “I don’t know that filing today will get you broad scope of claims,” says Haile. “If you’re using genetic manipulation, for example, I don’t believe that you’re necessarily going to get claims that include small molecules or proteins as well.” This narrowing could be further accelerated by the recent ruling handed down by the US Court of Appeals for the Federal Circuit for the high-profile Ariad v. Lilly patent lawsuit. The case revolves around the scope of the written description component of a patent application— historically interpreted as a requirement for a detailed summary that incorporates proof that an inventor has both successfully created and secured possession of a technology. The courts have required this description separate from the enablement requirement, which provides disclosure of the means for teaching manufacture and use of the technology to a skilled practitioner.

nature biotechnology volume 28 number 6 june 2010

The Federal Circuit has now formally upheld this strict interpretation by a 9–2 majority. Their decision makes it clear that this written description must provide actual proof of invention that goes beyond any mere hypothetical conception, stating that “patents are not awarded for academic theories, no matter how groundbreaking or necessary to the later patentable inventions of others,” and seems likely to greatly increase the burden of proof for applicants attempting to secure patents with far-reaching claims. A muddle of methods Against this backdrop, the patent landscape may ultimately be populated by numerous narrowly defined methodological patents rather than broad issuances with sweeping claims, although this will likely be steered as much by rapid technological evolution as cautious patent office policy. In fact, by the time Yamanaka’s initial patent issued, work had already been published on nonretroviral methods using fewer reprogramming factors and delivering them in a manner less disruptive to genomic integrity— and in some cases, didn’t use genes at all. “We’ve already seen protein- and biologic-based cocktails for reprogramming, and identified small molecules that can be added to improve both time to reprogram as well as reprogramming efficiency,” says Wolchko, “and we’re probably not too far off from seeing a pure small-molecule–based method for reprogramming cells.” The capacity to produce iPS cells with essentially ‘pristine’ genomes may make these cells considerably more palatable for clinical applications and Fate is banking heavily on such strategies, which have been championed by 545

© 2010 Nature America, Inc. All rights reserved.

N E W S f e atu r e another of their scientific cofounders, Sheng Ding of The Scripps Research Institute in San Diego. Ding’s team has achieved vector-free reprogramming using cell-penetrating variants of the four Yamanaka factors4 as well as progress in the use of chemical adjuncts that boost efficiency of reprogramming5, still an obstacle to commercial applications using reprogrammed cells. Ding first staked out this territory as early as 2004, in a patent application with colleague Peter Schultz that targets identification of “small molecules which induce dedifferentiation of mammalian somatic cells.” Several alternative nonintegrating methods are also in use, including a filing from Thomson for the cultivation of iPS cells that are “essentially free of exogenous vector elements,” based on the use of an episomal reprogramming system6. This technology is currently in use at Madison, Wisconsin–based Cellular Dynamics International (CDI), a company which was cofounded by Thomson and maintains close ties to the University of Wisconsin. “WARF is an investor in our company,” says Nick Seay, chief technology officer at CDI. “We have three licenses from WARF so far, and we consider them as an active partner.” These are just a handful of the applications now under consideration—as of this writing, 175 applications mentioning ‘induced pluripotent stem cells’ have been published by the World Intellectual Property Organization—and there is ample room for a dark horse to stake an unexpectedly important claim. Most watching the field consider it simply too early to speculate about who may ultimately have the strongest hand. “This is no different from other new platform technologies,” says Mason. “I’m sure there will be hundreds or thousands of patents and then eventually, just like monoclonal antibodies, it will get whittled down to a few key patents.” Just part of the process As long as the situation remains nebulous, the primary concern for companies is ensuring that research and product development can continue by whatever means are most practical and costeffective. “From our perspective, we are not interested in stifling research or anybody else’s innovation,” says Wolchko. “What we’re concerned about is ensuring that we have freedom to operate in the iPSC space.” According to Seay, the success of CDI is as dependent on strong relationships with outside inventors and researchers as it is on internally developed IP. “For anybody who’s got a business interest, it’s not an academic question of who owns which rights,” he says. “It’s a question of what specific technologies you need licensed for the specific product you want to make.” Some observers are cautiously optimistic that as the IP 546

landscape becomes more concrete, patent holders will see the wisdom of entering into crosslicensing arrangements that ensure broad access under reasonable terms. As an example, Haile cites the University of Massachusetts Medical School’s approach to RNA interference technology. “Part of the conditions for licensing to a company were that they make nonexclusive licenses available,” she says. “The whole idea was to get the technology developed and out to the public and to consumers, but not necessarily by one company.” Chang agrees, but remains skeptical. “From a business perspective, the smart way to do it would be to get lawyers out of the primary discussions and don’t use ego as your primary driver,” he says. On the other hand, this may be mitigated by the perception of present iPS cell work as a largely precompetitive step on the road to product development. “What I see a lot of lately is people focusing research not necessarily on the core technologies, but on using technologies to generate different cell and tissue types,” says Haile. For example, work at iPierian is focused on cells that can facilitate development of drugs that target neurodegenerative diseases, whereas Fate is using iPS cells to identify compounds that could modulate the behavior of stem cells in vivo. CDI, on the other hand, is actively producing iPS cell–derived human cardiomyocytes for use in toxicology and drug discovery studies. For all of these companies, the true monetary value of these cells lies in the development of differentiation and screening strategies, and the therapeutic compounds that result. Even in the competitive arena of stem cell differentiation, companies are not scrambling for control over methodological IP. “There will obviously be preferred mechanisms to differentiate cells,” says Wolchko, “but I do think there will be more than one way to take cells from point A to point B, and I think this will provide multiple companies with freedom to operate in this space.” And given the time gap between invention and patent issuance—which can span the better part of a decade—researchers are keen to benefit from early revenue and greater dissemination through prompt licensing, offering additional options for companies. “Virtually every cell type has many ways to make it, and many patents on different ways to make it—it’s just a question of which ones work best in your hands and are priced so that you can get access. It’s very easy to price yourself out of this market if you own one of those [patents] and you charge too much,” says Seay. Indeed, recent work in ‘transdifferentiation’ from scientists such as Harvard’s Douglas Melton, whose team directly reprogrammed mature pancreatic exocrine cells to yield betaislet cells7, offers evidence that stem cells could potentially become largely dispensable for many

research applications. “Instead of going back from a fibroblast to an iPS cell and then on to a liver or heart cell, you could go from skin to cardiac in one move,” says Mason. Kumbaya? For the time being, the iPS cell environment remains more collaborative than directly competitive, with current patent-holders expressing a commitment to promote development of the field in parallel with their commercial efforts. “We think this represents an opportunity for us to really be helpful by more of a Cohen-Boyer type of approach to licensing,” says Walker, referencing the famous Stanford University cloning patent, which generated hundreds of billions of dollars of licensing revenue while also enabling commercialization by a broad array of inventors8. Similarly, Yamanaka’s Japanese patent is now managed by iPS Academia Japan, a company he launched with the University of Kyoto in June of 2008, which offers nonexclusive licenses that are royalty-free for nonprofit research entities. Likewise, academic iPS cell researchers are maintaining close ties with the commercial sector, whether through participation in scientific advisory boards or through broader institutional collaborations. This past October, for example, iPierian announced a partnership with Johns Hopkins University on a $3.7 million National Institutes of Health Grand Opportunities Grant for a project using iPS cell–derived motor neurons and astrocytes to study ALS. This relatively collegial environment may seem surprising for such a ‘hot’ and potentially powerful technology, but Mason suggests the typically slow path from patent to product may help maintain the peace for some time to come. “Platform technologies, right across biotech, take 20 to 25 years to get from basic discovery of the technology into mainstream clinical products,” says Mason. “Typically companies are too smart to be fighting over patents early on, when there’s no benefit to be had—and I think this is just the very beginning.” Michael Eisenstein, Philadelphia

Note: Supplementary information is available on the Nature Biotechnology website. 1. Takahashi, K. & Yamanaka, S. Cell 126, 663–676 (2006). 2. Vierbuchen, T. et al. Nature 463, 1035–1041 (2010). 3. Vrtovec, K.T. & Scott, C.T. Nat. Biotechnol. 26, 393–395 (2008). 4. Zhou, H. et al. Cell Stem Cell 4, 381–384 (2009). 5. Lin, T. et al. Nat. Methods 6, 805–808 (2009). 6. Yu, J. et al. Science 324, 797–801 (2009). 7. Zhou, Q. et al. Nature 455, 627–632 (2008). 8. Feldman, M.P., Colaianni, A. & Liu, C.K. in Intellectual Property Management in Health and Agricultural Innovation: a Handbook of Best Practices (eds. Krattiger, A. et al.) 1797–1807 (Concept Foundation, Bangkok, 2007)

volume 28 number 6 june 2010 nature biotechnology

building a business

Beyond venture capital John Hollway

© 2010 Nature America, Inc. All rights reserved.

You don’t always have to go to venture capitalists to raise funds. Proper planning and research can help you bring in millions through other avenues.

O

ne of the fundamental challenges in running a biotech business is the temporal alignment of two initiatives—scientific advancement and fundraising—that have no natural affinity for one another. Sometimes companies are lucky enough to raise money on the back of a scientific accomplishment, which is when it’s easiest, but raising money is a constant hurdle, especially for young biotechs; there is no guarantee that the next scientific accomplishment will occur within your new financing window (or at all). At Achaogen, we’ve secured commitments for more than $100 million in alternative financing to complement the investments made by our venture capital supporters. This has not always been easy, and it has rarely been fast, but we’ve learned a host of lessons through our experience. Money tension First, some background. To date, venture capitalists (VCs) have provided a valuable supply of risk capital to the marketplace to fund high-risk, high-reward enterprises like R&Dstage life science companies. But a potential tension exists here, as venture portfolios and companies tend to approach risk differently. VCs invest in multiple companies, technologies and therapeutic areas, and thus they can distribute their risks by putting smaller amounts of capital in play among a variety of companies. Life science companies, on the other hand, typically place a more focused bet on a single technology or therapeutic area and may seek to mitigate their risks by diversifying their funding sources or assets or conducting broader experiments to prove scientific hypotheses. John Hollway is vice president of business development at Achaogen, South San Francisco, California, USA. e-mail: [email protected]

Box 1 The benefits of venture capital funding The fact that venture capital may be less easy to come by than non-dilutive financing (NDF), or that it comes at a cost of management equity in the business, certainly doesn’t mean you should eschew it. Indeed, venture capitalists (VCs) bring many potential advantages to a business that NDF providers cannot bring, including access to other investors (like large pharma partners), experience in managing companies that have faced similar challenges, access to networks of new hires, validation of the management team or the underlying science, flexibility in terms and far greater speed in consummating the investment than the typical alternative funding cycle. And certainly, money from VCs, which simply goes into a money-management account and is far more liquid than the project-based cost/reimbursement structure of most government contracts, can more rapidly be repurposed if your scientific direction changes. The bottom line is that early investing needs to be strategic. Whether you’re looking for NDF or investment from the VC community, one size definitely does not fit all. There are many options and strategies that may be valid, and knowing where the opportunities lie and deciding which to pursue and when can be the difference between having a great scientific idea and having a successful life science company.

The problem is that many risk-diversifying moves for companies can both increase the overall enterprise value of the company and reduce the short-term price of the company’s shares by either raising enterprise costs or delaying advancement of a program to a value inflection point. This may create tension with the company’s existing VCs, who do not like to see shares devalued. Another potential conflict may occur when the company requires subsequent rounds of financing on the road toward an exit—a situation in which the privately held company stock can be sold and/or made publicly liquid. VCs often seek to reduce their risk by investing just enough funding to permit business operations to continue until the next scientific milestone is reached. If the milestone is reached, it should increase the value and decrease the risk of the enterprise, making it easier to raise the next round of funding. If, as not uncommonly happens, management’s projections of the time and money needed to reach this milestone are inaccurate (often, though not always, due to

nature biotechnology volume 28 number 6 JUNE 2010

uncontrollable external factors), the company may need additional funding before significant enterprise value has been created. Thus, life science companies and VCs are often on opposite sides of the financing table, with companies looking to raise bigger chunks of capital less frequently, which allows them to focus on science instead of fundraising while still reaching multiple milestones. The described tensions are only heightened when the financing environment is tight. A host of issues are hurting small biotechs these days: investors’ appetites for higher-risk investments have decreased; the initial public offering (IPO) market has not been terribly receptive to life science companies; big pharma is focused on programs that have achieved clinical proof of concept and can generate significant revenue in 2013 (or sooner); and VCs are more limited in their access to capital and are seeking returns in timelines that make investments in basic research unattractive. In addition to all this, the costs required for R&D are the same as they were before investments became scarce.

547

b u i l din g a b u s in e s s

2004

September - $15M Series A

2005 2006

May - $2.1M grant from Defense Advanced Research Program Administration for new approaches to treat Bacillus anthracis (anthrax) October - $26M Series B October - $25M contract with Defense Threat Reduction Agency (DTRA) for SOS Pathway and new fluoroquinolone research

2007

June - $30–$34M contract with DTRA for LpxC inhibitors October - $2M contract with US Army Military Research Institute and Material Command (USAMRMC) for new therapies to treat Acinetobacter baumannii

2008

September - $7M contract with Wellcome Trust for advancement of new amnioglycosides October - $27M contract with National Institute of Allergy and Infectious Disease for advancement of new aminogylcosides

© 2010 Nature America, Inc. All rights reserved.

October - $2M contract with USAMRMC for new therapies to treat A. baumannii 2010

April - $56M Series C

Figure 1 Achaogen’s funding timeline

Ways around Venture capital funding remains a strong option for financing your venture (Box 1), but there are other ways to help bring in money. Our company, Achaogen, which is focused on small-molecule antibacterial therapies to treat multidrug-resistant infections, has had success raising money in various ways. To date, we have raised about $100 million in venture capital and have augmented that with over $100 million in contractual commitments for funding from alternative sources, such as the US Department of Defense in Washington, DC; the National Institutes of Health in Bethesda, Maryland; the Wellcome Trust in London and other organizations (Fig. 1). The company decided shortly after its founding to blend its venture capital funding with non-dilutive financing (NDF)—money from third parties that could be obtained without giving up stock. We felt we had programs the government would fund, and any time we could gain capital without relinquishing stock, we wanted to do it. Done properly, NDF provides an external validation of the market need for your science and of the scientific rigor of your company’s particular approach. It also serves to extend the company’s financial runway and provides valuable capital for additional experiments. Many companies are turning to alternative sources, such as government grants or contracts, venture philanthropy organizations or hybrids of these options. These alternatives can be highly effective ways to leverage a business, but as with any financing vehicle, they present challenges and complexities over the

548

long term that should be understood before any commitments are made. Here are nine lessons our team learned through our funding experiences. 1. ‘Non-dilutive’ and ‘paid for’ may not be the same thing. For any small company, retaining focus is crucial. Although success in any one program is unlikely, adding a second or third program in an unrelated area may actually increase the likelihood of failure in both programs due to additional costs and distractions. Alternative funding may address the cost part, but it is important to ensure that extra programs do not create distractions that undermine the company’s ability to function. One example would be the funded application of a platform technology in a therapeutic area that is not a focus of the company’s other development efforts (for example, if you are working in infectious disease but the National Institutes of Health wants to fund a program in inflammation research that is not commercially viable for you). It may be that the addition of a second program in a new therapeutic area is a boon to your company, providing useful validation and diversification of your portfolio in a cashneutral way, fully funded by your partner as opposed to your stockholders. Still, there are risks for a business that goes along with adding a disparate research program, including distractions and problems with resource allocation. And, because you’ve entered a contract to pursue the research, you may find in the future that if you want to streamline your operations and divest the new program, the right to ter-

minate may only go one way and it might not be yours. So consider things carefully before taking on a new program, even if it’s coupled with sizable funding. 2. Grant financing takes time. In general, our experience is that it takes 18 months from the day you start seeking government funding to the day any money is received. For larger government contracts (greater than $3 million), the process is started by a Broad Agency Announcement—essentially, a request for proposals for certain government initiatives. The timeline for submitting proposals is generally 4–6 months, with another 4–6 months (or more) provided for the government to review the proposals and decide on the ones that will receive tentative awards. If yours is one of the lucky proposals to receive a tentative award, congratulations! But you are only halfway there. You then have to negotiate the contract, which can take another 4–6 months or more. Only after all of these things have happened is the contractor authorized to initiate work on the program. The challenge is the same for Small Business Innovation Research/Small Business Technology Transfer (SBIR/STTR) submissions, which are more frequent but are for smaller monetary amounts. In that arena, proposals typically receive lower priority scores on their first submission but can be resubmitted, based on the reviewing panel’s comments, during the next available window for applications. 3. Proposals are themselves expensive. Proposals also take time and money to put together and require detailed budgets, specific work plans, quotes from subcontractors and the provision of abundant data, typically in a nonconfidential setting. Each proposal Achaogen has made has taken about one full-time employee (half scientific writer, half business development writer and document coordinator) 2–3 months to complete. Although giving the program that much thought has benefits, it also takes away from other tasks the employee could be doing. (However, submitting proposals does get somewhat easier after the first one, as there is quite a bit of boilerplate language desired by various government agencies.) Also, in Achaogen’s experience, the time needed to write a good grant is largely the same, regardless of the amount of money requested—so choose your proposal opportunities accordingly. 4. Always measure twice. The government, quite reasonably, requires that its contractors submit a detailed budget, complete with quotes from outsourced labor, that follows a precise

volume 28 number 6 JUNE 2010 nature biotechnology

b u i l din g a b u s in e s s

© 2010 Nature America, Inc. All rights reserved.

work plan for the length of the award. The budget for this work plan will be determined in the contract negotiation phase. Typically, the agencies will not hold companies to the strict amounts laid out in the budget for each type of experiment—but they will hold the line on the total budget amount, so measure carefully. Also, be careful of the Statement of Work portion of the proposal. Word it too vaguely and it won’t be approved; word it too precisely and minor changes to the research plan over time may require repeated amendments to the contract. These will take a long time to get through the bureaucracy and can also lead to reexamination of the total budget. 5. Beware the SBIR/STTR conundrum. As of this writing, companies that have accepted money from VCs are typically ineligible for funding from SBIR/STTR sources. This is due to an overly simplistic (and probably inaccurate) interpretation of a federal rule that requires SBIR/STTR recipients to be majority-owned by US citizens. This interpretation extends the rule to limited partners of VC partnerships. The US House of Representatives recently agreed to remove this limitation, but it remains to be seen if that action will become law (Nat. Biotechnol. 27, 1065–1066, 2009). Also, remember that these grants are typically of small size ($200,000–$1.5 million), and multiple awards would be necessary to fund a credible drug research program from R&D through an investigational new drug submission. Even with the frequency of these awards, each potential grant application may be rejected, which could cause delays in your work while you scramble to find additional funding, and the amounts are such that you may run out of cash before even being able to apply for the next round. 6. Alternative funding application outcomes are binary. One of the main challenges of NDF is that although it’s alluring, it’s certainly not guaranteed, and funding tends to be all or nothing on a project-by-project basis. It is possible to spend months putting together a proposal and months more eagerly waiting— only to get a negative outcome that renders the entire effort useless. The government’s priorities are specific but not static, and its pockets are deep but not limitless. Also, you will not be the only company applying to receive a specific pool of money, and you may not even have a completely unique technical approach. More than 200 companies applied for the first contract that Achaogen was awarded, and only 13 companies received funding. The government will provide valuable feedback if your proposal is

not accepted, so that you can begin this risky, binary process again. Make sure to set expectations with your board appropriately—every proposal has a less than 50% shot at success, and success can be incremental over several revised submissions. 7. Priorities for VCs and NDF providers differ. Typically, entities that offer alternative funding have an agenda that is noncommercial, or what is sometimes called ‘super-commercial’—it has a higher purpose than simply selling drugs. One needs to be careful that terms set during NDF do not conflict with your ability to raise funds from traditional VCs. The Institute for OneWorld Health in San Francisco, for example, focuses on medicines for the developing world, an arena in which the economics of the pharmaceutical industry have historically been challenged. This may lead to some difficult discussions and some challenging terms being placed in a funding agreement that could scare off the more commercial investors you might want to attract. (Most organizations try hard to blend a commercial return with a charitable purpose in a way that can be appealing to management, but this can be harder than it sounds. Be careful about ‘hooks’ in agreements like diligence requirements, ownership in developing markets or approvals over potential acquisition partners. These could scare off an acquirer or licensor down the road.) 8. A certain infrastructure is needed to support contracts. Receiving funding from the federal government can create a significant administrative burden for you. Government contractors are frequently audited, allowing the government to feel secure that business is being conducted in a way that it finds suitable. This could mean companies have to add infrastructure for time-based activity reporting, equal opportunity employment restrictions and reporting, time-limited intellectual property (IP) reporting or even the submission of quarterly (and in some instances, monthly) technical reports. The penalties for failing to comply with this dizzying array of regulation can be severe, and the government keeps close track of contract performance, using that as a significant factor in future awards. It is important to understand this clearly before signing a government contract, so weigh the risks before applying. (Grants, which are often for smaller dollar

amounts, also typically have far less burdensome compliance and communication requirements.) 9. There are implications for IP. Government contracts are relatively benign in this regard, with standard language in the Federal Acquisition Regulations allowing companies to retain ownership of patents while providing a license to the US government to use technologies invented under government-funded programs for legitimate federal purposes. Given the precision needed to manufacture most pharmaceuticals, the contracted company would remain the most likely customer of the government, but it is nonetheless important to realize what such government purposes may be, both now and in the future. Other funding organizations (the Wellcome Trust, the Cystic Fibrosis Foundation in Bethesda, the Gates Foundation in Seattle and so on) may place different requirements on your IP as a condition of funding, including the ability to obtain the program if certain diligence obligations are not met, sell a resulting product in markets that your company is not actively pursuing and veto a merger with a partner who does not share the same philosophies of the alternative funding group. These IP hooks can, in some cases, be sufficiently onerous to make the investment too risky for the company. Most of these enterprises, however, want you to develop the drug or else they wouldn’t have agreed to fund it in the first place. Thus, there is typically some room to find a mutually acceptable middle ground. Conclusions VCs are not the only means of funding your company. Achaogen’s strategy of combining NDF with venture capital has succeeded in taking the best from both worlds—the capital efficiency of the NDF and the broad, operational utility and excellent networks and managerial support of the VCs—to construct a company with a robust discovery and development pipeline, and we’ve done this without diluting our investors into oblivion. Of course, one needs investors who understand the risks of pursuing NDF aggressively, and it helps to have expertise and experience working with government officials and obtaining funding. Alternative funding organizations are out there, and they can be as valuable to you as they have been to Achaogen.

To discuss the contents of this article, join the Bioentrepreneur forum on Nature Network:

http://network.nature.com/groups/bioentrepreneur/forum/topics

nature biotechnology volume 28 number 6 JUNE 2010

549

correspondence

To the Editor: We wish to express our concern and dismay at the apparent lack of intergovernmental engagement by European governments regarding the proven positive roles of modern biotechnologies as key tools supporting efforts to address the issue of food security, especially in developing countries. This was shown clearly by the failure of 26 of the 27 members states of the European Union to send any official government delegations to participate and engage in the recent United Nations Food and Agriculture Organization (FAO; Rome) intergovernmental conference (ABDC-10) on ‘Agricultural biotechnologies in developing countries’ (http://www.fao. org/biotech/abdc/en/). The Netherlands was the only EU member state to send an official government delegate to ABDC-10. The conference1, which took place in Guadalajara, Mexico, on March 1–4, 2010, was concerned with the full range of agricultural biotechnologies used in food and agriculture, including the improvement of plant varieties and animal populations to increase their yields; characterization and conservation of genetic resources; plant or animal disease diagnosis; vaccine development; improvement of feeds; and the safety of foods. The meeting also crossed different sectors, covering crops, livestock, forestry, agro-industry, and fisheries and aquaculture. Around 300 policy makers (government representatives), scientists and representatives of intergovernmental and international nongovernmental organizations came together at the meeting from 68 different countries. The conference was co-sponsored by the International Fund for Agricultural Development and also involved the Consultative Group on International Agricultural Research, the Global Forum on Agricultural Research, the International Centre for Genetic Engineering and Biotechnology and the World Bank. Previous FAO International Technical Conferences on related topics, such as genetic resources

for food and agriculture, have been fully that the US government regards agricultural attended by EU member states and have biotechnologies as a key area in which its own led to agreement on International Plans of public and private sector R&D can be usefully Action. deployed to assist in the challenge of food Over the past few years, there has been security in developing countries. a great deal Many of rhetoric ABDC-10 from EU delegates governments expressed and national puzzlement at European the stark lack organizations of attendance about the from importance official EU of global food government security and representatives the need for a and the multi-pronged negative approach from message that both developed this conveyed nd and developing ABDC-2010 is the 2 FAO conference to focus on the about the potential of agbiotech in developing countries. countries. The willingness desirability of of European a “multifaceted and linked global strategy… countries to facilitate the exploitation of to ensure sustainable and equitable food European agricultural biotechnology research security” was highlighted recently by an for the strengthening of food security in eminent group of European experts that developing countries. This was even more included the UK government chief scientist, ironic given the fact that the International John Beddington2. It was therefore surprising Steering Committee for the FAO conference had significant representation of technical that only one European government thought and policy expertise from Europe (including it worthwhile to take advantage of the signatories of this letter). unique opportunity presented by the FAO For genetic modification (GM)-phobic conference to engage with several hundred European policymakers, it should be policy specialists and agricultural experts from over 50 developing countries in a forum emphasized that one clear message from the specifically targeted at developing approaches FAO conference was that modern agricultural biotechnologies are about much more than and alliances to increase global food security. genetic engineering. Indeed, although genetic In contrast to the no-shows from EU modification technologies are constantly member states, which collectively aspire being improved and are making important under the Lisbon Agenda to become a contributions to crop breeding, they are only knowledge-based bio-economy region, one component of the overall agricultural the United States sent a high-level official biotechnology toolkit required for science government delegation of >20 officials, and technology to strengthen food security in scientists and policymakers led by Roger developing countries. Other biotechnologies Beachy, director of the US Department of that have already contributed greatly to Agriculture National Institute of Food and developing country crop, forestry, fisheries Agriculture and a senior member of US and livestock improvement include advanced President Obama’s science team. It was clear

nature biotechnology volume 28 number 6 JUNE 2010

FAO

© 2010 Nature America, Inc. All rights reserved.

1 out of 27—European politicians score poorly in agbiotech

551

© 2010 Nature America, Inc. All rights reserved.

corr e s p ond e nc e tissue culture, artificial insemination and reproductive technologies, mutagenesis/ TILLING, marker-assisted selection and micropropagation. Like GM, all of these biotechnologies have also benefited from new advances in research over the past few decades. In addition to their vital contributions to breeding, biotechnologies are also playing key roles in improving the cultivation and management of crops, forestry, fisheries and livestock. For example, crop management is benefiting from new biotech-based strategies for pest and disease control (including diagnostics), as well as the increasing use of biofertilizers as an alternative to expensive nonrenewable chemical inputs. As stated in the conference report1, there was strong consensus at ABDC-10 that future progress for global food security will require the deployment of the whole range of both new and traditional biotechnologies, in combination with other less high-tech methods in the context of a more needsdriven rather than technology-led approach. Organizations representative of end-users, especially smallholders, should, where possible, participate in the process of crop, forestry, fisheries and livestock improvement. In the context of climate change and other environmental uncertainties that are likely to increase both biotic and abiotic stresses, there may be many cases where broad adaptability and yield robustness, rather than high yields per se, should be the primary focus of crop and animal improvement. It was also agreed that access to agricultural biotechnologies should be improved, for example via North-South collaborations and privatepublic partnerships. Finally, the sometimes inconsistent and onerous regulatory burdens that policy makers have devised regarding some biotechnologies (that is, GM) were felt to be a major impediment to any possibilities for their dissemination and exploitation by developing countries for the benefit of poorer smallholder farmers. In one of the final sessions of the FAO conference, a European participant confessed to being “ashamed” at the lack of participation by European governments. He was not alone. This was an opportunity missed by EU member states and has certainly raised questions in some developing countries regarding the willingness of EU member states to close the widening biotech gap between rich and poor countries, in a manner that could reduce poverty levels and strengthen food security in developing countries. Rather than focusing on inward-looking debates on issues such as the intricacies of GM crop regulation, 552

European governments and policy makers should realize that there is a broad range of agricultural biotechnologies (including, but by no means restricted to, GM) that can make a huge contribution to assisting humanity tackle the immense task of feeding itself sustainably in an era dominated by the uncertainties of population growth, climate change and rapidly escalating global demand for food, feed and energy. As our US colleagues might say, European governments and their policy makers should “wake up and smell the coffee.” COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Atanas Atanassov1, Godelieve Gheysen2, Denis J Murphy3, Olivier Sanvido4, Joachim Schiemann5, Charles Spillane6 & Roberto Tuberosa7

1AgroBio Institute, Sofia, Bulgaria. 2Department of Molecular Biotechnology, Ghent University, Ghent, Belgium. 3Division of Biological Sciences, University of Glamorgan, Pontypridd, UK. 4Agroscope Reckenholz Tänikon Research Station ART, Zürich, Switzerland. 5Institute for the Biosafety of Genetically Modified Plants, Julius Kühn Institute, Federal Research Centre for Cultivated Plants, Quedlinburg, Germany. 6Botany and Plant Science, National University of Ireland, Galway, Ireland. 7Department of Agronomy, University of Bologna, Bologna, Italy. e-mail: [email protected]

1. Food and Agriculture Organization of the United Nations (FAO). Agricultural biotechnologies in developing countries: options and opportunities in crops, forestry, livestock, fisheries and agro-industry to face the challenges of food insecurity and climate change (Abdc-10), Guadalajara, Mexico, 1–4 March 2010 (FAO, Rome, Italy, 2010). 2. Godfray, H.C. et al Science 327, 812–818 (2010).

Split approvals and hot potatoes To the Editor: The letter by Gerhart Ryffel in the April issue1 outlines some of the public perception concerns surrounding the European Union’s (EU; Brussels) recent sanctioning of the cultivation of a genetically modified (GM) potato—the first for any GM plant in 12 years. But readers should be far more concerned about the form of approval granted by EU authorities. Registration of BASF’s (Ludwigshafen, Germany) Amflora was only for commercial production of starch for industrial purposes, not for food use. This ‘split approval’ is a disaster waiting to happen. Amflora was created because of a limitation of conventional potato varieties. Such potatoes contain starch granules made up of two glucose polymers: amylopectin, a highly branched molecule, and amylose, which has a linear arrangement. Although the alignment of the linear amylose chains in potatoes may be useful in food preparation (e.g., for setting sauces on cooling) and contributes to the consistency of potatoes as a foodstuff, it is undesirable and must be removed in many industrial applications, such as making the coating on glossy printing paper. The availability of Amflora

means that potatoes with low-amylose starch appropriate for industrial uses will now be grown in Europe and offer economic benefits to local industry and farmers. All well and good. But the decision of EU regulators to provide a split approval, which permits animal feed or industrial uses but not human consumption, is likely to invite all sorts of mischief. One only need look no further than the debacle surrounding a similar decision by the US Environmental Protection Agency (EPA) over a decade ago on a recombinant DNA-modified corn variety called StarLink that contains a bacterial protein, Cry9C, toxic to certain lepidopteran insects2. Because of unresolved dubious concerns about possible allergenicity of the novel StarLink protein—which, similar to many known allergens, takes slightly longer than most proteins to be digested in a laboratory simulation of digestion—the EPA approved the variety only for animal but not human consumption. Following StarLink’s commercialization, an activist organization paid a laboratory to test a large selection of packaged food products made with corn (including corn chips, tortillas and taco shells) and found the

volume 28 number 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

corr e s p ond e nc e unintended presence of the Cry9C protein in some of them. After newspaper and television news reports announced that the unapproved protein—which EPA regulated as a pesticide—was found in food products taken from grocery store shelves, 28 people reported that they had experienced allergiclike reactions after eating food products that contained corn. However, an intensive investigation of adverse effects reports by the US Centers for Disease Control was not able to confirm a single allergic reaction: “Although the study participants may have experienced allergic reactions, based upon the results of this study alone, we cannot confirm that a reported illness was a foodassociated allergic reaction.” Despite this conclusion and the absence of other evidence of harm of any kind to anyone, because there was no regulatory approval for StarLink in human food, a class-action lawsuit alleging that consumers ate food unfit for human consumption was successfully concluded with a settlement against Aventis (Lyon, France), producer of the StarLink corn variety. The EPA has since decided that it will never again approve a recombinant DNAmodified crop for split use. Any crop intended for feed or industrial uses that could conceivably find its way into the food supply will have to meet the standards

for human food use to gain government approval. The StarLink saga should provide a cautionary tale to BASF, the creator of the Amflora potato: recombinant DNA-modified crops not approved for human consumption present the risk of legal liability, even if no consumer has suffered any toxic, allergic or other healthrelated harm. It should also concern EU regulators but likely will not, given their discriminatory stance against recombinant DNA technology applied to agriculture. The StarLink contretemps resulted from a fault not with the product itself or the legal system but from flawed regulatory policy and an unwise series of decisions by regulators. Such problems are the inevitable result of a regulatory approach that treats recombinant DNA–modified products as though they pose some inherent, unique risks, although all the evidence is to the contrary. COMPETING FINANCIAL INTERESTS The author declares no competing financial interests.

Henry I Miller The Hoover Institution, Stanford University, Stanford, California, USA. e-mail: [email protected] 1. Ryffel, G.U. Nat. Biotechnol. 28, 318 (2010). 2. Fox, J.L. Nat. Biotechnol. 19, 298–298 (2001).

Why drought tolerance is not the new Bt To the Editor: Given rapid uptake of Bacillus thuringiensis toxin (Bt) cotton by farmers in several developing countries, it is often assumed that poor farmers will clamor for droughttolerant varieties in an era of tightening water resources and climate change. There are, however, important differences between Bt-mediated insect resistance and drought tolerance. We would like to bring to the attention of your readers some of these differences, which, based on the output of a stochastic model that we published last month1, are predicted to hinder the uptake of drought-resistant cotton by smallholders. Few agricultural research objectives have ever attracted the intensity of attention and investment from private, public, academic and philanthropic sectors as drought tolerance. In the past decade, total investment in drought-tolerance research has

almost certainly surpassed $1 billion. With climate change, growing water scarcity and impending water disputes, the prospective welfare gains from effective drought-tolerant varieties are enormous. Among the poor, such varieties may limit catastrophic losses and help households recover from drought and famine. Many proponents also argue that the higher economic security afforded by drought-resistant crops will encourage the households of resource-poor adopters to become more entrepreneurial as a whole. Decades of research in economics and other social sciences have emphasized learning as a central process that influences the uptake of technology. Learning from experience is nowhere more important than with the highly heterogeneous agricultural production of smallholder agriculture in developing countries. Yet, marginal farmers—who typically face poor soils,

nature biotechnology volume 28 number 6 JUNE 2010

erratic weather and limited or no access to irrigation and other inputs—often lack the control required to perceive subtle differences in the value of competing varieties. This background of confounding factors may challenge marginal farmers’ ability to learn how to assess the net gains of droughttolerant varieties for two reasons. First, the relative yield benefit of drought-tolerant varieties is conditional on drought pressure. During seasons with good rainfall, both drought-resistant and conventional varieties appear identical; indeed, marginal farmers may even view drought-tolerant varieties as inferior if increased cost is not associated with benefit or if yield is comparatively inferior when water is plentiful. Second, the relative benefits of drought-tolerant varieties peak at moderate levels of drought; if drought severity increases, these benefits quickly fade. This not only makes it difficult for breeders to test drought-tolerant traits, but also makes it much more difficult for marginal farmers in completely uncontrolled environments to discern differences between drought-resistant and conventional varieties. Contrast this with insect-resistant Bt varieties, which confer perceptible benefits to poor households even when pest pressure is low due to imperfect baseline pest control in most farming regions. What’s more, the heavier the target pest pressure, the more exaggerated the relative performance of Bt varieties—a signal that easily outcompetes any other factor affecting the farmers’ perception of relative merits. In our recent publication1, we built a model to predict the effect of drought presence and farmer perception of relative yield benefit on the uptake of drought-resistant varieties over 100 seasons. In the model, farmers chose to plant either a drought-resistant or conventional variety. The yield of these varieties was determined by the underlying drought stress, which is random. If a farmer growing the drought-tolerant variety interacts with a farmer growing conventional variety (or vice versa), the model assumes the farmer notes that season’s yield difference. According to our model, the subsequent decision to adopt the drought-tolerant or conventional variety is based on this difference. Adoption grows as the population of drought-tolerant adopters grows relative to nonadopters (increasing the probability of observing drought-resistant crop performance) and as the number of seasons increases (offering more opportunities to observe yield differences). We also formulated the model so that the drought-tolerant variety stochastically dominates the conventional variety in terms of farmer expectations, so the 553

© 2010 Nature America, Inc. All rights reserved.

corr e s p ond e nc e question is never whether drought-tolerance is better (it is), but how long it takes a farmer to discover this. Although our model is not specific to a particular crop, we parameterized the model to reflect India’s experience with the diffusion of Bt cotton: after ~10 years, on-average diffusion of the crop was 90%. When we used the same parameters for drought-tolerant varieties, however, it took four times longer to reach the same level of diffusion. Similarly, when we looked at the effect of farmer aversion to risk, vulnerable (highly risk averse) farmers—the ostensible target clientele of many drought-tolerant research efforts in the public sector and in public-private partnerships—took four times longer to reach 90% diffusion than their less vulnerable (and less risk averse) peers. This trend was observed because risk-averse farmers are highly sensitive to extreme drought and to the background context that may occasionally make the drought-tolerant variety look worse than the conventional variety. For example, an extreme drought that stunts even drought-tolerant varieties would likely be catastrophic for a risk-averse farmer. And if, as expected, climate change increases the probability of severe drought, this possibility becomes even more likely and further hampers learning and adoption among vulnerable farmers in particular. Our analysis should in no way detract from the real potential of drought-tolerance research to help poor rural households cope

with and recover from drought. But we hope that our model can inform the research and development process of looming downstream challenges for drought-tolerant varieties. For example, our modeling exercise emphasizes not only the importance of generalized gains in water use efficiency and early maturation traits that confer benefits across a broader range of rainfall outcomes, but also the importance of pricing; indeed, diffusion of drought-tolerant crops is likely to be especially sluggish among vulnerable farmers if their seeds cost more than conventional crops. If we seek to ensure the efficient uptake of drought-resistant varieties, demonstrating effectiveness in laboratories and test plots will be only part of a solution. The quandary of a marginal farmer in drought-prone Africa trying to figure out whether his neighbor’s maize really did better than his own emphasizes how adoption of such varieties is unlikely to be as smooth and rapid as experienced previously with Bt cotton. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Travis J Lybbert1 & Adrian Bell2 1Agricultural & Resource Economics, University

of California, Davis, Davis, California, USA.

2Human Ecology, University of California, Davis,

Davis, California, USA. e-mail: [email protected]

1. Lybbert, T.J. & Bell, A.V. AgBioForum 13, 13–24 (2010).

Health impact in China of folate-biofortified rice To the Editor: Despite efforts to reduce the burden of malnutrition, large numbers of people still consume insufficient micronutrients, including folate1. Folate deficiency, characterized by a suboptimal daily intake of folate (<400 µg) may lead to the onset of diseases and disorders, such as neural-tube defects (NTD), megaloblastic anemia and aggravation of iron-deficiency anemia2. In line with the main micronutrient deficiencies (zinc, iron and vitamin A), folate deficiency is 554

more prevalent in less developed, nonWestern countries. In China, for instance, ~20% of the population is considered to be folate deficient3 (for an overview on folate deficiency and NTDs, see Supplementary Discussion, sections 2 and 3, respectively). In 2007, a report in this journal by Storozhenko et al.4 reported folate biofortification of rice by metabolic engineering. These authors proposed folate biofortification as an alternative to tackle deficiency of the micronutrient and its

adverse health outcomes, such as NTDs. In the following analysis, we present the first attempt to evaluate the health impact of folate-biofortified rice. Analogous to previous impact studies of other biofortified staple crops5, we apply the disability-adjusted life years (DALYs) approach6 to evaluate the potential health benefits of rice with a high folate content in China. Several interventions are available to increase folate intake levels in malnourished populations, including folic acid pills supplementation, folic acid fortification and dietary diversification to increase the consumption of folate-rich foods. Implementation of these approaches can often be problematic, however. For example, in poor, rural regions, such as Shanxi Province in Northern China or Balrampur District in Northern India, where industrial fortification is not well established, folate pill distribution often does not reach the targeted individuals and dietary habits are difficult to alter. In this context, folate biofortification (that is, improving the folate content of staple crops) offers an additional approach for alleviating the burden of folate deficiency (Supplementary Discussion, section 5)2. China is an important study location to evaluate folate-biofortified rice for two major reasons. First, it is not only the world leader in the production and consumption of this staple crop, but also considered one of the pioneers of R&D and commercialization of genetically modified (GM) rice7. In 2009, for instance, China’s Ministry of Agriculture issued a bio-safety certificate to pest-resistant Bacillus thuringiensis toxin (Bt) rice, which should lead to large-scale production of this transgenic crop in about 2 to 3 years8. This makes rice, the world’s main staple crop, an appropriate food vehicle for folate biofortification. Second, China as a riceconsuming country is characterized by large folate deficiencies and high NTD prevalence rates9. Each year, about 18,000 pregnancies in China are affected with an NTD (Supplementary Discussion, Table 4). Because of significant differences in rice consumption, folate status and the prevalence of NTDs between the northern and southern regions, a regional comparison of the health impact of folate biofortification in China will further underpin its ex-ante evaluation. Shanxi Province for instance, has one of the highest reported NTD prevalence rates in the world10, in part because folate intake

volume 28 number 6 JUNE 2010 nature biotechnology

corr e s p ond e nc e

able 1 The current burden of folate deficiency (DALYs lost), maternal folate intake and health impact scenarios (DALYs saved) after T folate biofortification of rice Impact scenario (annual DALYs saved)b Current burden (annual DALYs lost) Region

Total

Per 10,000 persons

Maternal intake after

biofortificationa

µg per day per woman of child-bearing age

Low

High Per 10,000 persons

Total

Total

Per 10,000 persons

North

260,223

4.68

716.92

96,152

1.73

213,149

3.83

Northeastc

197,849

4.31

735.41

73,105

1.59

162,058

3.53

Northwestd

62,374

6.39

638.39

23,047

2.36

51,090

5.23

South

53,957

0.71

1,418.56

19,937

0.26

44,197

0.58

Southeaste

40,252

0.71

1,486.83

14,873

0.26

32,971

0.58

Southwestf

13,705

0.69

1,228.86

5,064

0.25

11,226

0.56

314,180

2.38

1,120.41

116,090

0.88

257,345

1.95

Chinag

© 2010 Nature America, Inc. All rights reserved.

aTotal

folate intake after biofortification is the sum of the current folate intake and the additional absorbed folate due to biofortification, expressed in µg folate per day, per women of childbearing age (CBA). bThe low-impact scenario is defined by a low-coverage rate (36.95%), that is, women willing to accept GM rice and having access to favorable farmers. The high-impact scenario is defined by a high-coverage rate (81.91%), that is, women indifferent or amenable to GM rice and having access to indifferent or enthusiastic farmers (see Supplementary Discussion, section 6.2.). cNortheast China consists of ten administrative areas (Beijing, Tianjin, Hebei, Shanxi, Inner Mongolia, Liaoning, Jilin, Heilongjiang, Shandong and Henan) and is characterized by a high contribution level (85% of NTDs attributable to folate deficiency) and an average daily rice consumption of 186.5 g per person. dNorthwest China consists of five administrative areas (Shaanxi, Gansu, Qinghai, Ningxia and Xinjiang) and is characterized by a high contribution level (85% of NTDs attributable to folate deficiency) and an average daily rice consumption of 156.1 g per person. eSoutheast China consists of 11 administrative areas (Shanghai, Jiangsu, Zhejiang, Anhui, Fujian, Jiangxi, Hubei, Hunan, Guangdong, Guangxi and Hainan) and is characterized by a low contribution level (40% of NTDs attributable to folate deficiency) and an average daily rice consumption of 435.5 g per person. fSouthwest China consists of five administrative areas (Sichuan, Guizhou, Yunnan, Tibet and Chongqing) and is characterized by a low contribution level (40% of NTDs attributable to folate deficiency) and an average daily rice consumption of 348.8 g, per person. gChina consists of thirty-one administrative areas (see notes above), excluding two special administrative regions (Hong Kong and Macau) and Taiwan (Republic of China).

levels are sub-optimal11. In the following analysis, we apply the DALYs method to quantify the burden of a disease as a single index (that is, the number of DALYs lost). This number equals the sum of the ‘years lived with disability’ and ‘years of life lost’, which represent disability-weighted morbidity and cause-specific mortality, respectively. We use this approach to estimate the current burden of the functional outcomes of folate deficiency in a scenario with or without biofortified rice (Supplementary Discussion, section 4). Because of the lack of data on the contribution of folate deficiency to other folate-related health outcomes, only NTDs are included as functional outcomes in this impact study. NTDs result in malformations of the spine (e.g., spina bifida), skull and brain (e.g., anencephaly and encephalocele), which can be both fatal and nonfatal and are considered to be the world’s most common congenital malformations (responsible for one-third of all stillbirths in China12). On the basis of a folic acid supplementation study among pregnant women, Berry et al.13 have estimated that women in northern and southern China are able to reduce the risk of having a baby with an NTD by 85% and 40%, respectively, if they comply with the recommended intake of 400 µg of folate. In other words, 85% and 40% of all NTDs are attributable to folate deficiency in northern and southern China, respectively. The efficacy of folate-biofortified rice depends on the levels of folate that are

bioavailable within the rice. Depending on the transgenic line, folate content in biofortified rice can be 20 to 100 times higher than that in conventional rice. In our calculations, we use 1,200 µg per 100 g raw polished grains. Taking into account potential losses after processing and the bioavailability upon ingestion, the total folate content of biofortified rice comes to 300 µg per 100 g rice. Given this efficacy, the total folate intake after biofortification can be estimated based on the current folate intake and rice consumption data of the Chinese regions (Supplementary Discussion, Table 8). Because of the lack of scientific evidence on the relationship between absorbed folate (‘dose’) and the incidence rate of NTDs caused by folate deficiency (‘response’), each region is evaluated on the assumption that an average daily folate intake >400 µg prevents women from having a baby with an NTD caused by folate deficiency. The success of biofortification also depends on the coverage rate (that is, consumer acceptance of, and accessibility to, folatebiofortified rice). Our study refers to the percentage of women that switch completely to folate-biofortified rice, compared with a group of women that continues to consume traditional rice. A low and high coverage rate are included, based on previous research on acceptance of folate-biofortified rice14. Finally, based on its efficacy and coverage rate, the health benefits of folate biofortified rice can be assessed by comparing the number of DALYs lost

nature biotechnology volume 28 number 6 JUNE 2010

under the current situation and a scenario with biofortification (Supplementary Discussion, section 6). Although women of childbearing age are considered as the target group to reduce folate deficiency, the health impact refers to newborns that benefit from their mother’s biofortified diet. Table 1 gives an overview of the current burden of folate deficiency, the total folate intake of women of childbearing age and the health benefits after folate biofortification of rice. The six Chinese regions are characterized by significant differences in rice consumption and current folate intake levels. The maternal intake after biofortification refers to the introduction of rice with a folate content of 1,200 µg per 100 g rice. As the average folate levels in northeast and northwest China are significantly lower than in southern China, these regions could deploy a transgenic line with a higher folate content (e.g., 1,700 μg per 100 g). Application of the DALY approach shows that the current burden of folate deficiency in China amounts to a loss of 314,180 DALYs per year, of which 72.15% is caused by NTD mortality (Table 1). Although northeast China has the highest number of DALYs lost, the burden of disease is relatively higher in the northwest (that is, when DALYs are expressed per 10,000 persons). On a regional basis, the current situation is most problematic in Shanxi (northeast), Gansu (northwest), Anhui (southeast) and Guizhou (southwest) (Supplementary Discussion, Table 11). 555

© 2010 Nature America, Inc. All rights reserved.

corr e s p ond e nc e According to a low- and a high-impact scenario, implementing folate-biofortified rice in China would save, respectively, 116,090 and 257,345 DALYs per year. The health benefits of folate-enriched rice in China are based on daily folate intakes that are significantly higher than the recommended intake of 400 µg to tackle maternal folate deficiency and the associated risk of having a baby with an NTD caused by folate deficiency. Although rice consumption and daily folate intake in the northern regions are generally lower than in the south, the required daily folate intake can still be achieved if folate-biofortified rice is consumed. This explains the higher number of DALYs saved in high NTD risk regions, such as northeast and northwest China. The findings support folate biofortification of rice as a valuable strategy to reduce folate deficiency and its main adverse health outcome, NTDs. Especially in poor, rural regions where other interventions have little chance of success, folate-biofortified rice seems to be an effective, complementary approach to address folate deficiency. To further improve the evaluation of the health benefits of folate-biofortified staple crops, research is needed to determine the nonlinear relationship between folate deficiency and NTDs and the contribution of folate deficiency to other health outcomes. If folate-biofortified rice were to obtain approval, further thought would be needed as to the optimal method for its introduction in rural China. A possible scenario would be to cross the high-folate trait into rice varieties that have improved agronomic characteristics, such as the pest-resistant Bt rice, to ensure acceptance of farmers and politicians. Besides political approval, the success of folate biofortification in China will be mainly determined by the acceptance of consumers, the cost effectiveness of this intervention and the price of folatebiofortified rice. Even though folate biofortification is a pro-poor and pro-rural intervention, it is only one of many approaches for alleviating the burden of folate deficiency. A combination of policy interventions will probably be most feasible and effective in tackling folate deficiency in all sections of the population. Note: Supplementary information is available on the Nature Biotechnology website. ACKNOWLEDGMENTS With respect to rice consumption data, we acknowledge the assistance provided by F. Gale of the USDA Economic Research Service, Washington, DC. This investigation received financial support from Ghent

556

University, through the Special Research Fund (BOF GOA 1251204). COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Hans De Steur1, Xavier Gellynck1, Sergei Storozhenko2, Ge Liqun3, Willy Lambert4, Dominique Van Der Straeten2 & Jacques Viaene1 1Department of Agricultural Economics, Faculty

of Bioscience Engineering, Ghent University, Ghent, Belgium. 2Unit Plant Hormone Signalling and Bio-imaging, Department of Physiology, Ghent University, Ghent, Belgium. 3Rural Economy Research Institute, Liaoning Academy of Agricultural Sciences, Shenyang, Liaoning, P.R. China. 4Laboratory of Toxicology, Department of Bioanalysis, Ghent University, Ghent, Belgium. e-mail: [email protected] 1. McLlean, E., Benoist, B. & Allen, L. Food Nutr. Bull. 29, S38–S51 (2008).

2. Blancquaert, D. et al. Crit. Rev. Plant Sci. 29, 14–35 (2010). 3. Hao, L. et al. J. Nutr. 133, 3630–3635 (2003). 4. Storozhenko, S. et al. Nat. Biotechnol. 25, 1277–1279 (2007). 5. Stein, A., Sachdev, H. & Qaim, M. Nat. Biotechnol. 24, 1200–1201 (2006). 6. Stein, A. et al. Analyzing the Health Benefits of Biofortified Staple Crops by Means of the DisabilityAdjusted Life Years Approach. HarvestPlus Technical Monograph 4 (International Food Policy Research Institute, Washington, DC, 2005). 7. Wang, Y. & Johnston, S. Nat. Biotechnol. 25, 717–718 (2007). 8. Shuping, N. & Miles, T. Reuters. 2009–11–27 (2009). 9. Zhao, Y. et al. Matern. Child Nutr. 5, 104–116 (2009). 10. Dai, L. et al. Zhonghua Yu Fang Yi Xue Za Zhi (article in Chinese) 36, 402–405 (2002). 11. Zhang, B. et al. Biomed. Environ. Sci. 21, 37–44 (2008). 12. Li, Z., Ren, A., Zhang, L., Guo, Z. & Li, Z. Paediatr. Perinat. Epidemiol. 20, 43–53 (2006). 13. Berry, R. et al. N. Engl. J. Med. 341, 1485–14901 (1999). 14. De Steur, H. et al. Appetite 54, 118–125 (2010).

Alive and kicking To the Editor: As CEO of the companies involved, I would like to bring to the attention of your readers several inaccuracies in a News article in the March issue entitled ‘Resuscitated deCODE refocuses on diagnostics’1. The article erroneously reports that deCODE (Reykjavik, Iceland) “…already shuttered its Emerald Biosciences and Emerald Biostructures drug discovery operations in Bainbridge Island, Washington....” In fact, both Emerald BioSystems (which was misspelled as Emerald Biosciences in the original story) and Emerald BioStructures have never closed—they are vibrant, growing businesses. Both companies have been continuously operating since 1998 with the same management team. On November 13, 2009, they were sold to a group of investors in Delaware (Beryllium). Emerald BioSystems continues to sell research products for protein crystallization (http://www. emeraldbiosystems.com/) and Emerald BioStructures—which before deCODE’s bankruptcy, operated under the name ‘deCODE biostructures’—provides collaborative structural biology services to pharmaceutical companies, biotech companies and academic institutions (http://

www.emeraldbiostructures.com/). The two companies also remain active in the Protein Structure Initiative (http:// www.structuralgenomics.org/). Emerald BioStructures is the lead organization for the National Institutes of General Medical Sciences–funded Accelerated Technologies Center for Gene to 3D Structure (http:// www.ATCG3D.org/) specialized center, and Emerald BioSystems is actively commercializing technologies generated from this center. Emerald BioStructures is also a member of the National Institute for Allergy and Infectious Diseases–funded Seattle Structural Genomics Center for Infectious Disease (http://www. SSGCID.org/), which is solving hundreds of structures of novel anti-infective disease targets every year. COMPETING FINANCIAL INTERESTS The author declares competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturebiotechnology/.

Lance Stewart Emerald BioStructures, Inc. & Emerald BioSystems, Bainbridge Island, Washington, USA. e-mail: [email protected] 1. Ratner, M. Nat. Biotechnol. 28, 192 (2010).

volume 28 number 6 JUNE 2010 nature biotechnology

p at e n t s

Pluripotent patents make prime time: an analysis of the emerging landscape Brenda M Simon, Charles E Murdoch & Christopher T Scott

© 2010 Nature America, Inc. All rights reserved.

An examination of three patents in the fast-moving iPS space may help determine their ultimate value.

W

hen induced pluripotent stem (iPS) cells burst onto the scene in 2007, they brought along with them a new approach to stem cell research, which had previously been restricted to human embryonic stem (hES) cells1,2. Unlike hES cells, which were made from two-day old human embryos and would require cloning technologies to generate genetically matched lines for future therapeutic use, iPS cells were quickly deployed to do better, using somatic cells rather than embryos as the source material. The power of the technique lay in its ability to take any differentiated cell—diseased or otherwise—and reprogram it to an embryonic state, producing an immortal line with an exact genetic match to the donor cell3. The field has moved along at a blistering pace, and this is reflected in the international patent landscape. As of this writing, dozens of applications have been filed internationally, and in the past two years, the first three patents including claims to this technology have issued in Japan, the United Kingdom and the United States. (Tables 1 and 2) We briefly discuss the scope of the three issued patents by examining the extent of the protection as described by their claims. Granted patent landscape The first issued patent, granted in Japan to Shinya Yamanaka on September 12, 2008, was filed on December 12, 2006, with a priority date of December 13, 2005. The fast-tracked patent covers a method for preparing an iPS cell from a somatic cell by introducing the embryonic Brenda M. Simon and Charles E. Murdoch are at the Stanford University Center for Law and the Biosciences and Charles E. Murdoch and Christopher T. Scott are at the Stanford University Program on Stem Cells in Society, Stanford, California, USA. e-mail: [email protected] or [email protected]

transcription factors Oct 3/4, Klf4, c-Myc and Sox2. Claims directed to cells produced by this method were previously filed in Japan and several other countries4. The second patent was granted in the United Kingdom to Kazuhiro Sakurada on January 12, 2010. It was filed on June 13, 2008, with a priority date of June 15, 2007. The Sakurada patent covers a method of inducing human iPS cells from human postnatal tissue by forcing expression of some combination of Oct3/4, Sox2 and Klf4, providing that c-Myc is not included, and culturing in the presence of FGF-2 (ref. 5). According to Sakurada, these cells self-renew and differentiate into ectoderm, mesoderm and endoderm. The elimination of c-Myc in preparing iPS cells is a significant advancement for therapeutic application, as it is a potentially cancer-causing gene. The broadest claims of the Yamanaka and Sakurada methods have three genes in common: Oct 3/4, Klf4 and Sox2. Despite this overlap, the patents cover different methods. Most notably, the Sakurada patent claims expressly require the elimination of c-Myc in preparing iPS cells, whereas to infringe the Yamanaka issued claims, all four genes must be used. The patents have been cross-licensed, protecting against the unlicensed use of either method. Both the Sakurada and Yamanaka patents are part of the portfolio held by iPierian, a company recently formed by the merger of iZumi Bio, a San Francisco Bay Area biotech and Bostonbased Pierian (see p. 544). The iPS cell intellectual property landscape was radically realigned recently with the news of a granted patent that considerably predates the Sakurada and Yamanaka patents. Filed by Rudolf Jaenisch, the patent issued on March 23, 2010 (ref. 6). Jaenisch’s patent, also known as the ‘828 patent, was filed on November 24, 2004, with an earliest possible priority claim to November 26, 2003. The patent’s broadest independent claim— that claim with the largest scope which is not

nature biotechnology volume 28 number 6 JUNE 2010

ancillary to another claim—covers a somatic cell with an endogenous pluripotency gene linked to DNA encoding a selectable marker such that expression of the marker substantially matches expression of the endogenous gene, and an exogenously introduced nucleic acid encoding a pluripotency protein linked to a regulatory sequence. The endogenous gene is expressed in a pluripotent ES cell, is required for pluripotency of the ES cell and is downregulated as the ES cell differentiates. The pluripotency protein is expressed in a pluripotent ES cell and is downregulated as the ES cell differentiates. Another independent claim from the ‘828 patent covers a somatic cell similar to the broadest claim, but requires that the first endogenous pluripotency gene encode Oct4 or Nanog. It furthermore requires that the exogenously introduced nucleic acid encode Oct4, Nanog or Sox2 and be linked to a regulatory sequence. The patent also includes an independent composition of matter claim directed to the line itself, which includes a somatic cell as described in the second claim above, as well as a candidate agent of interest in its potential to reprogram a somatic cell, where the endogenous pluripotency gene encodes Oct4 or Nanog. Below we explore some of the underlying assumptions and limitations of this first patent related to IPS cells to issue in the United States. Is there adequate support for Jaenisch’s claims? The scope of the ‘828 patent might be limited by the disclosure requirements of the Patent Statute7. Those requirements state that to rely on the benefit of its earliest priority date, the Jaenisch patent application would have to adequately describe and enable the claimed invention as of 2003. To determine what invention the application covers, the written description and enablement requirements must first be met. To satisfy the 557

pat en t s Table 1 Characteristics of issued iPS cell patents

© 2010 Nature America, Inc. All rights reserved.

Patent no.

Earliest priority; Composition/ product Process filing date; claims claims Listed genes in broadest claims grant date

Related international publications

JP 2008-131577 Dec. 13, 2005; No (Yamanaka) Dec. 6, 2006; Sept. 12, 2008

Yes

Oct 3/4, Klf4, c-Myc and Sox2.

EP1970446, ZA200804673, US20090068742, KR20080095852, JP2009165480, JP2009165479, JP2009165478, JP2008283972, JP4183742, JP2009165481, WO2007069666, EA200870046, CN101356270, CA2632142, AU2006325975, US20090227032, US20090047263, US20100062533

GB2450603 (Sakurada)

June 15, 2007; No June 13, 2008; Jan. 12, 2010

Yes

Combination of Oct3/4, Klf4 and Sox2 (but not c-Myc)

US20090304646, US20090191159, JP2008307007, WO2009007852, WO2009006997, WO2009006930, AU2008273817

US 7,682,828 (Jaenisch)

Nov. 26, 2003; Yes Nov. 24, 2004; March 23, 2010

No

Endogenous gene expressed in pluripotent ES cell, required for pluripotency, and downregulated during differentiation Exogenous nucleic acid encoding pluripotency protein expressed in pluripotent ESC and downregulated during differentiation

International publications with this priority date not found WO2008124133 (earliest priority claim April 7, 2007)

written description requirement, we need to assess whether the application shows Jaenisch possessed the claimed invention. To determine enablement, we must consider to what extent the patent application, as filed in 2003, would enable one of ordinary skill in the art of stem cell research to make and use the claimed invention without undue experimentation. The US Court of Appeals for the Federal Circuit recently revisited an important 2009 decision on the issue of written description in Ariad v. Lilly8. In this decision, the Federal Circuit invalidated broad patent claims to methods of reducing activity of a transcription regulator. In particular, the court held that the patent failed to provide an adequate description of the molecules that could carry out this inhibition. The Federal Circuit has also provided guidance to determine whether a patent enables one of ordinary skill to make and use an invention without undue experimentation. In In re Wands, the Federal Circuit set forth the factors to be considered in determining whether a patent meets the enablement requirements: (i) the quantity of experimentation necessary, (ii) the amount of direction or guidance presented, (iii) the presence or absence of working examples, (iv) the nature of the invention, (v) the state of the prior art, (vi) the relative skill of those in the art, (vii) the predictability or unpredictability of the art and (viii) the breadth of the claims9. Courts apply these factors in assessing whether an applicant has provided sufficient disclosure to support the claims of the patent. In the discussion below, we draw broadly on these factors in determining the likely scope and impact of Jaenisch’s claims. Which types of somatic cells? The broadest claim of the ‘828 patent covers somatic cells with pluripotency genes, including mammalian cells. To support this claim, the application needs to show that in 2003, Jaenisch 558

possessed the ability to create induced pluripotent stem cells, including in mammals, and enabled others to make and use the invention without undue experimentation. Jaenisch claims a murine-based product that might be extended to any somatic cell line, including mammalian cells. This is reminiscent of a pre-Ariad decision, UC Regents v. Lilly, where the Federal Circuit determined that a patent describing only rat insulin cDNA would not support a claim directed to vertebrate and mammalian insulin cDNA10. The rationale for this limitation was based on the degeneracy of the genetic code. To determine if the ‘828 patent adequately describes mammalian cells, a court would need to determine if one with ordinary skill could visualize the members of the genus based on the description provided. More disclosure must be provided for more unpredictable members of a genus11. A notable feature of the ‘828 patent is its priority date of November 26, 2003. It was not until four years later in 2007 that two independent reports announced successful creation of iPS cells without the use of embryos or using human fibroblasts1,2. The ‘828 patent was also filed considerably earlier than the December 13, 2005 priority date of the Yamanaka patent and the June 15, 2007 priority date of the Sakurada patent. The large gap in filing raises the question of what the state of the art was at the time of the filings, particularly in assessing whether the Jaenisch claims extend to mammalian cells. The fact that his examples are limited to murine reprogramming suggests that the claims to mammalian cells may not be supported by a disclosure that shows possession or teaches others how to make and use the claimed invention. The relatively quick development from reprogramming murine to human cells, however, suggests that perhaps translation between murine and human may not have been unpredictable. Yamanaka’s murine iPS cell lines appeared in

the literature in 2006, followed quickly by his successful reprogramming of human fibroblasts in 2007. Contrast this rapid development to the species-barrier jump for hES cell lines: James Thomson’s 1995 priority date for hES cells was over 13 years after mouse embryonic stem cells were first reported in Nature12. The Thomson patents survived a 2008 reexamination, in which they were found to be nonobvious, because the technique to isolate mouse ES cells was unpredictable and not universally applicable to the isolation of ES cells from other species, particularly human13. In other words, Thomson’s invention could not have been informed by knowledge of the mouse literature at the time. Whereas the lack of iPS cell literature in 2003 similarly suggests that Jaenisch’s invention was nonobvious, it is unclear whether the ‘828 patent will support the claims to human cells with pluripotency genes. Recently, the Board of Patent Appeals and Interferences reversed the reexamination decision for one of the Thomson patents, finding that it would have been obvious to try the known mouse protocols to isolate hES cells14. Which pluripotent genes? Because the ‘828 patent is limited to a somatic cell that requires the introduction of genes, the claims should not cover the introduction of proteins without alteration of the cell’s genome. However, Jaenisch filed at least two continuations to cover different aspects of the invention described in his granted patent that also claim the priority date of November 2003 (ref. 15,16). The specification of the granted patent, which is the same as that of the continuations, mentions various categories of reprogramming agents, including chromatin remodeling agents, pluripotency proteins (protein products of the genes Nanog, Oct4, Stella), and genes important for maintaining pluripotency (Sox2, FoxD3, LIF, Stat3, BMP, PD098059). Although the continuation applications are not yet publicly available,

volume 28 number 6 JUNE 2010 nature biotechnology

pat en t s Table 2 Selected pending iPS cell patent applications Publication no.

Earliest priority

Comments

US20080233610 James Thomson, Junying Yu

March 23, 2007

hES cell foundational patent holder James Thomson’s earliest priority date US patent application relating to the reprogramming of primate cells.

US20090227032 Shinya Yamanaka, Kazutoshi Takahashi, Masato Nakagawa

Dec. 13, 2005 (continuation in part, so only some claims could claim priority to this date)

Earliest priority date US patent application for the Yamanaka team’s work on iPS cells and reprogramming factors.

US20090304646 Kazuhiro Sakurada, Hideki Masaki, Tetsuya Ishikawa, Shunichi Takahashi

June 15, 2007

Earliest priority date US patent application for Sakurada et al.’s work on iPS cells and reprogramming factors.

© 2010 Nature America, Inc. All rights reserved.

EP2145000

Inventor(s)

Rudolph Jaenisch, Jacob Hanna, Marius Wernig, April 7, 2007 Christopher Lengner, Alexander Meissner, Tobias Brambrink, Grant Welstead, Ruth Foreman

European application on iPS cells by Jaenisch’s team with a priority date markedly later than the ‘828 patent, but slightly earlier than several other notable applications.

US20090047263 Shinya Yamanaka, Kazutoshi Takahashi, Masato Nakagawa

Dec. 13, 2005 (continuation in part, so only some claims could claim priority to this date)

Earliest priority date US patent application for the Yamanaka team’s work on iPS cells and reprogramming factors.

US20100062533 Shinya Yamanaka

Dec. 13, 2005 (continuation in part, so only some claims could claim priority to this date)

Earliest priority date US patent application for the Yamanaka team’s work on iPS cells and reprogramming factors.

Jaenisch may attempt to claim these categories in these applications, as he was required to elect a subset of his claims during prosecution of the ‘828 patent17. In the recently issued patent, Jaenisch’s broadest independent claim, and many of the claims dependent on it, may face challenges because they do not specify which “pluripotency genes” are necessary for reprogramming. In the absence of demonstrating reprogramming, the mere mention of several types of pluripotency genes that might be effective may not have adequately described or enabled the invention. And, as Yamanaka demonstrated, uncovering the right combination of transcription factors was not trivial: his experiments used 24 genes in varying combinations. On the other hand, Jaenisch specifies in two other independent claims three potentially useful genes: Oct4, Sox2 and Nanog, which in the end were shown to reprogram cells, albeit inefficiently. Significantly, it now appears that the essential factors are Oct4 and Sox2—two of the genes that Jaenisch listed18. Therefore, if applications corresponding to the Yamanaka and Sakurada patents are examined in the United States, the ‘828 patent might render them obvious. Courts might find it obvious for one of ordinary skill to attempt to use such genes, provided there is a reasonable expectation of success in choosing them from a finite number of predictable solutions19. Given Jaenisch’s early filing date, his patent would then have an advantage in a priority race. Reprogramming with proteins Does the disclosure of the ‘828 patent foreclose later claims to reprogramming with proteins? Although not present in the granted claims, the specification of the patent mentions chromatin remodeling agents and pluripotency proteins (products of the genes Nanog, Oct4 and Stella). The question is whether using proteins

to reprogram would be obvious in light of this disclosure. Would one of ordinary skill have a reasonable expectation of success, and are there a finite number of predictable solutions? This seems doubtful, given the difficulty in constructing and purifying proteins at the time of filing and the inefficiencies encountered six years later20. In light of this need for considerable experimentation as well as the lack of examples, it is unlikely the ‘828 patent provided sufficient information about reprogramming using such protein products in 2003, despite the mention of “candidate agents of interest” in the claims. Finally, the Jaenisch patent relies on selectable, vector-mediated delivery of genes and nucleic acids whereas the field is moving away from gene delivery as a necessary caution for eventual use of iPS cells for human therapeutics. Conclusions If newer methods of reprogramming are not covered by a continuation patent, the reach of the ‘828 patent may be quite narrow, especially given the movement toward reprogramming with proteins rather than genes. Protein reprogramming has taken years since the ‘828 patent’s disclosure, suggesting that it was not entirely predictable. In sum, the ‘828 patent: (i) may render obvious reprogramming using pluripotency genes Oct4 and Sox2; (ii) is unlikely to support claims to all somatic cells, including mammalian cells, with pluripotency genes; and (iii) is unlikely to foreclose reprogramming with pluripotency proteins or chromatin remodeling agents. Although these determinations will ultimately be made at the discretion of a court or during a reexamination by the US Patent and Trademark Office, and are thus uncertain, our analysis suggests the ‘828 patent is not as far-reaching as some have feared and as Fate Therapeutics, a company founded by Jaenisch, triumphantly

nature biotechnology volume 28 number 6 JUNE 2010

pronounced21. However, even if the impact of the ‘828 patent is dulled somewhat, it may still have a lengthy reach. Just as H1 and H9 became standard lines in hES cell research, arguably so too might lines reprogrammed using the ‘828 patent factors. These lines could become the next set of experimental controls, and if they remain in widespread use, the patent could prove to have considerable value. ACKNOWLEDGMENTS Thanks to the Stem Cell Network (Canada) for their generous funding. A special thanks to E. Chiao for his extremely helpful insights. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 1. Takahashi, K. et al. Cell 131, 861–872 (2007). 2. Yu, J. et al. Science 318, 1917–1920 (2007). 3. Park, I. et al. Cell 134, 877–886 (2008). 4. Yamanaka, S. Japan patent JP 2008–131577 (2008). 5. Sakurada, K. et al. UK patent GB2450603 (2010). 6. Jaenisch, R. et al. US patent 7,682,828 (20104). 7. 35 USC § 112. 8. Ariad Pharms., Inc. v. Eli Lilly & Co., 598 F.3d 1366 (Fed. Cir. 2010) (en banc). 9. In re Wands, 858 F.2d 731 (Fed. Cir. 1988). 10. Regents of the University of California v. Eli Lilly & Co., 119 F.3d 1559 (Fed. Cir. 1997). 11. USPTO. Manual of Patent Examining Procedure §§ 2131.02, 2144.08, edn. 8th (US Patent and Trademark Office; 2008). 12. Evans, M.J. & Kaufman, M.H. Nature 292, 154–156 (1981). 13. Vrtovec, K. & Scott, C.T. Nat. Biotechnol. 26, 393–395 (2008). 14. USPTO Board of Patent Appeals and Interferences, Foundation of Taxpayer & Consumer Rights v. Patent of WARF, Appeal 2010-001854, Patent 7,029,913 (April 28, 2010). 15. Jaenisch, R. et al. United States Patent Application No. 12/703,061 (filed 2010). 16. Jaenisch, R. et al. United States Patent Application No. 12/703,015 (filed 2010). 17. USPTO Requirement for Restriction/Election in US Patent Application No. 10/997,146 (mailed May 24, 2006). 18. Yu, J. & Thomson, J.A. Genes Dev. 22, 1987–1997 (2008). 19. In re Kubin, 561 F.3d 1351 (Fed. Cir. 2009). 20. Zhou, H. et al. Cell Stem Cell 4, 381–384 (2009). 21. Normile, D. Science Insider (February 8, 2010).

559

patents

© 2010 Nature America, Inc. All rights reserved.

Recent patent applications in epigenetics Patent number

Description

Assignee

Inventor

CN 101525592

A novel human parthenogenetic embryonic stem cell line with two active X chromosomes prepared by promoting pronucleus from oocyte, inducing embryo formation, maintaining undifferentiated cells and obtaining stem cell line; useful in genetic and epigenetic research, and in regeneration therapy for treating various diseases.

Guangzhou Medical College (Guangzhou, China)

Sun X

Priority application date

Publication date

3/7/2008

9/9/2009

WO 2008021288, Diagnosing breast and/or colorectal cancer in a human EP 2069535 by determining in a test sample a somatic mutation in a gene or its encoded cDNA or protein, which is indicative of breast or colorectal cancer.

8/11/2006 Johns Hopkins University Barber T, Jones S, (Baltimore) Kinzler KW, Lin J, Mandelker D, Parsons DW, Sjoblom T, Velculesu VE, Vogelstein B, Wood LD, Parsons WD

WO 2009049916, A method of determining methylation at cytosine residues EP 2053131 in DNA comprising treating the test sample with a reagent that comprises electrophilic/oxidizing species and selectively reacts with 5-methylcytosine residues; useful for analyzing DNA methylation patterns in epigenetics and for quantifying methyltransferase activity.

Ludwig Maximilian University of Munich (Munich)

KR 2009035372

Korea University Industry Gil J, Kim J & Academy Cooperation Foundation (Seoul)

A method of inducing differentiation of spinal cord oligodendrocyte by culturing human embryonic stem cells to form an embryoid, and culturing the embryoid in proliferation induction culture medium, and then in differentiation induction culture medium. The method is useful for inducing differentiation of spinal cord oligodendrocyte from human embryonic stem cells to prepare cell therapy composition for treating spinal cord disease and spinal cord injury.

Bareyt S, Carell T, Mueller M

2/21/2008, 6/17/2009

10/19/2007 4/23/2009, 4/29/2009

10/5/2007

4/9/2009

WO 2009015283, A new polypeptide for forming a histone complex for WO 2009015283 identifying histone demethylase-histone binding modulator comprising a specific amino acid sequence of human-specific Phe-His-Asp (PHD) finger–containing protein of a lysine-specific histone demethylase complex.

Harvard College (Cambridge, MA, USA), Emory University (Atlanta)

Cheng X, Collins RE, Horton JR, Lan F, Shi Y

7/24/2007

1/2/2009, 4/9/2009

WO 2006031745, A method of sequencing a target nucleic acid by generatUS 20060073501 ing overlapping fragments of the target nucleic acid, contacting fragments with an array of capture oligonucleotides, measuring the mass of hybridized fragments and constructing a nucleotide sequence.

Boecker S, Sequenom (San Diego), van den Boom DJ

Boecker S, van den Boom DJ

9/10/2004

3/23/2006, 4/6/2006

US 6872868

Ohio University (Athens, OH, USA)

Hoppe PC, Wagner TE

5/24/1995

3/29/2005

Cooney CA, Wolff GL

Cooney CA, Wolff GL

6/12/1998, 12/16/1999, 3/29/2003 2/19/2004

A method for genetic transformation of zygotes by placing exogenous genetic material into the zygote nucleus.

WO 1999063943, A method of increasing methylation of DNA in an unborn US 20040033198 offspring, resulting in changes in the epigenetically determined phenotype, inhibition of parasitic DNA sequences and a decrease in the susceptibility to tumor formation.

Source: Thomson Scientific Search Service. The status of each application is slightly different from country to country. For further details, contact Thomson Scientific, 1800 Diagonal Road, Suite 250, Alexandria, Virginia 22314, USA. Tel: 1 (800) 337-9368 (http://www.thomson.com/scientific).

560

volume 28 number 6 JUNE 2010 nature biotechnology

news and views

Raising the bar for cancer therapy models Giulio Francia & Robert S Kerbel

The failure rate of double-blind, often placebo controlled randomized phase 3 trials is higher in oncology than in any other therapeutic area1. In non-small cell lung cancer, for example, with the exception of a bevacizumab (Avastin) trial, every one of over a dozen phase 3 trials combining a ‘targeted’ biologic agent with standard chemotherapy used for first-line treatment has failed to provide an overall survival benefit2,3. In this issue, Singh et al.4 suggest that one source of this dismal record—preclinical studies with animal models and, in particular, with genetically engineered mouse models (GEMMs)—could be improved through proper experimental design and data analysis. Aside from exposing thousands of people to ineffective therapies, failures in late-stage clinical trials contribute substantially to the high cost of most newly approved anti-cancer drugs1. As failures in large phase 3 trials are almost always preceded by encouraging results in smaller phase 2 trials and in earlier preclinical studies, most of which involve mouse tumor models, the need to improve the predictive power of both is obvious4–7. The desire to develop more useful animal models led to work on genetically engineered mouse models (GEMMs) of spontaneous cancer beginning over two decades ago. These models involve stable or conditional manipulation of various cancer-causing genes such that immunocompetent mice spontaneously develop tumors that in many cases are very similar histologically to their human counterparts. So far, however, GEMMs have not been shown to be consistently superior to less-expensive human tumor xenograft models, which have been used extensively for over 35 years5,7,8. Giulio Francia and Robert S. Kerbel are in Molecular & Cellular Biology Research, Sunnybrook Health Science Centre, Department of Medical Biophysics, University of Toronto, Toronto, Ontario, Canada. e-mail: [email protected] or [email protected]

Syngeneic

Subcutaneous xenograft

Orthotopic xenograft

GEMM

–Transplanted –Immunocompetent

–Transplanted –Immunodeficient

–Transplanted –Immunodeficient

–Spontaneous –Immunocompetent KRAS G12D

Metastasis Primary tumor

Metastatic

Nonmetastatic

Metastatic

Katie Vicari

© 2010 Nature America, Inc. All rights reserved.

Can mouse cancer models predict the results of phase 3 clinical trials?

Nonmetastatic

Figure 1 Types of mouse model used to test new cancer therapies. Experimental tumors can be induced by (i) transplantation of syngeneic cancer cells or whole tumors, (ii) subcutaneous transplantation of human cells, (iii) transplantation of human cancer cells in the same (‘orthotopic’) tissue from which the cancer cells originated and (iv) genetic engineering of mice to carry mutated oncogenes and/or tumor suppressor genes leading to a high incidence of spontaneous cancer (in this example, a primary lung cancer).

Singh et al.4 present an illuminating and exhaustive set of experiments showing that GEMMs, if used ‘properly’, may in fact improve the ability to predict the outcomes of phase 3 trials. The authors study two different GEMMs, both involving tumors driven by the mutant Kras oncogene: a non-small cell lung cancer model and a pancreatic carcinoma model4. They assess the outcomes of several different combination treatment regimens involving certain ‘standard-of-care’ (or ‘standard of care’–like) chemotherapy drugs and biologic agents such as erlotinib (Tarceva), an epidermal growth factor receptor antagonist, or monoclonal antibodies against mouse vascular endothelial growth factor, which are similar to bevacizumab. They then compare these results with those from several previous phase 3 trials involving erlotinib and bevacizumab in nonsmall cell lung cancer and pancreatic cancer, in which either a positive or negative efficacy outcome was reported. To undertake this kind of retrospective analysis—the first of its kind—the authors assess tumor growth and response to treatment by serial noninvasive imaging measurements, such as X-ray micro-computed tomography and

nature biotechnology volume 28 number 6 june 2010

high-resolution micro-ultrasound. They are thus able to show benefits in both overall survival and progression-free survival, using statistical criteria employed in clinical trials. This is an exceptional aspect of this work, because progression-free survival is an important consideration in clinical trials and one that has seldom been modeled in preclinical studies (which generally use only tumor growth delay as a surrogate marker). The authors find close, if not compelling, correlations between the clinical trial outcomes and the corresponding GEMM therapy results in most, though not all, of their models using the same or very similar treatment regimens. This concordance is found despite the homogenous genetic background of the mice used, in obvious contrast to the marked genetic heterogeneity of humans, raising the question of whether the impact of pharmacogenomics in large clinical trials is as great as some believe. Another obvious question, especially given the expense of GEMMs, is whether Singh et al.4 have proved that these models are clearly superior to human tumor xenografts. The answer is no—at least, not yet. The authors do not undertake an exhaustive comparative analysis using several xenograft models of mutant KRAS 561

© 2010 Nature America, Inc. All rights reserved.

n e w s a n d v ie w s non-small cell lung cancer or pancreatic cancer (especially metastatic models) and deploying all of the endpoints, statistical methods and imaging modalities used in the GEMM studies. Without such a comparative analysis, and without data showing that GEMMs predict future positive clinical trial results (it is always easier to predict the past), any claim that GEMMs are superior to the alternatives would seem premature. Until such analyses are done, it seems reasonable for investigators to first evaluate the technical, financial and time constraints associated with each type of preclinical model (Fig. 1) and then choose the model best-suited to the question they wish to address. Several previous studies have shown that, by optimizing clinically relevant parameters such as drug dose, exposure and pharmacokinetics, even subcutaneously transplanted primary human tumor xenografts can yield therapeutic results consistent with clinical drug-response experience5,8,9. In addition, transplanted tumors are often easy to surgically resect, making it possible to mimic not only postoperative adjuvant therapy of early-stage microscopic metastatic disease10 but also treatment of advanced visceral metastatic disease11. The different stages of disease progression can respond quite differently to cancer therapy12 and at present are difficult to duplicate in GEMMs. Extensive distant metastases are rare in most of these models, and surgery of the multiple and often asynchronously arising primary tumors is difficult and thus not commonly performed. Moreover, it is well known that bulky, visceral metastatic disease, especially in individuals previously exposed to anti-cancer therapies and whose tumors have become ‘refractory’, can be extraordinarily difficult to treat: such second- or third-line treatment scenarios may

be more readily duplicated experimentally using transplanted tumors, including human tumor xenografts. Regardless of the relative merits of GEMMs and human tumor xenografts, the efforts of Singh et al.4 are to be applauded as they highlight the critical importance of using multiple clinically relevant endpoints and methods to assess tumor therapies in mice. Although such approaches, whether applied to studies with GEMMs or xenografts, would be expensive, the payoffs could be dramatic when one considers the staggering cost of even a single failed randomized phase 3 trial—especially a long-term adjuvant therapy trial. Research funding agencies and especially pharmaceutical companies should take note. ACKNOWLEDGMENTS We thank U. Emmenegger as well as members of the Kerbel Lab, in particular C. Hackl, C. Milsom and W. Cruz-Munoz, for their comments and suggestions. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/ naturebiotechnology/. 1. Kola, I. & Landis, J. Nat. Rev. Drug Discov. 3, 711–715 (2004). 2. Sandler, A. et al. N. Engl. J. Med. 355, 2542–2550 (2006). 3. Rossi, A. et al. Curr. Drug Discov. Technol. 6, 91–102 (2009). 4. Singh, M. et al. Nat. Biotechnol. 28, 585–593 (2010). 5. Kerbel, R.S. Cancer Biol. Ther. 2, 108–113 (2003). 6. Frese, K.K. & Tuveson, D.A. Nat. Rev. Cancer 7, 645– 658 (2007). 7. Talmadge, J.E., Singh, R.K., Fidler, I.J. & Raz, A. Am. J. Pathol. 170, 793–804 (2007). 8. Peterson, J.K. & Houghton, P.J. Eur. J. Cancer 40, 837–844 (2004). 9. Inaba, M. et al. Cancer 64, 1577–1582 (1989). 10. Ebos, J.M.L. et al. Cancer Cell 15, 232–239 (2009). 11. Munoz, R. et al. Cancer Res. 66, 3386–3391 (2006). 12. Francia, G. et al. Clin. Cancer Res. 15, 6358–6366 (2009).

Scalable pluripotent stem cell culture Larry A Couture Large-scale production of human embryonic stem cells will require improved culture methods. Only a dozen years after they were first isolated, human embryonic stem cells (hESCs) are beginning to move from the research Larry A. Couture is at the Beckman Research Institute at City of Hope, Duarte, California, USA. e-mail: [email protected]

562

laboratory toward the clinic. Several biotech companies have initiated hESC clinical programs; at least two cell therapies based on hESCs have been submitted to the US Food and Drug Administration under Investigational New Drug (IND) applications; and hESC cultures are being retooled for disease modeling and drug screening. With large-scale applications of hESCs on

the horizon, a significant challenge is the establishment of controllable, reproducible and scalable culture methods that preserve growth rates, genetic stability and pluripotency. Several recent papers in Nature Biotechnology describe progress toward this goal, with three studies1–3 in this issue presenting fully defined surfaces for hESC culture and a report4 in April demonstrating the derivation and culture of hESCs in suspension (Fig. 1). These papers represent important steps in the development of culture technologies suitable for industrial applications of hESCs. In considering the challenges in adapting hESCs to common practices for scale-up manufacturing, it’s important to remember that these cells are not transformed cell lines but unique diploid primary cells with indefinite self-renewal capability. It is therefore not surprising that, like many other primary cell types, hESCs are highly sensitive to their cell culture microenvironment. This sensitivity, which includes a dependence on poorly understood cell-to-cell and extracellular matrix interactions, has made it difficult to maintain hESCs in the pluripotent state even at the research laboratory scale. Interactions with extracellular matrix, which are mediated by cell adhesion molecules such as integrins on the surface of hESCs, are often provided in culture by using feeder layers derived from mouse embryos or by coating cell culture vessels with animal-derived protein substrates. Even when they are described as ‘defined’ reagents, these substrate materials are typically human or animal tissue extracts that vary between lots and therefore require time-consuming testing. Cell-to-cell interactions, including those mediated by E-cadherin5 and possibly gap junctions6, can provide additional signaling to optimize hESC growth and may underlie the tendency of these cells to grow as colonies rather than as the uniformly distributed monolayers characteristic of transformed cell lines. Enzymatic dissociation of hESCs during passaging leads to the loss of these important interactions with the microenvironment and significant cell death. For this reason, many laboratories passage hESC lines by mechanically fragmenting cell colonies into clusters or clumps—a tedious, inefficient and difficult process with limited reproducibility. Mechanical passaging and reliance on poorly defined animal extracts in the substrate and media make common hESC culture practices inadequate for scale-up or for use in sensitive screening systems. Whether it is to support large-scale in vitro screening systems or the manufacture of cell products for preclinical and clinical studies, a robust hESC culture system would have the same basic attributes: it would allow control of cell density and distribution, be reproducible,

volume 28 number 6 june 2010 nature biotechnology

Katie Vicari

n e w s a n d v ie w s

Peptide-acrylate surface

Recombinant human laminin-511

PMEDSAH

Suspension

© 2010 Nature America, Inc. All rights reserved.

Figure 1 New methods for the culture of human pluripotent stem cells. Fully defined adherent culture substrates composed of synthetic1,3 or recombinant2 molecules and a suspension culture medium4 mark progress toward the establishment of reproducible and scalable culture systems for hESCs and iPSCs.

maintain critical cell characteristics such as pluripotency and viability, provide sufficient yields of cells and be free of undefined contaminants. Furthermore, these processes should not involve mechanical dissociation of cells during passaging. Although current hESC culture practices fall short of meeting these criteria, considerable progress has been made in several areas over the past few years, notably in the development of defined culture media. Several commercially available media (from Invitrogen and Stem Cell Technologies) and at least one cell substrate material (Invitrogen’s CeLLStart) consist of relatively well-defined components. Although some of these components include nonrecombinant animal or human-derived materials that may lead to lot-to-lot variability, they have proved sufficiently defined to be used in scale-up manufacturing in support of early clinical studies. The papers by Melkoumian et al.1, Rodin et al.2 and Villa-Diaz et al.3 address the challenge of identifying hESC culture substrates that are fully defined, which should promote product uniformity, reproducibility of experimental results and scalability. Melkoumian et al.1 and Villa-Diaz et al.3 use several synthetic materials, and Rodin et al.2 use a recombinant form of laminin-511, a component of extracellular matrix. All three papers demonstrate that their surfaces broadly support the maintenance of hESC pluripotency, viability and growth rate. Whereas Melkoumian et al.1 use a fully defined, though proprietary, medium with recombinant human growth factors, Rodin et al.2 and VillaDiaz et al.3 use defined media supplemented with nonrecombinant animal or human proteins. The use of proprietary reagents raises issues of supply chain reliability and limits the ability to adjust media components to address cell adaptation and manufacturing problems. In addition, each substrate in the three papers is tested on only a limited number of cell lines, and one paper3 reports variation in the achievable passage number over the lines tested, suggesting

that each new cell line may have to be screened against a panel of defined substrates. In a different approach, Steiner et al.4 circumvent the need to precoat culture vessels with exogenous substrates by demonstrating that hESCs can be derived and grown in suspension. Advantages of an effective suspension culture system would include increased control over cell seeding density and distribution, ease in scaling up to larger volumes, and facile feeding and harvesting. Although suspension adaption of hESCs has been reported7,8, these studies typically showed only low passage number or poor final population expansion. Steiner et al.4 begin with a medium used to support differentiation of hESCs into neural spheres and supplement it with several extracellular matrix components and neurotrophic and growth factors. They acknowledge that they have not confirmed the requirement for all of the components in their system, but they are able to demonstrate impressive results with three hESC lines. Notably, the medium supports the derivation of hESC lines from embryos with high efficiency in suspension, an accomplishment that promises to greatly simplify the generation of new lines. Unfortunately, although cell-doubling rates in the new medium are good, expansion rates are not on par with those of adherent culture. This appears to be due in large part to cell death associated with mechanical trituration during passaging. Mechanical passaging is used in all four studies1–4, underscoring the remaining, and perhaps greatest, challenge to be addressed in establishing reproducible and robust large-scale culture systems. Although mechanical dissociation of colonies during passaging is currently the most reliable method for maintaining hESC pluripotency and viability, it is ill suited to largescale manufacturing because it is labor intensive, inefficient and not amenable to standard operating procedures. How best to protect cells during enzymatic passaging remains to be determined,

nature biotechnology volume 28 number 6 june 2010

but efforts to identify the relevant signaling pathways, such as rho-associated kinase (ROCK) signaling9, suggest possible strategies. For example, a ROCK inhibitor, Y-27632, was shown to reduce apoptosis induced by dissociation of hESCs10. However, the long-term effects of ROCK inhibitors on hESCs are unclear, and Y-27632 has been associated with aneuploidy11, which may affect the cell cycle and is implicated in cell transformation. Another issue that must be considered in the development of defined, scalable hESC culture systems is whether they are compatible with current culture methods. Transferring cells from one culture condition to another may promote genetic and epigenetic changes that have long-term, perhaps subtle, consequences on hESC pluripotency or on the characteristics of differentiated progeny. Testing would address these concerns, but many research laboratories may be reluctant to risk ongoing efforts by adopting ‘improved’ reagents or may not have sufficient resources to perform the necessary comparative studies. Thus, until new culture reagents and methods are fully vetted by the broader research community, there may be unanticipated differences in manufactured hESCs and their differentiated progeny. In the foreseeable future, a number of products derived from hESCs or induced pluripotent stem cells (iPSCs) are expected to enter clinical testing. Last year, Geron and Advanced Cell Technology submitted IND applications, and the California Institute of Regenerative Medicine provided preclinical translational grants totaling over $220 million to 14 projects, many of which involve hESCs or iPSCs. The objective of these grants is to bring stem cell therapies to the IND stage within the next 4 years. These and other efforts underscore the need to establish scalable hESC manufacturing processes to allow preclinical development of hESC-derived therapeutics under conditions similar to those that will be required for manufacturing. COMPETING FINANCIAL INTERESTS The author declares no competing financial interests. 1. Melkoumian, Z. et al. Nat. Biotechnol. 28, 606–610 (2010). 2. Rodin, S. et al. Nat. Biotechnol. 28, 611–615 (2010). 3. Villa-Diaz, L.G. et al. Nat. Biotechnol. 28, 581–583 (2010). 4. Steiner, D. et al. Nat. Biotechnol. 28, 361–364 (2010). 5. Xu, Y. et al. Proc. Natl. Acad. Sci. USA 107, 8129– 8134 (2010). 6. Wong, R. et al. Stem Cell Rev. 4, 283–292 (2008). 7. Singh, H. et al. Stem Cell Res. published online, doi:10.1016/j.scr.2010.03.001 (12 March 2010). 8. Olmer, R. et al. Stem Cell Res. (Amst.); epub ahead of print doi:10.1016/j.scr.2010.03.005 (2010). 9. Krawetz, R. et al. Bioessays 31, 336–343 (2009). 10. Watanabe, K. et al. Nat. Biotechnol. 25, 681–686 (2007). 11. Riento, K. & Ridley, A.J. Nat. Rev. Mol. Cell Biol. 4, 446–456 (2003).

563

n e w s a n d v ie w s

Complex molecular dynamics in the spotlight Lois Pollack & Watt W Webb

The ability to measure events at the singlemolecule level promises to reveal the workings of biological machines in unprecedented detail1. Among the various technologies that can achieve single-molecule resolution, the zero-mode waveguide (ZMW) is emerging as a powerful method with unique capabilities. Recently, a team at Pacific Biosciences (Menlo Park, CA, USA) reported the use of ZMWs for single-molecule DNA sequencing2; now, researchers from this company in collaboration with academic scientists3 have applied these nanometer-sized chambers to monitor the tran sit of individual fluorescently labeled transfer RNAs (tRNAs) through the ribosome4. The new study highlights the potential of this technology to elucidate a range of biological processes beyond DNA replication. Single-molecule measurements can ensure detection of one molecule at a time by using very low sample concentrations. But this requirement poses a substantial challenge in studying the operation of complex machines. For instance, understanding the ribosome would require monitoring the arrival and departure of tRNAs. If tRNA is present at the low concentrations of traditional single-molecule experiments, long delays between the arrivals of tRNAs could allow molecules to become photobleached while present in the illuminated volume and no longer detectable by their fluorescence. Thus, the signals of interest could vanish during the experiment, defeating its purpose. ZMWs provide an elegant solution to these problems. Each waveguide has a subwavelength-diameter nanofabricated hole in metal film that restricts light to a zeptoliter volume, typically ~50 nm deep. The dimensions of the waveguide are too small to permit light propagation in ‘modes’—hence the term ‘zeromode’. Waveguides are arranged in a regular array in an otherwise opaque microlayer on a glass substrate5. Light samples to a depth of ~30 nm from the bottom of the ZMW. Using sophisticated biomolecular attachments, macromolecular machines such as a ribosome or DNA polymerase can be localized to the floor of the waveguide, within the Lois Pollack and Watt W. Webb are in the School of Applied and Engineering Physics, Cornell University, Ithaca, New York, USA. e-mail: [email protected]

564

illuminated volume. A great advantage of the ZMW approach becomes evident when comparing the tRNA transit time through the ribosome with the photobleaching time. As excitation volumes are limited to 10−21 liter, the concentration of fluorescently labeled tRNA can be orders of magnitude higher than in more traditional single-molecule studies, ensuring tRNA concentration–dependent reaction times that are far shorter than photobleaching times. Freely diffusing labeled tRNAs move rapidly through the excitation volume, whereas molecules that interact with the machine have longer-lived fluorescence signals due to their long residence times. In this way, the ZMW provides the benefits of single-molecule studies while enabling measurements of frequent intermolecular interactions. As demonstrated in the recent report of DNA sequencing with ZMWs2, because many waveguides can be sampled simultaneously, single-molecule sensitivity is achieved while accumulatingdata for statistically significant numbers of events. Uemura et al.3 employ three distinctly labeled tRNA complexes to monitor occupation of the three ribosome sites (Fig. 1a). Captive tRNAs signal their presence by emitting light of a particular color (red for one amino acid, blue for the second, green for the third) and remain in the miniscule, illuminated volume as they transit through the ribosome, moving from one internal ribosomal site to the next. Dwell times are measured directly and can be correlated with the distinct function of each of the three internal sites: codon recognition (A site), amino acid attachment (P site) and departure (E site). Thus, the ZMW provides a platform for testing theories about the interplay of distinct tRNA binding sites, including co-occupancy. Interestingly, simultaneous occupation of the three sites was rarely observed in these experiments, suggesting rapid tRNA release from the ribosome under normal circumstances. This technique is bound to contribute to our understanding of frequent or ‘ordinary’ events that enable translation. Advances in understanding the motions of the entire machine (reviewed in ref. 6), notably those relying on positioning fluorescent labels within the ribosome itself7, can be used to correlate passage of the tRNA with molecular motion, revealing the mechanisms of translation in

Mark A. Williams

© 2010 Nature America, Inc. All rights reserved.

Zero-mode waveguides illuminate the process of translation.

Figure 1 Zero-mode waveguide (ZMW) technology reveals insights into the actions of single molecules and nanomachines. (a) Application of ZMW technology to monitor single ribosomes during translation. A translating ribosome is anchored to the floor of a nanopore waveguide. Because of the small diameter of the nanopore, light excitation from below is confined to a spotlight that illuminates a nanoscopic volume. Fluorescently labeled tRNAs transit through the ribosome as they translate codons (represented as shapes along the template strand) into amino acids (colored balls). Because transit times through the ribosome exceed the diffusion times of free tRNAs, long bursts of light accompany the translation process. Information about translation is acquired by observing the sequence and color of long flashes. (b) In a ‘leaky’ ZMW, a thin fibril of polarizable dielectric (silicon) is coated with a conducting metal (gold) surface layer. It is internally illuminated by laser light entering the base end and dimensionally designed to optimize the maximized spatial resolutions of the ‘leaky’ exiting external radiation in both axial and lateral directions for optimal localized intensity near the exit surface (to ~90 and ~30 nm).

more detail than current knowledge allows. In addition, ZMW experiments enable detection of ‘rare’ events, such as co-occupancy of all three sites, and can correlate site population with changes in translation. Insights can be gained into important questions such as

volume 28 number 6 june 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

n e w s a n d v ie w s the potential role of tRNA in ‘shifting’ the underlying substrate by one letter (one-third of a codon), which results in modification of all subsequent ‘words’, known as ‘frameshifting’8. This process allows a single message to code for more than one protein, depending on the selection of the sentence’s first letter. Another question that might be addressed is translational ‘errors’ resulting from tRNA misincorporation, estimated to occur at rates of ~1 event in 1,000–10,000 (e.g., ref. 9). This powerful technology is not restricted to monitoring biological machines. Other interesting applications of ZMWs are readily envisioned. For instance, the small, aluminumcoated pores of the ZMW can be filled with a highly polarizable dielectric that can transmit bright nanoscopic light to form a thin, smalldiameter focal volume across the nanoscopic open end (Fig. 1b). Such a ‘leaky’ ZMW provides a new nanoscopic optical resolution for precisely scannable three-dimensional resolved light microscopy. Computations suggest external focal volumes with dimensions smaller than 30 nm in diameter and 60 nm

in depth10 just outside the tip of the ZMW (Fig. 1b). This technology can extend the capabilities of near-field scanning optical microscopy, which scans spherical focal volume illumination through a small aperture in a thin metal coating on a glass fiber11. This enables high-resolution imaging on flat target surfaces like electronic circuits but not on biological preparations due to force-field perturbations11. Waveguides that are selectively transmissive could permit applications in fragile systems such as high-resolution monitoring of the dynamics of intercellular interactions or of protein conformation fluctuations in reactive systems. Studies that rely on zeptoliter-effective focal volumes have become a focus of intensive research endeavors. According to the ISI Web of Knowledge, ~115 articles aimed at sub-wavelength optical resolutions were published in the last 15 months. ‘Leaky’ ZMWs, in conjunction with switchable fluorescent proteins (e.g., ref. 12), may contribute to this field by providing novel high-resolution molecular imaging strategies and sensitive diagnostics.

Biophysicists can look forward to further improvement of ZMW-based methods in future efforts to resolve the nanoscopic, zeptoliter and single-molecule challenges facing contemporary biology. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/ naturebiotechnology/. 1. Weiss, S. Science 283, 1676–1683 (1999). 2. Eid, J. et al. Science 323, 133–138 (2009). 3. Uemura, S. et al. Nature 464, 1012–1017 (2010). 4. Green, R. & Noller, H.F. Annu. Rev. Biochem. 66, 679–716 (1997). 5. Levene, M.J. et al. Science 299, 682–686 (2003). 6. Marshall, R.A., Aitken, C.E., Dorywalska, M. & Puglisi, J.D. Annu. Rev. Biochem. 77, 177–203 (2008). 7. Fei, J., Kosuri, P., MacDougall, D.D. & Gonzalez, R.L. Mol. Cell 30, 348–359 (2008). 8. Gesteland, R.F. & Atkins, J.F. Annu. Rev. Biochem. 65, 741–768 (1996). 9. Zaher, H.S. & Green, R. Cell 136, 746–762 (2009). 10. Xu, H., Zhu, P., Craighead, H.G. & Webb, W.W. Opt. Commun. 282, 1467–1471 (2009). 11. Betzig, E. et al. Biophys. J. 49, 269–279 (1986). 12. Lippincott-Schwartz, J. & Patterson, G.H. Trends Cell Biol. 19, 555–565 (2009).

Detecting methylated bases in real time In many organisms the primary DNA structure is covalently modified to regulate, for example, gene expression and genome structure. In eukaryotes, the dominant modification is methylcytosine, although others, such as hydroxymethylcytosine, have been detected. In bacteria, both methylcytosine and methyladenine are observed frequently. None of the currently available sequencing platforms can directly detect modified bases, and researchers rely on indirect methods such as bisulfite treatment, methylation-sensitive restriction enzyme mapping or affinity precipitation methods. In a recent paper1 in Nature Methods, researchers at Pacific Biosciences (Menlo Park, CA, USA) have now shown that a

single-molecule, real-time, sequencing-by-synthesis platform based on their zeromode waveguide technology2 can distinguish methylcytosine, hydroxymethylcytosine and methyladenine from unmodified deoxynucleotides in sequences whose methylation patterns are known. As described previously2, the base sequence is determined by monitoring incorporation into the growing chain by a single DNA polymerase of nucleotides tagged with four different fluorescent colors. In the new study, the presence of covalent modifications in the template strand is identified through two kinetic parameters: the time interval between the addition of adjacent nucleotides and the length of each catalytic cycle (beginning with the binding of

nature biotechnology volume 28 number 6 june 2010

the fluorescent base to the enzyme and ending with the release of the fluorophore attached to the terminal phosphate of the nucleotide). Both parameters are influenced by the presence of methylcytosine, hydroxymethylcytosine and methyladenine, not only at positions opposite the incoming nucleotide but also at several adjacent positions. The authors use synthetic templates and DNA purified from Escherichia coli to define the kinetic signatures of specific modifications at a given position. Although de novo determination of methylation patterns is not reported, the detection of methyladenine

seems feasible using circular templates that allow the repeated interrogation of each base. Robust detection of methylcytosine and hydroxymethylcytosine and of multiple modified bases in close proximity (as in CpG islands) will require further optimization of the method. Markus Elsner 1. Flusberg, B.A. et al. Nat. Methods, published online, doi:10.1038/ nmeth.1459 (9 May 2010). 2. Eid, J. et al. Science 323, 133–138 (2009).

565

r e s e a r c h h i g h li g h t s

© 2010 Nature America, Inc. All rights reserved.

Taking aim at transcription factors The transcription factor BCL6 facilitates the generation of antibody diversity in B cells by repressing the DNA damage–sensing apparatus, thereby creating genomic instability. But when BCL6 activity goes awry by mutation or translocation, unregulated B-cell growth can ensue and is often associated with diffuse large B-cell lymphomas (DLBCL). Now Cerchietti and colleagues have isolated small molecules that interfere with the interaction between BCL6 and its co-repressors that show activity against DLBCL cells in vitro and in vivo. Although one of a family of transcription factors with a particular binding region called BTB, BCL6 has a unique lateral groove that interacts with co-repressor molecules. Using computeraided design, the researchers screened over a million commercially available small molecules for those that might bind that region; molecules were grouped according to their structure, and some from the largest group were chosen for testing. The selected molecules were found to specifically bind BCL6; no binding was observed with other BTB-containing transcription factors. In BCL6positive cell lines, the molecules blocked repression of several BCL6 targets, among them tp53, cd69 and cd44, which are involved in checkpoint maintenance. Finally, the molecules killed BCL6-dependent lymphoma cells in culture and when transplanted into severe combined immune-deficient mice. Whereas small molecules that target protein-protein interactions have been previously described, this is the first to target a transcription factor. Although more work needs to be done to maximize the potential of the active molecules, this study does suggest a new approach to treating B-cell lymphomas. (Cancer Cell 17, 400–411, 2010) LD

Deeper tumor-specific drug delivery The efficacy of many anti-cancer drugs is compromised by their inability to penetrate tumors more than a few cell diameters from the vasculature. The tumor-penetrating peptide iRGD is known to home to tumors by binding to αv integrins, and to then penetrate cancerous tissue by virtue of exposure of a motif that confers affinity for neuropilin-1. Chemical conjugation of iRGD to drugs can promote tumor-selective uptake, but it is laborious, may not be feasible for the full range of approved chemotherapies and might even impair drug activities. Ruoslahti and colleagues show that systemic coadministration of unconjugated iRGD with either free doxorubicin, liposome-borne doxorubicin, trastuzumab (Herceptin) or nanoparticle albumin-bound paclitaxel (Abraxane) promotes drug uptake by tumors as much as 40-fold in mouse models of breast and prostate cancer. Free iRGD also boosts uptake of both iron oxide and phage-based nanoparticles by prostate tumor xenografts in mice. There is no evidence that the peptide increases tumor metastasis. (Science 328, 1031–1035, 2010) PH

Targeting hepatitis C assembly More than 20 years after the discovery of hepatitis C virus, no drug specifically targeting viral proteins is approved for clinical use. Gao et al. and Lemm et al. have now identified compounds that target the viral protein NS5A. Whereas most previous anti-viral molecules have targeted viral Written by Kathy Aschheim, Laura DeFrancesco, Markus Elsner & Peter Hare

566

enzymes, NS5A has no known enzymatic function. It is involved in amplification of viral DNA and regulates the assembly of infectious particles, although details remain unknown. The lead compound, BMS-790052, is active against all hepatitis genotypes tested in cell culture. Early clinical trial results in eight individuals infected with genotype 1a or 1b viruses are promising. Oral administration of a single dose leads to an almost 2,000-fold reduction of viral titers and the low levels are maintained for 1 week. The mechanism of action of BMS-790052 still needs to be elucidated. However, the location of resistance mutations suggests that it might disrupt the formation of dimers of NS5A. In vitro results imply synergistic effects between BMS-790052 and inhibitors of the viral protease NS3 or the DNA polymerase NS5B. (Nature 465, 96–100, 2010; J. Virol. 84, 482–491, 2010) ME

miRNAs in cancer networks It is widely appreciated that single microRNAs (miRNAs) frequently control expression of multiple genes and that single mRNA transcripts can be controlled by multiple miRNAs. Yet, instead of aiming to comprehend the complex coordination of miRNA activities, most efforts to elucidate the functions of miRNAs have studied them in isolation. Croce and colleagues illustrate the potential of a systems biology approach to understanding the roles of miRNAs in gene regulation. Using miRNA expression profiles from ~1,000 human samples collected from 50 normal tissues, they show that each cell type is characterized by a distinctive network, with certain miRNAs playing a more critical role than others. Comparison of these networks with those obtained after analysis of >3,000 neoplastic samples from 51 cancer types reveals that all tested cancer types fragment the miRNA network found in healthy cells into several smaller clusters of miRNAs with coordinated activities. The authors conclude that independently regulated miRNAs defined by discrete miRNA subnetworks in cancer cells identify genes involved in cancer-related pathways. They validate this proposal by showing that deregulated miRNAs associated with leukemia in a Mir155 transgenic mouse model map to the vicinity of the miR155 hub in the cancer network. (Genome Res. 20, 589–599, 2010) PH

Cancer metabolism modulator In the 1920s, biochemist Otto Warburg identified a puzzling feature of cancer metabolism. Whereas normal cells catabolize glucose through oxidative phosphorylation, generating >30 molecules of ATP per molecule of glucose, cancer cells favor the less-efficient fermentation pathway, which yields only 2 molecules of ATP. Why this is so remains elusive nearly a century later, but the possibility of exploiting this difference in molecularly targeted therapies appears promising. Building on their earlier preclinical studies, Michelakis et al. have begun to test one such approach in a small-scale clinical trial for glioblastoma. The compound dichloroacetate was known to shift metabolism away from fermentation toward oxidative phosphorylation (by inhibiting an inhibitor of the mitochondrial pyruvate dehydrogenase complex) and has been studied as a treatment for lactic acidosis in metabolic disorders unrelated to cancer. Working with five patients, the authors identified a dosing regimen that altered glioblastoma cell metabolism in vivo without causing serious side effects. Although the trial was not designed to measure anti-tumor efficacy, by comparing patient samples from different time points, the authors documented increased activity of pyruvate dehydrogenase, depolarization of mitochondria, increased apoptosis of glioblastoma cells, activation of p53 and decreased angiogenesis. (Sci. Transl. Med. 2, 31ra34, 2010) KA volume 28 number 6 JUNE 2010 nature biotechnology

A n a ly s i s

Comparative assessment of methods for aligning multiple genome sequences

© 2010 Nature America, Inc. All rights reserved.

Xiaoyu Chen & Martin Tompa Multiple sequence alignment is a difficult computational problem. There have been compelling pleas for methods to assess whole-genome multiple sequence alignments and compare the alignments produced by different tools. We assess the four ENCODE alignments, each of which aligns 28 vertebrates on 554 Mbp of total input sequence. We measure the level of agreement among the alignments and compare their coverage and accuracy. We find a disturbing lack of agreement among the alignments not only in species distant from human, but even in mouse, a well-studied model organism. Overall, the assessment shows that Pecan produces the most accurate or nearly most accurate alignment in all species and genomic location categories, while still providing coverage comparable to or better than that of the other alignments in the placental mammals. Our assessment reveals that constructing accurate whole-genome multiple sequence alignments remains a significant challenge, particularly for noncoding regions and distantly related species. With the rapid sequencing of many related genomes, comparative sequence analysis has emerged as one of the most important areas of computational biology. The fundamental tool of comparative sequence analysis is multiple sequence alignment. As an example of alignments that are intended for comparative sequence analysis, consider the whole-genome multiple sequence alignments of the UCSC Genome Browser1. Sophisticated analyses rely implicitly on the correctness of such alignments. For instance, it is standard practice to search for regulatory elements by scanning the regulatory regions of such whole-genome alignments to identify short windows that are well conserved across the species2,3. Similar conservation-based applications include gene prediction 4,5, noncoding RNA prediction6,7 and, more generally, predicting genomic elements that are under purifying selection8–13. In regions where the sequences are misaligned, these methods may fail to find conserved sites that exist. Downstream applications of genomic multiple sequence alignments are not limited to identifying regions under purifying constraint. Other important applications include inference of phylogeny14,15, estimates of substitution rates15,16, understanding of evolutionary mechanisms17,18 and identification of regions under positive selection11,19–23. Department of Computer Science and Engineering, Department of Genome Sciences, University of Washington, Seattle, Washington, USA. Correspondence should be addressed to M.T. ([email protected]). Received 21 December 2009; accepted 27 April 2010; published online 23 May 2010; doi:10.1038/nbt.1637

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

Because misaligned sequences could easily produce false signals of evolutionary change, these downstream applications are at greater risk of a loss of accuracy when sequences are misaligned15. The many existing multiple-alignment tools often produce quite different alignments when applied to the same set of input sequences10,15,24, leading users to wonder which alignment, if any, is ‘right’. Because of this, a number of recent reviews and articles10,25–30 have made compelling pleas for methods to assess the accuracy of genomic multiple sequence alignments and to compare the alignments produced by different tools. We address this issue here. Recently, the ENCODE Multi-Species Sequence Analysis team used four different pipelines to align 1% of the human genome to 27 other vertebrate genomes10. The four alignment tools are TBA31, MAVID32, MLAGAN33 and Pecan34. The four ENCODE alignments provide a rich resource for comparison of whole-genome alignment tools. What makes these alignments an unprecedented test bed for comparison is that four expert teams used four different alignment methods to align the same 28 vertebrate sequences, spanning 554 Mbp of sequence in total. What makes such a comparison a challenge is the number of dimensions to be taken into account: how much agreement is there among the alignments? Which method is most accurate in aligning distantly related species? How do the methods compare in accuracy in coding and noncoding regions? Which methods align more input sequence than the others? When one method aligns more input sequence than the others, how accurate are these additional aligned regions? Margulies et al.10 performed the first comparative analyses of these four alignments. They compared estimates of sensitivity, which is the fraction of orthologous residues that are correctly aligned (using as proxies coverage of human coding sequences and ancestral repeats) and estimates of specificity, which is the fraction of aligned residues that are truly orthologous (using as proxies coding sequence periodicity and nonalignment to human Alu sequences). Our comparative assessment is more comprehensive than the initial assessment of Margulies et al. They estimated the alignment coverage and accuracy by extrapolating from coding regions and repeats. In contrast, we compare alignment coverage and accuracy at all sites, broken down by location into four categories: coding, UTR, intronic and intergenic. Margulies et al. restricted their analyses to mammalian alignment, omitting chicken, Xenopus, tetraodon, fugu and zebrafish. We include all aligned vertebrate species, and discover that some of the most dramatic differences occur in these distant species. Our analyses are divided into three types. 1. We measure precisely the agreement and disagreement among the alignments. The purpose of this analysis is to establish that differences among the alignments are substantial; it is not 567

A n a ly s i s

Human residues (kbp)

Coding

TBA

MAVID

MLAGAN

Pecan

600 500 400 300 200 100 0

Human residues (kbp)

b

UTR

500 400 300 200 100 0

Human residues (kbp)

c

Intronic

12,000 10,000 8,000 6,000 4,000 2,000 0

d Human residues (kbp)

intended to expose relative advantages and disadvantages of the alignments. 2. We compare the coverage of each alignment, which is the number of human residues aligned to each species. 3. We compare the accuracy of each alignment. To estimate accuracy, we use a statistical method called StatSigMA-w35, which identifies suspiciously aligned portions of each.

a

Intergenic

14,000 12,000 10,000 8,000 6,000 4,000 2,000 0

e 28,000 Each analysis is broken down by species and 24,000 by location with respect to annotated human 20,000 genes. This provides the most comprehensive 16,000 comparison of large-scale alignments to date 12,000 and suggests a methodology for future com8,000 parisons. Finally, we exploit the availability of 4,000 alternative alignments by demonstrating how 0 often the alignment of a region identified as suspicious can be improved by some alternative alignment. We use StatSigMA-w35 to measure the accuracy of genome-size alignments. In the past, two other approaches were used to measure accuracy. The first uses sequences constructed by simulating evolution31,34. The strength of this is that the correct alignment is known, so that alignment sensitivity and specificity can be measured accurately. The drawback is its sensitivity to assumptions in the simulation about underlying evolutionary processes, particularly genomic rearrangements, that are not well understood. The second approach measures the accuracy with which known homo logous features are aligned. For example, known orthologous exons are often used10,32,33,36, as are known repeat families10,34. Such features represent a narrow spectrum of the genome, and evaluations based on them may not extrapolate well to other genomic regions. In particular, the use of orthologous coding exons has the drawback that they are usually well conserved and most tools align them accurately. In contrast, StatSigMA-w allows direct evaluation of the accuracy at all aligned sites, rather than being limited to a small number of genomic features. There are many alignment scoring functions that measure conservation and cannot serve as measures of alignment accuracy, including

Overall

568

D o ep g ha nt C ow R ab b M it ou se R a Sh t re Te w M on nre od c el p Pl his at yp C us hi ck Xe en no Te pu tra s od on Fu Ze gu br af is h El

t illo

Ba

ad

m Ar

M

C

Ba

hi

m p bo ac on a M qu e ar m os e t G al ag o

Human residues (kbp)

© 2010 Nature America, Inc. All rights reserved.

Figure 1 Comparison of coverage of the alignments. The comparison is broken down by species and by location category; also provided is an overall chart that aggregates all four location categories. Species are displayed on the horizontal axis in order of increasing total branch length from human, according to a phylogeny estimated from fourfold-degenerate sites of third codon positions in the ENCODE regions10. The vertical axis represents the number of human residues aligned to each species given on the horizontal axis, in units of kilobase pairs (Kbp). Note that the vertical scales are different in each of the charts. The figure shows that TBA, MLAGAN and Pecan all have comparable coverage in all the placental mammals (chimp through tenrec) across all location categories. For all alignments, note that the coverage decreases approximately as species distance from human increases, particularly in the noncoding location categories.

sum of pairs, percent identity, entropy, binCons9, phastCons13, GERP8, Gumby12 and phyloP11. In a perfectly accurate alignment, where the measure of alignment accuracy should be high throughout, conservation scores will be high in regions under purifying selection and low in regions evolving neutrally or under positive selection. Conversely, in an alignment that is not perfectly accurate, there can be regions that have high conservation across nearly all sequences, with the remaining sequences misaligned (Fig. 2 and Table 1 of ref. 35). In such regions, the alignment accuracy will be low, but conservation scores will be high. These facts together suggest that any conservation score is a poor measure of alignment accuracy. RESULTS Alignment coverage Given alignment A and nonhuman species S, A’s “coverage” by S is the number of human residues aligned by A to a residue or gap in S (after removing gaps longer than 20 bp; see Online Methods). Figure 1 compares the alignment coverage for all species in the four location categories. (See also Supplementary Coverage Spreadsheet.) VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

a n a ly s i s

80

80

60

60

40

40

20

20

0

0

d

Intronic

Intergenic

100

100

80

80 60 40 20

1. “agree%,” the percentage of human residues aligned by A to S that are aligned to the same coordinate in S by some other alignment. 2. “unique%,” the percentage of human residues aligned by A to S that are not aligned to S by any other alignment. 3. “disagree%” = 100% – agree% – unique%. This is the percentage of human residues aligned by A to S that are aligned to S differently by some other alignment and not aligned to the same coordinate in S by any other alignment. Note that these are percentages of coverage, defined in the previous section. Figure 2 illustrates these comparison percentages for three ENCODE alignments. (See also Supplementary Comparison Percentage Spreadsheet.) The first observation is that there are no major differences in comparison percentages among the alignments. MLAGAN has somewhat greater unique% in the intronic and intergenic regions of nonmammals, consistent with its higher coverage in these regions. There are clear trends relating the location categories. Firstly, the intronic and intergenic categories have similar comparison percentages. If the species is kept fixed, agree% decreases and unique% increases as one moves from the coding to UTR to intronic and intergenic categories, reflecting the increased difficulty of aligning noncoding regions. nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

t illo D El og ep ha M nt ou s M Sh e on re od w el p C his hi ck Xe en no Ze pus br af is h ad

Ba

m

o

Unique-TBA Unique-MAVID Unique-MLAGAN

Ar

ag

aq

al

G

aq

al

G

ac

Level of agreement among alignments Next we investigate the level of agreement among alignments. For each alignment A and nonhuman species S, consider the following measures (defined precisely in Online Methods):

Agree-TBA Agree-MAVID Agree-MLAGAN

ue

0

ue ag o Ar B m at ad illo El Do ep g ha n M t ou se M Sh on re od w el p C his hi c Xe ken no Ze pu br s af is h

For all alignments, the coverage decreases 60 approximately as species distance from 40 human increases, particularly in noncoding location categories. Minor exceptions 20 are seen for dog, mouse, rat and mono 0 delphis. For mouse, rat and monodelphis, the explanation may be that more sequence was available to the aligners than for any other nonprimate 10. MAVID consistently has the lowest coverage in nearly all species and location categories. For distant species, MAVID often has only half the coverage of other alignments, even in coding regions. The other alignments have comparable coverage in all placental mammals (chimp through tenrec) and location categories. These observations are consistent with earlier findings10. In the intronic and intergenic regions of more distant species, MLAGAN has the highest coverage, followed in order by TBA, Pecan and MAVID. The most extreme case occurs in Xenopus intergenic regions, where MLAGAN has over four times the coverage of any other.

ac

Agree, unique, disagree (%)

c

UTR

100

M

© 2010 Nature America, Inc. All rights reserved.

b

Coding 100

M

a

Agree, unique, disagree (%)

Figure 2 Comparison percentages agree%, unique% and disagree% for TBA, MAVID and MLAGAN. (See Online Methods for the explanation of why Pecan is excluded.) The comparison is shown for 12 representative species and broken down by location category. Species are displayed on the horizontal axis in order of increasing total branch length from human. Note the trend that agree% decreases and unique% increases as the species distance from human increases and also as one moves from coding to UTR to intronic/intergenic categories.

Disagree-TBA Disagree-MAVID Disagree-MLAGAN

Fixing next the location category, there are clear trends relating the species. In all noncoding categories, as the species distance from human increases, agree% decreases and unique% increases, reflecting increased difficulty of aligning more diverged sequences 25,31. Compared to placental mammals, the more distant species have sharply decreased agree% and increased unique% in noncoding location categories. Most nonmammals have agree% < disagree% < unique% in intronic and intergenic regions, demonstrating little agreement among alignments. Because mouse is an important model organism, and because human-mouse alignments are widely used in research, the level of agreement for mouse is of particular interest. Intronic and intergenic regions account for 95% of the human sites aligned to mouse (Fig. 1). In these categories combined, agree% for mouse is disturbingly low, ranging from MAVID’s 46% to TBA’s 62% (Fig. 2). The situation is even worse in the distant species, which have much lower agree% values. Such low levels of agreement indicate that constructing a reliable wholegenome multiple sequence alignment remains a significant challenge, particularly for noncoding regions and distantly related species. Alignment accuracy Wherever alignments do not agree, which alignment, if any, is correct? This is difficult to assess because the true alignment (the one that aligns all and only orthologous residues) is inherently unknown. We use StatSigMA-w35 to estimate alignment accuracy. Given an alignment A and a nonhuman species S, StatSigMA-w identifies “suspiciously aligned regions for S,” which have at least 50 columns and statistical evidence that S is no better aligned in this region than a random sequence (see Online Methods for details). The percentage of aligned sites of species S that fall in suspicious regions for S is denoted “suspicious%.” Figure 3 compares suspicious% values of the four alignments for all species. (See also Supplementary Suspicious Percentage Spreadsheet.) When we first compare the four alignment methods for fixed species and fixed location category, MLAGAN has the highest (or nearly highest) suspicious% and Pecan the lowest (or nearly lowest) suspicious% 569

A n a ly s i s

a

Coding

35

TBA

MAVID

MLAGAN

Pecan

Suspicious%

30 25 20 15 10 5

b

0 35

UTR

30 Suspicious%

Figure 3 Comparison of accuracy of the alignments, as measured by suspicious%. The comparison is broken down by species and by location category, plus an overall chart that aggregates all four location categories. Species are displayed on the horizontal axis in order of increasing total branch length from human. For each alignment and each noncoding category, suspicious% generally increases as species distance from human increases, with a noticeable jump between the placental mammals and more distant species. Note that Pecan has the lowest or near lowest suspicious% for every species and location category.

25 20 15 10

Suspicious%

Intronic

Suspicious%

Intergenic

Overall

570

bi t ou se R a Sh t re w Te M on nre od c el p Pl his at yp C us hi ck Xe en no Te pu tra s od on Fu Ze gu br af is h

ab

R

M

nt ow C

og

D

El

ep

ha

B m at ad illo Ar

C hi Ba mp bo M o ac n a M qu ar e m os G et al ag o

Suspicious%

© 2010 Nature America, Inc. All rights reserved.

5

for every species and location category. For 0 Pecan, suspicious% is <10% in every species c 35 and category, whereas MAVID’s is as large as 30 16% (fugu intergenic), TBA’s as large as 25% 25 20 (chicken intronic) and MLAGAN’s as large as 15 33% (fugu intergenic). TBA’s and Pecan’s suspi10 cious% values are comparable for all placental 5 mammals and location categories: both have 0 suspicious% ≤2.5% in all such categories. 35 d However, TBA’s suspicious% rises precipitously 30 in noncoding regions of more distant species, 25 up to 14–25% depending on the species. 20 Turning to the trends in noncoding regions 15 as the species varies, with alignment and 10 5 location category fixed, suspicious% gener0 ally increases as species distance from human increases. There is a jump in suspicious% e 35 30 when one moves from placental mammals 25 to more distant species, which is particularly 20 noticeable in TBA and MLAGAN. These 15 trends again reflect the increased difficulty 10 of correctly aligning distant species. 5 We turn finally to the trends as location 0 category varies. As in comparison percentages, there is little difference in suspicious% values between intronic and intergenic categories. Generally, suspicious% increases as one moves from coding to UTR to intronic and intergenic categories, reflecting increased difficulty of aligning noncoding regions correctly. In coding regions, MLAGAN has greater suspicious% than the other alignments, sometimes exceeding 10%. Each of the other three has suspicious% <2.5% in the coding regions of every species. Our accuracy comparison disagrees sharply with that of Margulies et al.10 on the nonplacental mammals monodelphis and platypus. As Figure 3 illustrates, suspicious% increases in these species in the order Pecan, MAVID, TBA, MLAGAN. In intronic and intergenic regions, TBA’s suspicious% is 5 times that of Pecan in platypus and 10–12 times that of Pecan in monodelphis (Fig. 3). In contrast, in terms of Alu exclusion for these two species, Margulies et al.10 showed that TBA is best, with Pecan and MAVID close behind. In their analysis, unlike ours, monodelphis and platypus do not show patterns of alignment accuracy significantly different from those of cow, dog, armadillo, elephant, tenrec, shrew, bat and rabbit. See Supplementary Text Section 1 for further discussion. Taken together, the results of this section suggest that the accuracy of all alignments decreases in more distantly related species and in noncoding regions. Pecan appears to be most accurate overall.

Improving suspicious alignments The ENCODE alignments provide an interesting test bed for determining whether suspiciously aligned regions can be improved, also adding evidence supporting StatSigMA-w’s predictions of misalignment. (Evidence given in previous work35,37 includes poor protein BLAST E-values in suspicious coding regions and results on simulated data. Additional evidence in Supplementary Figs. 1 and 2 shows that suspicious regions are highly depleted in alignment-agreeing coordinates and enriched in alignment-unique coordinates.) As a first step towards improving suspiciously aligned regions, we plotted pairwise alignment scores of suspicious alignments versus alternative alignments. Figure 4 shows scatter plots of pairwise alignment scores for suspicious MLAGAN alignments versus nonsuspicious alternative alignments of the same human region (details in Online Methods). Scatter plots with each of the other alignments replacing MLAGAN have similar patterns (Supplementary Fig. 3). In the baboon plot, nearly every point lies above y = x, suggesting that suspiciously aligned baboon regions can be improved by an alternative alignment. In the mouse and zebrafish plots, the majority of points lie above y = x, suggesting that most suspiciously aligned regions can VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

a n a ly s i s a

b

c

be improved by an alternative alignment. Points lying below this line can be explained by two possibilities: (i) StatSigMA-w’s prediction of misalignment is incorrect, or (ii) StatSigMA-w’s prediction of misalignment is correct, but the alternative alignment is no better, either because the human sequence has no ortholog in the target species or because it is difficult to identify and align the correct ortholog. The fact that nearly all points in the baboon plot lie above the diagonal supports the latter explanation, because human sequences are more likely to have orthologs (that are not difficult to align) in baboon than in mouse or zebrafish. With either explanation for the points below y = x, it is natural to ask why StatSigMA-w does not identify the alternative alignment as suspicious as well. The explanation is that StatSigMA-w makes conservative calls of suspicious regions (details in Online Methods), which suggests that there are other misalignments besides the regions StatSigMA-w labels suspicious. Taken together, the results of this section suggest that most suspiciously aligned regions can be improved by an alternative alignment method. DISCUSSION For four multi-vertebrate alignments of the 30-Mbp human ENCODE regions, we performed three comprehensive analyses: we measured the level of agreement among alignments, we compared their coverage, and we compared their accuracy. In the first of these analyses, we found a surprisingly low level of agreement among the alignments of human noncoding regions to nonplacental mammals and more distant species. Even for mouse, an important model organism, only about half the sites aligned in one alignment agree with some other alignment. This suggests caution for users of whole-genome alignments. (Even though Pecan could not be included in this first analysis due to missing information, this analysis was not intended to compare the quality of the alignments, which is determined by the comparisons of coverage and accuracy below. The intent of this first analysis is rather to appreciate the lack of agreement that exists among alignment methods.) In a comparative assessment, the goal is to learn which method is best. To answer this, alignment coverage and accuracy must be considered together. Because we have used the suspicious% measure of StatSigMA-w35 to estimate alignment inaccuracy, the ideal alignment is one with high coverage (Fig. 1) and low suspicious% (Fig. 3). Figure 5 summarizes suspicious% versus coverage, but only for the aggregation of all location categories. MAVID has the lowest coverage for nearly all species and location categories. The other alignments nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

have comparable coverages in all species and location categories (with one exception, discussed below). Pecan has the lowest or nearly lowest suspicious% for all species and location categories, less than 10% in each of these 22 × 4 categories. Taken together, these results suggest that the Pecan alignment is the best among the four ENCODE alignments. Given the number of dimensions for alignment comparison, it is surprising that any one alignment appears best in nearly every category. The exception to comparable coverage for TBA, MLAGAN and Pecan occurs in intronic and intergenic categories of nonplacental species (monodelphis, platypus, chicken, Xenopus, tetraodon, fugu, and zebrafish), where MLAGAN has the highest coverage, TBA the next highest and Pecan the lowest. Among these seven species and two location categories, MLAGAN has up to 4 times the coverage of TBA (averaging 1.9 times the coverage over all 14 categories), and TBA has up to 2 times the coverage of Pecan (averaging 1.5 times the coverage over all 14 categories). However, in each of these 14 categories, MLAGAN’s suspicious% is greater than TBA’s, which is greater than Pecan’s, and in most categories these differences are great. This suggests that the additional coverage in these categories may not be worth the decreased accuracy. For example, averaged over these 14 categories, TBA’s suspicious% is 4 times Pecan’s, whereas TBA’s coverage is only 1.5 times Pecan’s. 30 25

Suspicious%

© 2010 Nature America, Inc. All rights reserved.

Alternative alignment score

Baboon Mouse Zebrafish Figure 4 Pairwise alignment scores of suspicious regions versus those for alternative 1.0 1.0 1.0 alignments of the same human region. For three representative species S (baboon, mouse 0.5 0.5 0.5 and zebrafish) and one representative target 0 alignment (MLAGAN), scatter plots show all 0 0 points (x′, y′), where x′ is the pairwise human-S −0.5 alignment score of an MLAGAN alignment region that is suspicious for species S, and y ′ −0.5 −0.5 −1.0 is the pairwise human-S alignment score of −1.5 one of the other three alignments for the same −1.0 −1.0 −1.5 −1.0 −0.5 0 0.5 1.0 −1.0 −0.5 0 0.5 1.0 −1.0 −0.5 0 0.5 1.0 human region that is not suspicious for S. MLAGAN alignment score MLAGAN alignment score MLAGAN alignment score (See Online Methods for the scoring function and Supplementary Fig. 3 for other target alignments.) Alignment scores are normalized by alignment length. The dashed black diagonal line has equation y = x. The solid blue line has equation y – x = μ, where μ is the mean value of y′ – x′ for all points (x′, y′) in the plot. The dotted blue lines have equations y – x = μ ± ó, where ó is the standard deviation of y′ – x′ for all points (x′, y′) in the plot. Note that most points lie above the line y = x, suggesting that most of the suspiciously aligned regions can be improved by one of the alternative alignments.

20 15 10 5 0 12

13 Primates

14

15 ln(coverage)

16

Other placental mammals TBA

MAVID

MLAGAN

17

18

Distant species Pecan

Figure 5 Summary plot of suspicious% vs. coverage, aggregated over all four location categories. The horizontal axis is on a logarithmic scale. For a given species, points that are lower and farther right represent better performance. Note the comparable performance of Pecan and TBA on placental mammals (orange and green triangles) and the superior accuracy of Pecan on the distant species (orange circles).

571

© 2010 Nature America, Inc. All rights reserved.

A n a ly s i s For placental mammals, TBA’s coverage and suspicious% are comparable to Pecan’s in every species and location category. In this realm, TBA and Pecan emerge together as best. Focusing finally on coding regions, all alignments except MLAGAN seem very accurate, with suspicious% <2.5% for every species. MAVID’s lower coverage suggests that TBA and Pecan are best in coding regions. TBA’s overall suspicious% values in Figure 3 are consistent with those reported in ref. 35 for the 17-vertebrate MULTIZ alignment of human chromosome 1. In particular, both demonstrate the same precipitous rise in suspicious% as one moves from placental mammals to more distant species. For the 14 nonprimates present in both alignments, the overall TBA ENCODE suspicious% values range from 0.4 to 1.16 times those of the MULTIZ whole-chromosome alignment, depending on the species, with an average ratio over all 14 species of 0.7. One reason why ENCODE suspicious% values may be less than those of the MULTIZ whole-chromosome alignment is that each ENCODE region is so much shorter than chromosome 1, and the orthologous sequences for each individual ENCODE region were prepared and supplied to the aligners. In conclusion, we provide the most comprehensive comparison of large-scale alignment methods to date, and we propose a methodo logy for future comparisons of whole-genome multiple alignments. These comparisons provide critical accuracy feedback to alignment tool designers. Our assessment reveals that constructing accurate whole-genome multiple alignments remains challenging, particularly for noncoding regions and distant species. Users should exercise caution when assuming alignment correctness in these situations. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Note: Supplementary information is available on the Nature Biotechnology website. Acknowledgments We thank P. Green, W. Noble, W.L. Ruzzo and especially A. Prakash for helpful discussions and technical advice. We thank the US National Institutes of Health and the Natural Sciences and Engineering Research Council of Canada for financial support. AUTHOR CONTRIBUTIONS X.C., design, implementation, experimentation, analysis; M.T., design, analysis. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Kent, W. et al. The human genome browser at UCSC. Genome Res. 12, 996–1006 (2002). 2. Woolfe, A. et al. Highly conserved non-coding sequences are associated with vertebrate development. PLoS Biol. 3, e7 (2005). 3. Xie, X. et al. Systematic discovery of regulatory motifs in conserved regions of the human genome, including thousands of CTCF insulator sites. Proc. Natl. Acad. Sci. USA 104, 7145–7150 (2007).

572

4. Gross, S.S. & Brent, M.R. Using multiple alignments to improve gene prediction. J. Comput. Biol. 13, 379–393 (2006). 5. Siepel, A. et al. Targeted discovery of novel human exons by comparative genomics. Genome Res. 17, 1763–1773 (2007). 6. Pedersen, J.S. et al. Identification and classification of conserved RNA secondary structures in the human genome. PLOS Comput. Biol. 2, e33 (2006). 7. Washietl, S., Hofacker, I.L., Lukasser, M., Hüttenhofer, A. & Stadler, P.F. Mapping of conserved RNA secondary structures predicts thousands of functional noncoding RNAs in the human genome. Nat. Biotechnol. 23, 1383–1390 (2005). 8. Cooper, G.M. et al. Distribution and intensity of constraint in mammalian genomic sequence. Genome Res. 15, 901–913 (2005). 9. Margulies, E. et al. Identification and characterization of multi-species conserved sequences. Genome Res. 13, 2507–2518 (2003). 10. Margulies, E.H. et al. Analyses of deep mammalian sequence alignments and constraint predictions for 1% of the human genome. Genome Res. 17, 760–774 (2007). 11. Pollard, K.S., Hubisz, M.J., Rosenbloom, K.R. & Siepel, A. Detection of nonneutral substitution rates on mammalian phylogenies. Genome Res. 20, 110–121 (2010). 12. Prabhakar, S. et al. Close sequence comparisons are sufficient to identify human cis-regulatory elements. Genome Res. 16, 855–863 (2006). 13. Siepel, A. et al. Evolutionarily conserved elements in vertebrate, insect, worm, and yeast genomes. Genome Res. 15, 1034–1050 (2005). 14. Felsenstein, J. Inferring Phylogenies (Sinauer Associates, 2004). 15. Wong, K.M., Suchard, M.A. & Huelsenbeck, J.P. Alignment uncertainty and genomic analysis. Science 319, 473–476 (2008). 16. Siepel, A. & Haussler, D. Phylogenetic estimation of context-dependent substitution rates by maximum likelihood. Mol. Biol. Evol. 21, 468–488 (2004). 17. Murphy, W.J. et al. Resolution of the early placental mammal radiation using Bayesian phylogenetics. Science 294, 2348–2351 (2001). 18. Nikolaev, S. et al. Early history of mammals is elucidated with the ENCODE multiple species sequencing data. PLoS Genet. 3, e2 (2007). 19. Bird, C.P. et al. Fast-evolving noncoding sequences in the human genome. Genome Biol. 8, R118 (2007). 20. Kim, S. & Pritchard, J. Adaptive evolution of conserved non-coding elements in mammals. PLoS Genet. 3, e147 (2007). 21. Nielsen, R. et al. A scan for positively selected genes in the genomes of humans and chimpanzees. PLoS Biol. 3, e170 (2005). 22. Pollard, K.S. et al. An RNA gene expressed during cortical development evolved rapidly in humans. Nature 443, 167–172 (2006). 23. Prabhakar, S., Noonan, J.P., Pääbo, S. & Rubin, E.M. Accelerated evolution of conserved noncoding sequences in humans. Science 314, 786 (2006). 24. Dewey, C.N., Huggins, P.M., Woods, K., Sturmfels, B. & Pachter, L. Parametric alignment of Drosophila genomes. PLOS Comput. Biol. 2, e73 (2006). 25. Blanchette, M. Computation and analysis of genomic multi-sequence alignments. Annu. Rev. Genomics Hum. Genet. 8, 193–213 (2007). 26. Kumar, S. & Filipski, A. Multiple sequence alignment: in pursuit of homologous DNA positions. Genome Res. 17, 127–135 (2007). 27. Lunter, G. et al. Uncertainty in homology inferences: assessing and improving genomic sequence alignment. Genome Res. 18, 298–309 (2008). 28. Margulies, E.H. Confidence in comparative genomics. Genome Res. 18, 199–200 (2008). 29. Margulies, E.H. & Birney, E. Approaches to comparative sequence analysis: towards a functional view of vertebrate genomes. Nat. Rev. Genet. 9, 303–313 (2008). 30. Rokas, A. Lining up to avoid bias. Science 319, 416–417 (2008). 31. Blanchette, M. et al. Aligning multiple genomic sequences with the threaded blockset aligner. Genome Res. 14, 708–715 (2004). 32. Bray, N. & Pachter, L. MAVID: constrained ancestral alignment of multiple sequences. Genome Res. 14, 693–699 (2004). 33. Brudno, M. et al. LAGAN and Multi-LAGAN: efficient tools for large-scale multiple alignment of genomic DNA. Genome Res. 13, 721–731 (2003). 34. Paten, B., Herrero, J., Beal, K., Fitzgerald, S. & Birney, E. Enredo and Pecan: genome-wide mammalian consistency-based multiple alignment with paralogs. Genome Res. 18, 1814–1828 (2008). 35. Prakash, A. & Tompa, M. Measuring the accuracy of genome-size multiple alignments. Genome Biol. 8, R124 (2007). 36. Dubchak, I., Poliakov, A., Kislyuk, A. & Brudno, M. Multiple whole-genome alignments without a reference organism. Genome Res. 19, 682–689 (2009). 37. Prakash, A. & Tompa, M. Assessing the discordance of multiple sequence alignments. IEEE/ACM Trans. Comput. Biol. Bioinformatics 6, 542–551 (2009).

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

ONLINE METHODS Alignments. Many computational tools are available for computing multiple sequence alignments. The reason so many diverse tools exist is that computing the optimal multiple sequence alignment is inherently an intractable computational problem38. This is true even for relatively short alignments consisting of hundreds or thousands of columns, such as alignments of proteins or genomic promoter regions. The problem becomes even harder when one wants to compute a whole-genome multiple sequence alignment, which may consist of millions or billions of alignment columns and, in addition, must contend with the complication of arbitrary genome rearrangements such as translocations, duplications, inversions and so on29. Each of the four whole-genome alignment programs studied here is actually integrated with other programs to form a pipeline for building the alignments10. For convenience, we use the names TBA, MAVID, MLAGAN and Pecan throughout to represent their respective pipelines. Although the ENCODE alignments contain sequence from 28 vertebrate genomes, we omitted from all results the alignments of five species, colobus monkey, dusky titi, owl monkey, mouse lemur and hedgehog, because for each of them less than 3.5 Mbp of sequence was available for alignment to the 30-Mbp human ENCODE sequence. For other mammals at least 17 Mbp of sequence was available10. This left six primates (human, chimpanzee, baboon, macaque, marmoset and galago), ten other placental mammals (bat, armadillo, dog, elephant, cow, rabbit, mouse, rat, shrew and tenrec), two nonplacental mammals (monodelphis and platypus) and five nonmammals (chicken, Xenopus, tetraodon, fugu and zebrafish) for all the analyses. For the TBA, MAVID and MLAGAN alignments, any column containing the gap character in human was removed. This was done for consistency with the Pecan alignment, which contains no gaps in human. Multiple sequence alignments based on the September 2005 sequence freeze of the ENCODE Multi-Species Sequence Analysis10 were downloaded from http://hgdownload.cse.ucsc.edu/goldenPath/hg17/encode/alignments/SEP2005/alignments/ for TBA, MAVID and MLAGAN and from http://www. ebi.ac.uk/~bjp/pecan/encode_sept_pecan_mfas_proj.tar.bz2 for Pecan. The phylogeny with branch lengths that was input to StatSigMA-w and used to determine the species order in all figures was downloaded from http:// hgdownload.cse.ucsc.edu/goldenPath/hg17/encode/alignments/SEP-2005/ phylo/tree_4d.tba.v2.nh. Preprocessing long gaps. There are great differences in gap length distribution among the alignment programs: MAVID, MLAGAN and Pecan tend to use very long gaps, whereas TBA prefers to simply omit the species from that portion of the alignment rather than assign it a very long gap. Supplementary Figure 4 shows the length distribution of gaps for these four alignments in ENm003, a representative ENCODE region. In the TBA alignment, most of the gaps have length 1–5 bp, and there is no gap longer than 200 bp. The other alignments, in contrast, have a much greater fraction of gaps longer than 50 bp. The aligner’s decision of whether to omit a species S from a given region or assign a long gap to S in that region is arbitrary, and this arbitrary decision could have affected our comparisons. For example, in our measurement of level of agreement among the alignments, it could make the difference between a human coordinate h being labeled disagree (if species S has a long gap at h in some other alignment) or unique (if species S is omitted at h in the other alignments). More importantly, it would affect the accuracy assessment, as StatSigMA-w treats gaps very differently from the way that it treats absent species35. Therefore, to put the alignments on equal footing for comparison, we preprocessed all the alignments to remove gaps longer than 20 bp, as though the species containing such a long gap is simply absent from that alignment region. For example, if a human subsequence was aligned to a gap of length 30 bp in mouse, we treated this human subsequence as unaligned in mouse after the

doi:10.1038/nbt.1637

removal of long gaps, though of course still possibly aligned to other species. The threshold of 20 bp was chosen so as to make the gap length distributions of the alignments much more comparable (see Supplementary Fig. 4). Genomic location categories. Our comparisons are all broken down into the four distinct location categories of coding, UTR, intronic and intergenic. Human sites are categorized according to annotated UCSC Known Genes downloaded from the UCSC Genome Browser (assembly July 2007). Because two known genes may overlap (for example, because of alternative splicing), a single human site may fall into more than one category. In order to assign exactly one category to each human site, we use the following priority order for the four location categories, listed from highest to lowest priority: coding, UTR, intronic, intergenic. For example, if a site is contained in a coding exon in one isoform and in an intron in another, that site will be categorized as coding. Comparison percentages. The ENCODE alignments are human-centric and represented in the coordinates of the human sequence. We therefore use the human sequence as our reference when measuring the level of agreement of the alignments. We compare, for each coordinate h in the human sequence and each nonhuman species S, the coordinate in S (if any) that is aligned to h by the alignments. By “coordinate” we mean the chromosome (or scaffold) name together with the position within that chromosome (or scaffold). Given an alignment A (for example, TBA), a nonhuman species S (for example, mouse), and a human coordinate h, when A aligns human coordinate h to coordinate s in S, there are three possible cases: 1. If there is at least one other alignment that also aligns h to s, we say that A “agrees.” 2. If A is the only alignment that aligns human coordinate h to the target species S, we say that A is “unique.” 3. If there is some other alignment that aligns human coordinate h to something in S, but the aligned coordinate in S is not s for any other alignment, we say that A “disagrees.” Another case that must be considered is when A aligns human coordinate h to a gap “–” in S. If two alignments both align h to a gap, we do not simply conclude that these two alignments agree on h. Instead, we take into consideration the contexts of the aligned gaps. Suppose, for example, that both TBA and MAVID align a human coordinate h to a gap in mouse. Let the first aligned mouse coordinate to the left and right of the gap be mL and mR, respectively, for TBA, and m′L and m′R, respectively, for MAVID. For the human coordinate h, TBA and MAVID will be considered to agree if and only if mL = m′L and mR = m′R. Otherwise, the two alignments will be considered to disagree, because TBA’s gap and MAVID’s gap are actually inserted between different pairs of mouse coordinates. For a given target species S, the absolute number of human coordinates for which alignment A agrees, disagrees or is unique is significantly influenced by the coverage of S in A, as defined in the section on “Alignment coverage.” Instead, we calculate the percentage of the human coordinates in each of those three categories among all human coordinates aligned to S by A. These comparison percentages are denoted “agree%,” “unique%” and “disagree%,” respectively. Note that (i) the sum of these three percentages is 100% for any fixed alignment A and species S; (ii) the comparison percentages may differ for varying alignments A; and (iii) the definition of agree% only requires the target alignment to agree with one other alignment. All the ENCODE alignments except Pecan’s provide coordinates for the aligned species. Because this information is absent from the Pecan alignment, we must omit Pecan from the agreement comparisons. Because our goal is to measure the extent to which different large-scale genomic alignments agree and disagree with each other, the trends are clear enough even without Pecan.

nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

It is worth noting that the comparison labels assigned to different alignments for the same human coordinate are not independent. For example, if a particular human coordinate is labeled agree for one alignment, this coordinate must be labeled the same way for at least one of the other alignments. StatSigMA-w. Given any multiple sequence alignment and a phylogeny of the aligned sequences, StatSigMA-w35 assesses the accuracy of the alignment and identifies suspiciously aligned regions. It is based on a statistical model that generalizes the Karlin-Altschul theory39 from pairwise to multiple sequence alignment. More specifically, StatSigMA-w assigns a “discordance score” to every site of the alignment and identifies a set of worst-aligned species for that site. (See ref. 35 for details. StatSigMA-w actually identifies a branch of the phylogeny whose removal would separate the species aligned at that site into two subsets that may be misaligned to each other, depending on the value of the discordance score; any species separated from human by this branch is referred to as a ‘worstaligned species’.) The discordance score, much like a P value, ranges between 0 and 1 and measures how likely it is that a worst-aligned species at that site is misaligned to the human sequence, with higher score indicating greater likelihood of misalignment. In practice, discordance scores show a bimodal behavior, with nearly all alignment columns having score either >0.1 or <10−4 (see Fig. 1 of ref. 35). This bimodality makes discordance values somewhat more intuitive, classifying alignment columns neatly into those that appear well aligned (score <10−4) and those that suggest poor alignment (score >0.1). For each nonhuman species S, StatSigMA-w next identifies “suspiciously aligned regions,” which are regions of the alignment (i) that have at least 50 sites, (ii) in which all sites have discordance score at least 0.1, with S being a worst-aligned species at each site, and (iii) that do not contain too many gaps. (See ref. 35 for details.) Given the bimodality described above, if the threshold of 0.1 were changed to 10−2 or 10−4, the suspicious-region predictions would hardly change. The threshold of 50 sites focuses attention on those moderate to long regions where S appears to be misaligned. Using the phylogeny generated for the ENCODE regions10, we ran StatSigMA-w on all four ENCODE alignments. The identified suspicious regions are available as UCSC Genome Browser custom tracks, for all four alignments and all 22 species, at http://bio.cs.washington.edu/encode-msa/. As a summary figure, the percentage of aligned sites of species S that fall in StatSigMA-w suspicious regions for S is denoted “suspicious%.” We use the suspicious% values of each species to compare the accuracy of the four ENCODE alignments.

nature biotechnology

Comparing suspicious and alternative alignments. This section describes the procedure that was used to create the scatter plots of Figure 4 and Supplementary Figure 3. Given a target alignment (say, MLAGAN), a target species (say, baboon), and an alternative alignment (say, Pecan), we performed the following analysis for each region of the MLAGAN alignment that StatSigMA-w identified as suspiciously aligned for baboon. Let h be the human genomic region in this suspicious alignment. If Pecan, the alternative alignment, does not align h to some sequence in baboon, or if this region overlaps a suspicious region for baboon in the Pecan alignment, discard h and go on to the next suspicious region. Otherwise, let AM and AP be the human-baboon alignments of MLAGAN and Pecan, respectively, for the human region h, and let BM and BP be the baboon sequences aligned to h by MLAGAN and Pecan, respectively. If either of BM or BP is a substring of the other, discard h and go on to the next suspicious region. (This is a proxy for MLAGAN and Pecan agreeing on part of their alignment, because Pecan does not supply nonhuman genomic coordinates.) At this point we are left with a suspicious MLAGAN human-baboon alignment AM and a nonsuspicious Pecan human-baboon alignment AP that do not agree. We then compute pairwise alignment scores SM and SP, respectively, of these two alignments using the following BLASTN scoring function: for mouse and zebrafish, the scoring function is +1 for match and −1 for mismatch or gap; for baboon, the mismatch score is −2 to reflect the smaller divergence between human and baboon40. Add a length-normalized point (SM/LM, SP/LP) to the scatter plot of Figure 4, where Li is the length of the alignment Ai. Repeat the procedure with TBA and MAVID replacing Pecan as the alternative alignment. To ensure that the differences between the baboon scatter plots and those of mouse and zebrafish are not due to differences in the alignment scoring function, we also created scatter plots for baboon using the same scoring function as those used for mouse and zebrafish. This had little noticeable effect on the patterns of the scatter plots (Supplementary Fig. 3). 38. Wang, L. & Jiang, T. On the complexity of multiple sequence alignment. J. Comput. Biol. 1, 337–348 (1994). 39. Karlin, S. & Altschul, S.F. Applications and statistics for multiple high-scoring segments in molecular sequences. Proc. Natl. Acad. Sci. USA 90, 5873–5877 (1993). 40. States, D.J., Gish, W. & Altschul, S.F. Improved sensitivity in nucleic acid database searches using application-specific scoring matrices. Methods: A Companion to Methods in Enzymology 3, 66–70 (1991).

doi:10.1038/nbt.1637

review

Rationalizing the development of live attenuated virus vaccines © 2010 Nature America, Inc. All rights reserved.

Adam S Lauring1,4, Jeremy O Jones2,4 & Raul Andino3 The design of vaccines against viral disease has evolved considerably over the past 50 years. Live attenuated viruses (LAVs)— those created by passaging a virus in cultured cells—have proven to be an effective means for preventing many viral diseases, including smallpox, polio, measles, mumps and yellow fever. Even so, empirical attenuation is unreliable in some cases and LAVs pose several safety issues. Although inactivated viruses and subunit vaccines alleviate many of these concerns, they have in general been less efficacious than their LAV counterparts. Advances in molecular virology—creating deleterious gene mutations, altering replication fidelity, deoptimizing codons and exerting control by microRNAs or zinc finger nucleases—are providing new ways of controlling viral replication and virulence and renewing interest in LAV vaccines. Whereas these rationally attenuated viruses may lead to a new generation of safer, more widely applicable LAV vaccines, each approach requires further testing before progression to human testing.

The basic goal of vaccination is to stimulate protective immunity while avoiding disease from the vaccine itself. The first generation of viral vaccines relied on empirical attenuation by repeated passage in cultured cells. Several LAVs meet both criteria for vaccines; they elicit a strong and protective immune response with a low risk of disease from the vaccine itself. Despite recent successes in the development of LAVs for rotavirus and several arboviruses, the classical attenuation process is somewhat unpredictable and has not always been applicable. In the present regulatory environment, the use of LAVs has also been limited by safety concerns, including reversion to wild-type virulence. Because LAVs are shed from vaccinees, they sometimes present a risk to unvaccinated individuals with impaired immunity. These safety concerns have led to a shift toward the use of inactivated viruses or viral subunits as vaccines. Despite notable successes like the inactivated poliovirus vaccine1, inactivated viruses are generally less immunogenic than their LAV counterparts, and this strategy is limited to viruses for which there are good culture and production systems. Subunit vaccines, which use viral proteins as immunogens, have become a major focus of vaccine development and have led to several successfully licensed vaccines, including vaccines against hepatitis B virus, influenza viruses and papillomaviruses2. Production is more easily controlled and efficient than that of LAVs or inactivated viruses. Even so, this strategy has not achieved universal success, as many subunit vaccines have failed to elicit a protective immune response in the host. Although adjuvants have increased the immunogenicity of subunit vaccines, newer methods of subunit delivery mimic a natural immune

response by incorporating more viral components. There are several approaches to this end, including liposome delivery of antigens3,4, viruslike particles5 and virosomes, which are reconstituted viral envelopes lacking any viral genetic material6. Another approach to increase the immunogenicity of subunit vaccines is to recombinantly encode a pathogenic antigen in a nonpathogenic, yet infectious poxvirus or adenovirus vector7,8. Although there have been some notable successes, the major concern with this strategy is that the vector vaccines will not induce adequate immunological responses in hosts who have preexisting antibodies against the vector. As vaccines become more complex and ‘virus-like,’ unsurprisingly, live, attenuated vaccines have received a second look. Advances in molecular virology and the advent of recombinant-virus systems have led to the identification of many viral genes associated with virulence and immunogenicity. Researchers have used this information to better control the replication and pathogenesis of vaccine candidates, thereby avoiding the unpredictability of empirical attenuation. Here we first review attempts to use mutation or deletion of replication genes to create attenuated virus. We then discuss the application of four new methods—altered replication fidelity, codon deoptimization, and control by microRNAs (miRNAs) or zinc finger nucleases (ZFNs)—to rationally design vaccines. These novel LAV designs each allow limited viral replication and antigen production, and because the host immune response is not required to limit viral spread, such LAVs may be safer than classic LAVs, even in immunocompromised patients.

1Department

Attenuation through deletion or mutation The identification of genes essential for viral replication and assembly led to the first generation of rationally designed, live-virus vaccines (Table 1). Deletion or mutation of these genes results in a ‘defective virus,’ which cannot replicate in the host (for an excellent review of defective-virus vaccines, see Dudek and Knipe8). These defective

of Medicine, University of California, San Francisco, California, USA. of Cellular and Molecular Pharmacology, University of California, San Francisco, California, USA. 3Department of Microbiology and Immunology, University of California, San Francisco, California, USA. 4These authors contributed equally to this work. Correspondence should be addressed to R.A. ([email protected]). 2Department

Published online 7 June 2010; doi:10.1038/nbt.1635

nature biotechnology volume 28 number 6 JUNE 2010

573

revie w

© 2010 Nature America, Inc. All rights reserved.

Table 1 Current vaccine strategies Vaccine approach

Construction

Empirically attenuated virus

Blind passage in different cell types. By adapting to a new environ- Host immunity is able to limit the virulence and spread of the ment, the virus accumulates mutations that mediate attenuation attenuated virus

Inactivated virus

Virus is inactivated by chemical treatment (for example, formaldehyde)

Disruption of viral proteins and/or genetic material

Subunit vaccine

Recombinant expression of one or several viral proteins

No viral genetic material is included

Viral vectors

One or several genes from a virus are inserted into the genome of a second nonpathogenic virus (the vector). Viral particles produced by the vector transduce these genes into target cells and direct their expression

The vector itself is attenuated (see above) but is able to express antigens derived from the pathogenic virus

Replication-defective viruses

One or several genes required for genome replication are deleted in the vaccine strain. The virus vaccine is produced in a helper cell line that expresses the missing protein(s) in trans

The administered virus is unable to replicate its genome

Single-cycle viruses

One or several genes required for viral assembly and spread are deleted in the vaccine strain. Distinguished from replicationdefective viruses by their competence for genome replication

The virus is able to replicate its genome but is defective for assembly or spread

viruses are propagated in ‘helper’ cells that express the missing gene(s). Although the virus is unable to replicate its genome, viral genes are still expressed, which can induce a strong immune response in the inoculated host. ‘Single-cycle viruses,’ which are defective in a viral protein required for assembly or spread, are a variation on this theme. Although these viruses can replicate their genome through a single cycle, they do not produce infectious virus9. The first example of a replication-defective virus used as a vaccine was a herpes simplex virus-1 (HSV-1) strain with a deletion of a gene essential for genome replication10. This virus stimulated an immune response similar to natural infection and protected against wild-type virus challenge in a mouse model of infection11. Replication-defective HSV-2 strains, which lack genes essential for viral DNA synthesis (UL5) and viral replication (UL29), have also been described12. These viruses were more effective than subunit vaccines in eliciting protective immunity in mice13 and did not establish latency14, an important consideration in herpesviruses. HSV-1 and HSV-2 strains have also been created that lack glycoprotein H and are unable to spread from cell to cell or produce infectious progeny. These single-cycle viruses protect against wild-type challenge in rodent models15,16, but the block in viral spread may be leaky. Similar strategies are now being applied to viruses other than HSV. The newer smallpox vaccines are replication-defective viruses8, and an influenza nuclear export protein (formerly referred to as the NS-2) knockout and hemagglutinin cleavage site mutants have been shown to provide protective immunity in mice17,18. Likewise, flaviviruses with a deletion in the core nucleocapsid protein C function are single-cycle viruses as they cannot spread between cells or encapsidate virus. These viruses can elicit a potent immune response and protect against wild-type challenge19. Even with progress in the attenuation of viruses by deleterious gene mutation, this approach has not led to a safe and effective vaccine for human disease. On the one hand, this can be attributed to the relatively short time this field has been in existence; on the other, vaccines based on deleterious gene mutation also often evoke only a weak immune response because the antigen is only expressed at the site of inoculation. There are also safety concerns about the completeness of the block in viral spread in single-cycle viruses19. As with conventional LAVs, it has proven very difficult to balance immunogenicity with safety, even with the rational design of replication-defective viruses. Riboviral replication fidelity—failure then success Although LAV vaccines have been developed for many RNA viruses, the mutability of these pathogens presents unique challenges for vaccine 574

Safety

design. The RNA-dependent RNA polymerases of RNA viruses exhibit characteristically low fidelity, with measured mutation rates of 10−3 to 10−5 mutations per nucleotide copied per replication cycle20. These mutation rates are orders of magnitude greater than those of nearly all DNA-based viruses and organisms. Because the genomes of RNA viruses typically comprise <10,000 nucleotides (nt), this mutation rate translates to roughly 0.1–10 mutations per genome replicated. Work in our laboratory (R.A. and colleagues21) estimates that each viral replication cycle generates every possible point mutation and many double mutations, which may be present within the population at any time. This impressive diversity has important biological implications. First, lowfrequency variants within the population may contain, or quickly acquire, mutations in key epitopes, which mediate escape from vaccine-elicited neutralizing antibody or cytotoxic T cells22. Antigenic drift within the hemagglutinin and neuraminidase proteins of influenza virus is the best example of this process and the primary reason for annual reevaluation of vaccine strains23. Second, many RNA viruses, including HIV and the hepatitis C virus, exhibit such marked intra- and interindividual genetic diversity that it has been difficult to identify stable, conserved epitopes that provide universal protection against all strains22. Finally, the mutability of RNA viruses has triggered real concerns about the potential reversion of live, attenuated vaccines to pathogenic strains. Both mutation and recombination probably have a role in this process, and the sporadic emergence of vaccine-derived polioviruses is a cautionary tale24. A large body of work in recent years suggests that because of their mutation rates, the evolution of RNA viruses may differ fundamentally from that of DNA-based organisms. Much of this work builds on the mathematical framework of quasi-species theory and seeks to understand the importance of genetic diversity at the population level25. According to quasi-species theory, the mutation rates of RNA viruses place them near a critical ‘error threshold.’ Below this threshold, the mutant spectrum within the population favors adaptability, and lowfitness variants are tolerated so long as the majority remains viable. Beyond the error threshold, too many mutations accumulate, genetic information is lost and the population becomes inviable26. Indeed, mutagenic nucleosides increase viral mutation rate and cause population collapse, thus providing an effective treatment for several RNA viruses. Even so, several groups have identified mutants in poliovirus and foot-and-mouth disease virus that were resistant to nucleoside analogs27–29. Further studies in our laboratory (R.A. and colleagues30) have revealed that these variants replicate with higher fidelity by virtue of mutations within the viral RNA-dependent RNA polymerase. As a volume 28 number 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

revie w result, drug-resistant mutants give rise to populations with appreciably described in many viral genomes. In bacteriophage, codon usage closely less genetic diversity than the wild type30. Significantly, this decrease in mirrors that of the host37. The bias is more pronounced in the highly diversity was responsible for attenuation in a transgenic mouse model expressed structural genes, suggesting optimization for translational of infection30,31. efficiency38,39. Most mammalian viruses also have a strong preference 32 On the basis of these observations, our group (R.A. and colleagues ) against CpG dinucleotides, although their overall GC content is highly proposed that the attenuation of these high-fidelity variants could be variable38. Studies of HIV and influenza suggest that codons in highly exploited for vaccine design. We focused on glycine 64 of the poliovirus variable surface proteins may be optimized for their volatility, the probpolymerase, which regulates fidelity through a complex hydrogen-bond ability that a codon will mutate to a different amino acid class40,41. This network and mediates sensitivity to nucleoside analogs27. Of the 19 would presumably facilitate immune escape and suggests that there has possible amino acid substitutions at this position, only 13 gave rise to been selection for genetic plasticity in these highly mutable viruses. Recent studies of poliovirus have addressed the importance of codon viable virus and 8 of these were unstable. The other 5 mutants had lower mutation rates than wild type and were less adaptable in cell culture. In bias for viral replication and pathogenesis. Burns and colleagues perthe transgenic mouse model, these high-fidelity variants were mark- formed large-scale mutagenesis of the Sabin 2 vaccine strain, replacing edly attenuated and shed less efficiently than wild type. Three of the up to 50% of the capsid codons with synonymous codons that are less viruses stimulated high titers of neutralizing antibody in infected mice, preferred in the human genome42. Although these codon-deoptimized an order of magnitude greater than the Sabin 1 vaccine strain. They also viruses exhibit minimal defects in viral gene expression, they produce induced long-lasting immunity. Mice vaccinated with G64S, G64A or fewer infectious progeny and overall fitness is markedly reduced. A G64L survived a lethal challenge of wild-type virus at 1 or 6 months via subsequent study found that synonymous changes that increase the frethe intraperitoneal or intramuscular route. quency of CpG and UpA dinucleotides have similar effects on viral fitThis work suggests that controlling replication fidelity is a promis- ness43. Mueller and colleagues44 took a similar approach but used gene ing approach for engineering live, attenuated vaccines. Even so, several synthesis technology to design poliovirus genomes with completely important questions remain. Although it is clear that such a strategy deoptimized codons in the capsid region. They also found a marked could be successful for other picornaviruses, which have structurally reduction in replicative fitness and a reduction in infectious progeny. conserved polymerases, it may be difficult to identify the relevant resi- Even so, their data imply that codon-deoptimized viruses have reduced dues in other viral RNA-dependent RNA polymerases. In these cases, translational efficiency compared with wild type. They obtained simiselection for nucleoside analog resistance may be an unbiased way of lar results with viruses in which synonymous changes are determined discovering promising mutants for further characterization. Reversion by codon pair bias. In both cases they found that codon-deoptimized to wild type is another potential problem, because viruses contain- polioviruses are attenuated by 1,000-fold on a per-particle basis coming the lower fidelity wild-type polymerase appear to have a selective pared with wild type. advantage. Although high-fidelity variants would certainly revert at a lower frequency, their mutation rate is still considerably higher than that of DNA viruses20,30. We found no evidence for reversion of the G64S mutation Pre-miRNA No virus Drosha after either 20 passages in HeLa cells or five Viral RNA with miRNA target sequences Dicer produced mouse-to-mouse passages over 25 days32. miRNA gene Although these results are encouraging, furmiRNA ther experiments along these lines will probRISC ably be required before regulatory approval. Finally, the mouse model for poliovirus pathoCleaved RISC binds to target sequence viral genesis is an imperfect one, and the level of RNA attenuation observed here may not reflect the situation in human vaccinees. Nevertheless, the high-fidelity variants could still be useful in the ongoing polio eradication campaign, as safer seed strains will be needed for large-scale production of the inactivated polio vaccine in Not expressed a postpolio world33. Viral production, Attenuation by substitution It is well known that many organisms exhibit a codon bias, using some synonymous codons or codon pairs more frequently than others34. In bacteria and simple eukaryotes, codon preference is related to amounts of the corresponding transfer RNA and affects translational efficiency35,36. The reasons for the observed codon bias are less clear in mammals. Because viruses rely on the host-cell machinery for nearly all aspects of replication, it is not surprising that codon bias has been

immune response

miRNA gene

Viral replication

Figure 1 The microRNA (miRNA)-virus vaccine strategy. Viral replication can be regulated in a tissuespecific manner by incorporating miRNA target sites into the viral genome. In cells that express the miRNA (e.g., brain, top cell), the miRNAs are processed and transported to the cytoplasm, where they mediate cleavage of viral RNA. Viral replication is restricted to cells in which the miRNA is not expressed (e.g., intestine, bottom cell). The engineered virus can therefore trigger a natural immune response in target tissues without the associated risk of dissemination and disease.

nature biotechnology volume 28 number 6 JUNE 2010

575

© 2010 Nature America, Inc. All rights reserved.

revie w Because all changes are synonymous, the proteins expressed from codon-deoptimized viruses are identical to wild type and similarly immunogenic. Mueller and colleagues45, therefore, proposed that their marked attenuation would make them ideal live vaccines. In their second study, they show that deoptimized viruses provoke a robust neutralizing antibody response following three weekly intraperitoneal inoculations. All immunized mice survive subsequent lethal challenge with wild-type poliovirus, demonstrating the vaccine efficacy of the engineered viruses. As a general strategy for vaccine development, codon deoptimization offers several advantages. First, attenuation does not affect antigenicity, and the immune response should closely mimic a natural infection. Second, because attenuation is systematic and not empirical, it may be easily applied to other viruses. Finally, codon-deoptimized viruses encode hundreds of point mutations, each with a fairly small individual effect on fitness. Consequently, there is little risk of reversion to virulence with even a handful of point mutations. Both Mueller et al.42 and Burns et al.44 found that codon-deoptimized viruses are genetically stable and remain attenuated after repeated passage. The marked sequence divergence of such engineered viruses from circulating strains may also reduce the frequency of recombination and the risk of pathogenic, vaccine-derived variants. Much work remains to be done before codon-deoptimized viruses are employed as live, attenuated vaccines. Although the results among the studies are consistent, the mechanism of attenuation is still debated. This would certainly be an issue for regulatory bodies, and the lack of clarity makes it difficult to determine whether codon-based attenuation is a unique aspect of picornaviruses, or a more generalizable approach to vaccine design. As in the case of the high-fidelity variants, the mouse model may not be the best system for assessing vaccine efficacy and safety. Nevertheless, codon deoptimization is a promising approach that has already generated significant interest in the virology community. miRNA-controlled LAVs The discovery of RNA interference (RNAi) just a decade ago has resulted in an explosion of research into this novel form of gene regulation. The two main effectors of RNAi are small interfering RNA (siRNA) and miRNA46. Although there has been intense interest in using siRNAs to combat mammalian RNA viruses47, miRNAs are now also being used to limit viral pathogenesis. miRNAs are genomically encoded and have a major role in endogenous gene regulation48. They are transcribed as long precursor pri-miRNAs, which are processed by the nuclear RNase Drosha to ~60-nt hairpin intermediates, which are then transported to the cytoplasm where they are trimmed by Dicer to roughly ~22 nt (Fig. 1). Similar to siRNAs, mature miRNAs are loaded into the RNA-induced silencing complex, where they mediate either degradation or translational repression of target messages. The human genome encodes over 400 miRNAs, many of which have tissue-specific or developmental expression patterns. Several DNA viruses also express miRNAs49. These virally derived miRNAs modulate pathogenesis and host immunity through regulation of viral and cellular transcripts, respectively. The diversity and complexity of cellular miRNAs means that many cell types will have a unique miRNA profile50. Several investigators have taken advantage of this property to better target viral gene therapy vectors51. Silencing of specific transcripts or the entire genome can be accomplished by inclusion of miRNA binding sites in the vector sequence. In many cases the miRNA system is used to provide a second level of control beyond receptor expression or tissue-specific promoter activity. For example, Brown and colleagues52 eliminated off-target expression from a hepatocyte-specific promoter in antigen-presenting 576

cells by incorporating miR-142-3p binding sites in their lentiviral construct52. In a related study, muscle-specific miRNA binding sites were used to limit secondary replication of a coxsackievirus in a murine tumor model53. Improved targeting of adenoviral vectors has also been achieved by the addition of miRNA binding sites to the 3′ untranslated region of the E1A transcript54,55. In LAV design, empirical attenuation of viruses is often accomplished by changing the tissue tropism of a virus through repeated passage in a new cell type. We (R.A. and colleagues)56 proposed that the same result could be achieved through miRNA restriction of poliovirus replication. Although poliovirus replicates in many tissues, disease onset is linked to lytic infection of the central nervous system (CNS). By incorporating binding sites for either let7a (a ubiquitous miRNA) or miR124 (a CNS-restricted miRNA) into the RNA genome of wildtype poliovirus (Fig. 1), we showed that viral replication is restricted in a cell type–dependent manner and that the effect is dependent on the cellular RNAi machinery. The miRNA-targeted viruses are largely restricted from the CNS in a murine model of infection and markedly attenuated as a result56. Experiments with viruses containing mutant target sequences confirm that the altered tropism is due to miRNA. The degree of attenuation exceeds 5 orders of magnitude, and neither let7anor miR124-targeted viruses are pathogenic in immunocompromised mice lacking the α/β-interferon receptor56. Both viruses are able to replicate in non-neuronal tissues and stimulate a strong neutralizing antibody response after a single intraperitoneal inoculation. The level of protection is impressive, as even the interferon receptor–knockout mice are protected from subsequent challenge with 10,000 times the lethal dose of wild-type virus56. Although miRNA targeting is a promising approach to rational design of LAVs, the study has several caveats worth mentioning. The let7a virus replicates poorly in most tissues, whereas the mir124 virus is restricted only in the CNS56. As a result, the former stimulates a weaker immune response and is a less effective vaccine. On the other hand, widespread replication of the mir124 virus in non-neuronal tissues could allow the virus to accumulate mutations within the miRNA target sequence and thereby escape degradation. Indeed, several mice in the study had low titers of mir124 virus in the spinal cord, and sequence analysis showed mutations within the miRNA target sequences56. Work from our laboratory suggests that a single let7a site can accumulate escape mutations in as little as 24–48 h32. The risk of miRNA escape could be minimized by the inclusion of several target sequences for the same miRNA or different miRNAs with the same tissue distribution. Another way of minimizing escape is highlighted in another article on species-specific restriction of influenza virus for vaccine production57. In this study Perez and colleagues57 incorporated nonavian miRNA target sequences into a region of the viral nucleoprotein open reading frame. Because the miRNA target sequence also serves as codons for conserved amino acids, escape mutations would alter protein structure and probably have a deleterious effect on viral replication. We expect that as the RNAi field matures, investigators will find other ways of controlling the replication and mutability of miRNA-targeted vaccines, although the potential for reversion to wild type will have to be mitigated to the satisfaction of regulatory bodies. Zinc-finger nuclease-controlled LAVs Zinc-finger (ZF) domains mediate nucleotide-specific binding of proteins to DNA, a property that defines a large family of DNA-binding proteins58. Each finger makes contact with a separate DNA triplet, and natural or recombinant ZFs have been created that can recognize almost any triplet59. The modular nature of the ZFs allows them to be joined in useful combinations. Typically, three ZFs are combined to bind to volume 28 number 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

revie w a specific 9–base pair (bp) DNA sequence, and these ZFs have been stages: immediate-early (before most of viral protein synthesis), early coupled to various functional domains to create artificial transcription (before viral replication) and late (after viral replication begins)67. Other factors that can activate or repress gene transcription with remark- DNA viruses for which ZFNs would be useful are similarly regulated. able promoter specificity60. ZFs have also been fused to the nuclease There is also the potential to encode ZFNs behind inducible promoters, domain of the restriction enzyme FokI to cleave double-stranded DNA so that ZFN expression would commence upon the administration of at specific sequences61. The nuclease domain must dimerize to cleave a small molecule68. Nuclease activity can also be controlled directly DNA, and because the dimer interface is weak, two nuclease domains by addition of small molecule–sensitive residues to the ZFN69. These are typically brought into close proximity by pairs of ZFs binding to strategies would provide an ideal way to optimize the balance between neighboring 9-bp sites, spaced 6 bp apart (Fig. 2)62,63. In this configura- ZFN-virus replication and nuclease activity. The ability to create a ZFN vaccine that can prevent and eliminate tion the engineered ZFN recognizes a specific 18-bp sequence, which is long enough, by a few orders of magnitude, to be unique in the human persistent viral infections is a long way from being realized. As with genome. Because of this specificity, this same technology could be used any LAV, safety issues are always a concern. The ZFN vaccine approach would probably be limited to nonintegrating, DNA viruses, as ranto distinguish between human and virus DNA. Several groups have used recombinant ZF proteins to control aspects dom breaks in host chromosomal DNA caused by ZFN cleavage of of the viral life cycle. ZF proteins fused to the ZF protein 10 gene integrated viral DNA could be catastrophic. There are many nonin(ZNF10/KOX1) repression domain have been created that target the tegrating human viruses, includes the herpesviruses, polyomaviruses, HSV-1–infected cell polypeptide 4 (ICP4) promoter64. These proteins adenoviruses and papillomaviruses, that establish a persistent infection bind the promoter with nanomolar affinity, with one able to markedly and provide particularly difficult challenges for the treatment of their repress viral protein 16 (VP16)–activated transcription in vitro. This respective diseases. ZFN-based vaccines may offer a way to prevent or ZF-ZNF10 fusion, when delivered in trans into HSV-1–infected cells, eliminate these hard-to-treat latent infections. Reversion to wild type is was able to limit HSV-1 replication and reduce viral titer by 90%. In a another concern, but the risk can be reduced by including ZFNs against similar strategy, recombinant ZF proteins have been designed to recog- multiple, essential viral sequences to ensure that the intrinsic mutation nize the human papillomavirus 18 (HPV-18) replication origin65. When rate of the virus will not allow the mutation of every ZFN target site. It is expressed in vitro, these ZFs were able to compete with the replication also possible that DNA cleaved by ZFNs could be repaired via homoloprotein E2 for binding to viral DNA. This competitive antagonism led to gous recombination using uncleaved viral genomes. However, if the reduced HPV replication in transient replication assays in mammalian sequence is repaired accurately it would be subject to repeated cleavage; cells. By fusing the origin-targeted ZF protein to a nuclease domain, this ZFN was able to cleave viral DNA and reduce viral replication in cultured cells66. These experiments demonstrate that ZFNs can effectively target and eliminate viral DNA in mammalian cells. It may be feasible to deliver a therapeutic Nine base pairs Linker Nine base pairs virus-specific ZFN in trans to eradicate latent viral DNA. Delivery of the ZFNs to all latently infected cells is, however, technically challengCircular viral ing. Alternatively, virus-specific ZFNs could episome be delivered using the viral genome itself and serve as a vaccine. In the ZFN-vaccine strategy, ZFNs targeting sequences for viral Packaging replication and other essential viral processes signal would be introduced into the viral genome Origin of replication (Fig. 2). Following inoculation, immunoCleavage to genic viral genes and virus-specific ZFNs linear DNA Essential would be expressed. While the viral proteins latency sequence would stimulate a natural immune response, the ZFNs would cleave viral DNA, and limit replication. ZFN-LAVs have potential both as prophylactic vaccines, protecting against wildExpression of type challenge, as well as therapeutic vaccines, immunogenic Essential delivering ZFNs to cells already harboring viral proteins replication latent viral DNA. gene The immunogenicity of ZFN vaccines can be controlled by temporal and spatial regulation of ZF expression to balance viral pro- Figure 2 Zinc-finger nuclease (ZFN)-virus vaccine strategy. The ZFN is composed of two arrays of three tein expression with the ability of the ZFNs ZF domains fused to a DNA nuclease domain (blue lightning bolt). The nuclease must dimerize to be to eliminate all replication-competent viral activated, so each ZFN array is designed to bind the adjacent 9-bp sequences in the virus genome, spaced 5–6 bp apart, allowing the nuclease domains to dimerize and cleave the viral double-stranded DNA. This could best be accomplished using DNA. ZFNs can be designed to target multiple, essential viral sequences. By encoding the ZFNs in the promoters that are temporally controlled by viral genome itself and temporally controlling the expression of the ZFNs using viral promoters, the virus the virus itself. For instance, herpesvirus gene can express immunogenic proteins before ZFN cleavage of circular episomal DNA to linear DNA, which transcription occurs in at least three distinct is incapable of replication and establishment of latency. nature biotechnology volume 28 number 6 JUNE 2010

577

revie w

© 2010 Nature America, Inc. All rights reserved.

Table 2 Approaches to viral attenuation for vaccine design Vaccine approach

Advantages

Disadvantages

Empirically attenuated virus

Excellent immunogenicity, few doses required

Limited applicability, reversion to wild type, Measles, mumps, rubella (MMR); oral breakthrough disease poliovirus vaccine, influenza, rotavirus, yellow fever, varicella

Examples

Subunit vaccine

Widely applicable, very safe

Poor immunogenicity, multiple doses usually required

Hepatitis B virus, human papillomavirus

Viral vectors

Good immunogenicity, delivery of multiple antigens

Neutralizing antibodies to vector, possible safety issues

Many examples (experimental)

Defective viruses

Good immunogenicity, known mechanism of attenuation

Limited to inoculation site, possible safety issues

HSV-1, HSV-2, influenza (experimental)

Replication fidelity

Strong immunogenicity, known mechanism of attenuation, not susceptible to antigenic shift/drift

RNA viruses only, possible reversion to wild type

Poliovirus (experimental)

Codon deoptimization

Strong immunogenicity, no reversion to wild Possible safety concerns type, possibly applicable to many viruses

Poliovirus (experimental)

miRNA-controlled virus

Strong immunogenicity, known mechanism Limited to some RNA viruses of attenuation, prevent latent infection

Poliovirus, adenovirus, coxsackievirus, influenza (experimental)

ZFN-controlled virus

Strong immunogenicity, known mechanism Limited to nonintegrating DNA viruses of attenuation, prevent latent infection

HSV-1, human papilloma virus (experimental)

HSV-1, HSV-2, herpes simplex virus-1 and -2; miRNA, microRNA; ZFN, zinc-finger nuclease.

if it is repaired inaccurately the virus should be nonviable because of mutation of an essential sequence. Ideally, we will arrive at a live virus strain that will have limited replication, not establish latency and elicit a protective immune response. In essence, we would turn an otherwise detrimental latent infection into an asymptomatic, acute infection. Conclusions LAV vaccines have provided ideal protection from several major diseases but have not lived up to their potential as a result of limited applicability and safety concerns. Advances in molecular biology have opened the door to novel approaches to viral attenuation and may lead to a new generation of safer LAVs (Table 2). Although replication-defective LAVs have encountered some problems, this approach to attenuation is on the cusp of providing safe, effective vaccines for several diseases. Several other approaches to attenuation are poised to overcome other problems specifically associated with vaccine design for RNA and DNA viruses. For many RNA viruses wherein high mutation rates limit the efficacy of vaccines, altering the replication fidelity can attenuate the entire virus population, leading to population collapse without mutation of key immunogenic epitopes. Codon deoptimization provides a systematic means by which to attenuate any virus. Substituting synonymous codons throughout a viral genome avoids loss of immunogenicity and confers little risk of reversion to wild type. ZFNs and miRNAs can be used to control the replication of DNA and RNA viruses, respectively. Controlling viral replication temporally or spatially can cause a strong, natural immune response to be elicited before the virus is eliminated. These may be particularly useful approaches for designing vaccines against persistent or latent viruses, as ZFNs and miRNAs lead to the elimination of all viral DNA or RNA, thus preventing chronic infection. Each of these approaches is aimed to address long-standing problems with LAV vaccine design. Although they could potentially change the way we think about attenuation, significant hurdles lie ahead. Live vaccines present an inherent trade-off between safety and efficacy, and regulatory bodies are right to be concerned about viral escape or reversion to wild type. The studies described here have largely been carried out in murine models with relatively short-term measures of immunogenicity and limited characterization of viral genetic stability. Much more work is needed in relevant animal models before an initial dosing and safety trial in humans can be contemplated. We expect that each strategy will have to be modified to optimize its safety and 578

efficacy profile. Nevertheless, the efficacy demonstrated by available LAVs, particularly the recent success in developing safe and effective live, attenuated rotavirus, influenza and varicella zoster vaccines, is a strong incentive to redouble efforts to improve the safety characteristics of this type of vaccine. Rational attenuation may also facilitate the development of inactivated vaccines for high-risk agents by providing safer seed stocks for large-scale production. The next several years will clearly be an exciting time in vaccine research as advances in molecular biology are further translated into preventive strategies for viral disease. ACKNOWLEDGMENTS This work was supported by grants from the National Institute of Allergy and Infectious Diseases to R.A. (R01 AI36178 and R01 AI40085) and A.S.L. (K08 AI081754-01). COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/reprintsandpermissions/. 1. Salk, J. Polio vaccines and polioviruses. BMJ 2, 765 (1977). 2. Arvin, A.M. & Greenberg, H.B. New viral vaccines. Virology 344, 240–249 (2006). 3. Rogan, D. & Babiuk, L.A. Novel vaccines from biotechnology. Rev. Sci. Tech. 24, 159–174 (2005). 4. Kersten, G.F. & Crommelin, D.J. Liposomes and ISCOMs. Vaccine 21, 915–920 (2003). 5. Jennings, G.T. & Bachmann, M.F. The coming of age of virus-like particle vaccines. Biol. Chem. 389, 521–536 (2008). 6. Felnerova, D., Viret, J.F., Gluck, R. & Moser, C. Liposomes and virosomes as delivery systems for antigens, nucleic acids and drugs. Curr. Opin. Biotechnol. 15, 518–529 (2004). 7. Panicali, D. & Paoletti, E. Construction of poxviruses as cloning vectors: insertion of the thymidine kinase gene from herpes simplex virus into the DNA of infectious vaccinia virus. Proc. Natl. Acad. Sci. USA 79, 4927–4931 (1982). 8. Dudek, T. & Knipe, D.M. Replication-defective viruses as vaccines and vaccine vectors. Virology 344, 230–239 (2006). 9. Loudon, P.T. et al. Preclinical safety testing of DISC-hGMCSF to support phase I clinical trials in cancer patients. J. Gene Med. 3, 458–467 (2001). 10. Nguyen, L.H., Knipe, D.M. & Finberg, R.W. Replication-defective mutants of herpes simplex virus (HSV) induce cellular immunity and protect against lethal HSV infection. J. Virol. 66, 7067–7072 (1992). 11. Morrison, L.A. & Knipe, D.M. Mechanisms of immunization with a replication-defective mutant of herpes simplex virus 1. Virology 220, 402–413 (1996). 12. Da Costa, X., Kramer, M.F., Zhu, J., Brockman, M.A. & Knipe, D.M. Construction, phenotypic analysis, and immunogenicity of a UL5/UL29 double deletion mutant of herpes simplex virus 2. J. Virol. 74, 7963–7971 (2000). 13. Hoshino, Y. et al. Comparative efficacy and immunogenicity of replication-defective, recombinant glycoprotein, and DNA vaccines for herpes simplex virus 2 infections in mice and guinea pigs. J. Virol. 79, 410–418 (2005).

volume 28 number 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

revie w 14. Da Costa, X.J., Jones, C.A. & Knipe, D.M. Immunization against genital herpes with a vaccine virus that has defects in productive and latent infection. Proc. Natl. Acad. Sci. USA 96, 6994–6998 (1999). 15. McLean, C.S. et al. Protective vaccination against primary and recurrent disease caused by herpes simplex virus (HSV) type 2 using a genetically disabled HSV-1. J. Infect. Dis. 170, 1100–1109 (1994). 16. Farrell, H.E. et al. Vaccine potential of a herpes simplex virus type 1 mutant with an essential glycoprotein deleted. J. Virol. 68, 927–932 (1994). 17. Watanabe, T., Watanabe, S., Neumann, G., Kida, H. & Kawaoka, Y. Immunogenicity and protective efficacy of replication-incompetent influenza virus-like particles. J. Virol. 76, 767–773 (2002). 18. Stech, J. Attenuated influenza A viruses with modified cleavage sites in hemagglutinin as live vaccines. Expert Rev. Vaccines 7, 739–743 (2008). 19. Widman, D.G., Frolov, I. & Mason, P.W. Third-generation flavivirus vaccines based on single-cycle, encapsidation-defective viruses. Adv. Virus Res. 72, 77–126 (2008). 20. Holland, J. et al. Rapid evolution of RNA genomes. Science 215, 1577–1585 (1982). 21. Vignuzzi, M., Stone, J.K. & Andino, R. Ribavirin and lethal mutagenesis of poliovirus: molecular mechanisms, resistance and biological implications. Virus Res. 107, 173–181 (2005). 22. Walker, B.D. & Burton, D. Toward an AIDS vaccine. Science 320, 760–764 (2008). 23. Couch, R.B. Seasonal inactivated influenza virus vaccines. Vaccine 26 Suppl 4, D5–D9 (2008). 24. Minor, P. Vaccine-derived poliovirus (VDPV): impact on poliomyelitis eradication. Vaccine 27, 2649–2652 (2009). 25. Domingo, E. et al. Viruses as quasispecies: biological implications. Curr. Top. Microbiol. Immunol. 299, 51–82 (2006). 26. Biebricher, C.K. & Eigen, M. The error threshold. Virus Res. 107, 117–127 (2005). 27. Arnold, J.J., Vignuzzi, M., Stone, J.K., Andino, R. & Cameron, C.E. Remote site control of an active site fidelity checkpoint in a viral RNA-dependent RNA polymerase. J. Biol. Chem. 280, 25706–25716 (2005). 28. Pfeiffer, J.K. & Kirkegaard, K. A single mutation in poliovirus RNA-dependent RNA polymerase confers resistance to mutagenic nucleotide analogs via increased fidelity. Proc. Natl. Acad. Sci. USA 100, 7289–7294 (2003). 29. Sierra, M. et al. Foot-and-mouth disease virus mutant with decreased sensitivity to ribavirin: implications for error catastrophe. J. Virol. 81, 2012–2024 (2007). 30. Vignuzzi, M., Stone, J.K., Arnold, J.J., Cameron, C.E. & Andino, R. Quasispecies diversity determines pathogenesis through cooperative interactions in a viral population. Nature 439, 344–348 (2006). 31. Pfeiffer, J.K. & Kirkegaard, K. Increased fidelity reduces poliovirus fitness and virulence under selective pressure in mice. PLoS Pathog. 1, e11 (2005). 32. Vignuzzi, M., Wendt, E. & Andino, R. Engineering attenuated virus vaccines by controlling replication fidelity. Nat. Med. 14, 154–161 (2008). 33. Chumakov, K. & Ehrenfeld, E. New generation of inactivated poliovirus vaccines for universal immunization after eradication of poliomyelitis. Clin. Infect. Dis. 47, 1587–1592 (2008). 34. Gustafsson, C., Govindarajan, S. & Minshull, J. Codon bias and heterologous protein expression. Trends Biotechnol. 22, 346–353 (2004). 35. Grantham, R., Gautier, C., Gouy, M., Jacobzone, M. & Mercier, R. Codon catalog usage is a genome strategy modulated for gene expressivity. Nucleic Acids Res. 9, r43–r74 (1981). 36. Sharp, P.M., Tuohy, T.M. & Mosurski, K.R. Codon usage in yeast: cluster analysis clearly differentiates highly and lowly expressed genes. Nucleic Acids Res. 14, 5125– 5143 (1986). 37. Carbone, A. Codon bias is a major factor explaining phage evolution in translationally biased hosts. J. Mol. Evol. 66, 210–223 (2008). 38. Jenkins, G.M. & Holmes, E.C. The extent of codon usage bias in human RNA viruses and its evolutionary origin. Virus Res. 92, 1–7 (2003). 39. Shackelton, L.A., Parrish, C.R. & Holmes, E.C. Evolutionary basis of codon usage and nucleotide composition bias in vertebrate DNA viruses. J. Mol. Evol. 62, 551–563 (2006). 40. Plotkin, J.B. & Dushoff, J. Codon bias and frequency-dependent selection on the hemagglutinin epitopes of influenza A virus. Proc. Natl. Acad. Sci. USA 100, 7152– 7157 (2003). 41. Stephens, C.R. & Waelbroeck, H. Codon bias and mutability in HIV sequences. J. Mol. Evol. 48, 390–397 (1999). 42. Burns, C.C. et al. Modulation of poliovirus replicative fitness in HeLa cells by deoptimization of synonymous codon usage in the capsid region. J. Virol. 80, 3259–3272 (2006).

nature biotechnology volume 28 number 6 JUNE 2010

43. Burns, C.C. et al. Genetic inactivation of poliovirus infectivity by increasing the frequencies of CpG and UpA dinucleotides within and across synonymous capsid region codons. J. Virol. 83, 9957–9969 (2009). 44. Mueller, S., Papamichail, D., Coleman, J.R., Skiena, S. & Wimmer, E. Reduction of the rate of poliovirus protein synthesis through large-scale codon deoptimization causes attenuation of viral virulence by lowering specific infectivity. J. Virol. 80, 9687–9696 (2006). 45. Coleman, J.R. et al. Virus attenuation by genome-scale changes in codon pair bias. Science 320, 1784–1787 (2008). 46. Carthew, R.W. & Sontheimer, E.J. Origins and mechanisms of miRNAs and siRNAs. Cell 136, 642–655 (2009). 47. Haasnoot, J., Westerhout, E.M. & Berkhout, B. RNA interference against viruses: strike and counterstrike. Nat. Biotechnol. 25, 1435–1443 (2007). 48. Bartel, D.P. MicroRNAs: target recognition and regulatory functions. Cell 136, 215– 233 (2009). 49. Gottwein, E. & Cullen, B.R. Viral and cellular microRNAs as determinants of viral pathogenesis and immunity. Cell Host Microbe 3, 375–387 (2008). 50. Lagos-Quintana, M. et al. Identification of tissue-specific microRNAs from mouse. Curr. Biol. 12, 735–739 (2002). 51. Brown, B.D. & Naldini, L. Exploiting and antagonizing microRNA regulation for therapeutic and experimental applications. Nat. Rev. Genet. 10, 578–585 (2009). 52. Brown, B.D., Venneri, M.A., Zingale, A., Sergi Sergi, L. & Naldini, L. Endogenous microRNA regulation suppresses transgene expression in hematopoietic lineages and enables stable gene transfer. Nat. Med. 12, 585–591 (2006). 53. Kelly, E.J., Hadac, E.M., Greiner, S. & Russell, S.J. Engineering microRNA responsiveness to decrease virus pathogenicity. Nat. Med. 14, 1278–1283 (2008). 54. Cawood, R. et al. Use of tissue-specific microRNA to control pathology of wild-type adenovirus without attenuation of its ability to kill cancer cells. PLoS Pathog. 5, e1000440 (2009). 55. Ylösmäki, E. et al. Generation of a conditionally replicating adenovirus based on targeted destruction of E1A mRNA by a cell type-specific microRNA. J. Virol. 82, 11009–11015 (2008). 56. Barnes, D., Kunitomi, M., Vignuzzi, M., Saksela, K. & Andino, R. Harnessing endogenous miRNAs to control virus tissue tropism as a strategy for developing attenuated virus vaccines. Cell Host Microbe 4, 239–248 (2008). 57. Perez, J.T. et al. MicroRNA-mediated species-specific attenuation of influenza A virus. Nat. Biotechnol. 27, 572–576 (2009). 58. Iuchi, S. Three classes of C2H2 zinc finger proteins. Cell. Mol. Life Sci. 58, 625–635 (2001). 59. Wright, D.A. et al. Standardized reagents and protocols for engineering zinc finger nucleases by modular assembly. Nat. Protoc. 1, 1637–1652 (2006). 60. Dhanasekaran, M., Negi, S. & Sugiura, Y. Designer zinc finger proteins: tools for creating artificial DNA-binding functional proteins. Acc. Chem. Res. 39, 45–52 (2006). 61. Porteus, M. Design and testing of zinc finger nucleases for use in mammalian cells. Methods Mol. Biol. 435, 47–61 (2008). 62. Smith, J. et al. Requirements for double-strand cleavage by chimeric restriction enzymes with zinc finger DNA-recognition domains. Nucleic Acids Res. 28, 3361– 3369 (2000). 63. Mani, M., Smith, J., Kandavelou, K., Berg, J.M. & Chandrasegaran, S. Binding of two zinc finger nuclease monomers to two specific sites is required for effective double-strand DNA cleavage. Biochem. Biophys. Res. Commun. 334, 1191–1197 (2005). 64. Papworth, M. et al. Inhibition of herpes simplex virus 1 gene expression by designer zinc-finger transcription factors. Proc. Natl. Acad. Sci. USA 100, 1621–1626 (2003). 65. Mino, T. et al. Inhibition of DNA replication of human papillomavirus by artificial zinc finger proteins. J. Virol. 80, 5405–5412 (2006). 66. Mino, T., Mori, T., Aoyama, Y. & Sera, T. Inhibition of human papillomavirus replication by using artificial zinc-finger nucleases. Nucleic Acids Symp. Ser. (Oxf) 52, 185–186 (2008). 67. Fields, B.N., Knipe, D.M., Howley, P.M. & Griffin, D.E. Fields Virology 4th edn (Lippincott Williams & Wilkins, Philadelphia, 2001). 68. Zhu, Z., Zheng, T., Lee, C.G., Homer, R.J. & Elias, J.A. Tetracycline-controlled transcriptional regulation systems: advances and application in transgenic animal modeling. Semin. Cell Dev. Biol. 13, 121–128 (2002). 69. Pruett-Miller, S.M., Reading, D.W., Porter, S.N. & Porteus, M.H. Attenuation of zinc finger nuclease toxicity by small-molecule regulation of protein levels. PLoS Genet. 5, e1000376 (2009).

579

B r i e f c o m m u n i c at i o n s

Synthetic polymer coatings for long-term growth of human embryonic stem cells

© 2010 Nature America, Inc. All rights reserved.

Luis G Villa-Diaz1,2,12, Himabindu Nandivada3,12, Jun Ding1, Naiara C Nogueira-de-Souza1,4, Paul H Krebsbach2,5, K Sue O’Shea6, Joerg Lahann3,5,7,8 & Gary D Smith1,9–11 We report a fully defined synthetic polymer coating, poly[2(methacryloyloxy)ethyl dimethyl-(3-sulfopropyl)ammonium hydroxide] (PMEDSAH), which sustains long-term human embryonic stem (hES) cell growth in several different culture media, including commercially available defined media. The development of a standardized, controllable and sustainable culture matrix for hES cells is an essential step in elucidating mechanisms that control hES cell behavior and in optimizing conditions for biomedical applications of hES cells. Considerable progress has been made in the development of defined hES cell media1,2; however, long-term culture still requires use of recombinant extracellular matrix proteins3 or animal-derived matrices4, which are sources of variability5 and xenogeneic contamination6. A compositionally defined matrix that supports hES cell expansion in defined media is critical for determining factors that regulate stem cell growth and differentiation, expanding the use of hES cells in biotechnologies and enabling potential clinical applications. Thus far, synthetic polymer matrices have sustained only short-term hES cell propagation7–10. Here, we synthesized six polymer coatings by surface-initiated graft polymerization and tested their ability to promote attachment and proliferation of undifferentiated hES cells (Fig. 1a and Supplementary Methods). First, we examined growth using mouse embryonic fibroblast (MEF)-conditioned media (CM). H9 hES cells were mechanically harvested from cultures grown on MEFs and placed onto dishes coated with polymer or Matrigel. Matrigel supported adhesion and colony formation of >90% of hES cell aggregates, but no attachment was observed on poly[carboxybetaine methacrylate]. hES cells adhered, but spontaneously differentiated, during the first two passages on poly[[2-(methacryloyloxy)ethyl]trimethylammonium chloride], poly[3-sulfopropyl methacrylate], poly[2-hydroxyethyl methacrylate],

and poly[poly(ethylene glycol) methyl ether methacrylate] (Fig. 1a). However, PMEDSAH (Fig. 1b for polymer characterization) supported attachment and proliferation of two hES cell lines (BG01 and H9), long-term growth, as well as normal genetic, proteomic and differentiation potential. Throughout 25 passages, H9 hES cells seeded on PMEDSAH expressed characteristic hES cell markers, displayed a normal karyotype and retained pluripotency (Fig. 1c–f and Supplementary Fig. 1). At passage 20, expression levels of the hES cell markers OCT3/4 (91 ± 3%) and SOX2 (92 ± 2%) were similar to levels expressed by cells cultured on Matrigel (Fig. 1c). Microarray analysis revealed similar expression levels of hES cell–specific genes in cells grown on PMEDSAH or Matrigel (Supplementary Table 1). Further validation using qPCR revealed similar RNA expression levels of NANOG, OCT3/4 (also known as POU5F1) and SOX2 (Fig. 1d; primers listed in Supplementary Table 2). The pluripotency of H9 hES cells was confirmed at passage 25 by tri-lineage differentiation in teratomas (Fig. 1f). Taken together, these results demonstrate the ability of PMEDSAH to support long-term culture of undifferentiated, pluripotent hES cells. Next, we grew BG01 and H9 cells on PMEDSAH-coated dishes in the presence of a xeno-free (free of nonhuman animal products) commercially available human cell–conditioned medium (Supplementary Methods). Throughout 15 passages, BG01 and H9 cells showed similar cell population–doubling times (38 ± 6 h and 37 ± 4 h, respectively) that compared well to values reported for Matrigel11, expressed hES cell markers, retained normal karyotypes and remained pluripotent (Supplementary Fig. 2). On PMEDSAH, a significantly higher H9 hES cell–aggregate adhesion was observed for all passages using human cell–conditioned medium (86 ± 6%) compared to MEF-conditioned media (15 ± 1%; Fig. 2a). Cellaggregate adhesion was also significantly higher for H9 cells in human cell–conditioned medium than for BG01 cells cultured under the same conditions (47 ± 5%; Supplementary Fig. 2), suggesting that there may be important biological differences between cell lines in their expression of adhesion receptors. Finally, we examined the ability of PMEDSAH to support hES cell cultures in two serum-free defined media1,2,4,12 (StemPro and mTeSR; Supplementary Methods). Cells grown in mTeSR could not be passaged, but StemPro medium was able to support ten passages of H9 hES cells (Fig. 2). BG01 hES cells could not be passaged beyond three passages in StemPro (Supplementary Fig. 3). Throughout ten passages of H9 cells in StemPro medium, cell population–doubling times, expression of hES cell markers and normal karyotypes were confirmed

1Department

of Obstetrics and Gynecology, University of Michigan, Ann Arbor, Michigan, USA. 2Department of Biologic and Materials Sciences, University of Michigan, Ann Arbor, Michigan, USA. 3Department of Chemical Engineering, University of Michigan, Ann Arbor, Michigan, USA. 4Human Biology Research Laboratory, Barretos Cancer Hospital, Sao Paulo, Brazil. 5Department of Biomedical Engineering, University of Michigan, Ann Arbor, Michigan, USA. 6Department of Cell and Developmental Biology, University of Michigan, Ann Arbor, Michigan, USA. 7Department of Materials Science and Engineering, University of Michigan, Ann Arbor, Michigan, USA. 8Department of Macromolecular Science and Engineering, University of Michigan, Ann Arbor, Michigan, USA. 9Department of Urology, University of Michigan, Ann Arbor, Michigan, USA. 10Department of Molecular and Integrated Physiology, University of Michigan, Ann Arbor, Michigan, USA. 11Reproductive Sciences Program, University of Michigan, Ann Arbor, Michigan, USA. 12These authors contributed equally to this work. Correspondence should be addressed to J.L. (materials: [email protected]) and G.D.S. (cell culture: [email protected]). Received 14 September 2009; accepted 5 April 2010; published online 30 May 2010; doi:10.1038/nbt.1631

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

581

a

UV-ozone activation

PCBMA

PPEGMA PCBMA

PPEGMA

PHEMA

PSPMA

PMETAC

PMEDSAH

63.3 ± 3.1

56.0 ± 3.7

50.2 ± 4.1

40.5 ± 5.7

17.1 ± 1.1

Reduced elastic modulus (GPa)

3.3 ± 0.1

3.9 ± 0.1

3.0 ± 0.5

3.0 ± 0.1

3.3 ± 0.1

2.5 ± 0.1

Cell-aggregate adhesion (%)

0

5±1

12 ± 1

14 ± 2

8±1

15 ± 1

Number of passages

0

1

2

2

2

25

0.01

5.6

Oxygen (O 1s)

532

20.9

27.8

Sulfur (S 2p)

168

3.2

5.6

d

95 90 85

e

1.5 1.0 0.5 0

3

4

9

5

10

11

16

17

22

X

NANOG

OCT3/4

SOX2

f

12 18

Endoderm

Ectoderm

Mesoderm

Y

to animal-derived matrices such as Matrigel, PMEDSAH is chemically defined, can be synthesized reproducibly and has long-term stability. Unlike natural and recombinant matrices, PMEDSAH-coated dishes can be handled and stored with relative ease. This matrix therefore SSEA-4

TRA-1-60

TRA-1-81

Phase

c

RNA +

0 Defined medium (StemPro)

+

–

+

–

+

–

-c

–

rin

SOX2

he

OCT3/4

ad

2

Matrigel PMEDSAH

8

P2

P0

3

80

2.0

tin

hCCM

3.2

ac

MEF-CM

72.7

402

β-

0

61.1

285

T1

10

Percent theoretical

Carbon (C 1s)

Matrigel PMEDSAH

SOX2

Percent experimental

Relative transcript level

OCT3/4

XPS Peak position

Nitrogen (N 1s)

0 3,800 3,300 2,800 2,300 1,800 1,300 800 –1 Wavenumbers (cm ) 100

Element

P0 3 P2 0 P0 P23 0

c

0.02

0 P0 3 P2 0

Absorbance

b

0.03

d

VE

20

PMEDSAH

KR

20

PMETAC

A4

30

40

PSPMA

G AT

60

b Cell doubling time (h)

80

Cell adhesion Cell doubling P = 0.0001 50 P = 0.0001 P = 0.031 40

PHEMA

71.6 ± 4.8

(Fig. 2a,b). Moreover, H9 cells maintained the 6 7 8 ability to differentiate into endoderm, meso13 14 15 derm and ectoderm (Fig. 2c,d). The ability of PMEDSAH to support hES 19 20 21 cell culture in defined media suggest the utility of this coating for commercial hES cell expansion. Additional studies will be required to determine contribution of the various physicochemical properties of the polymer, such as wettability, mechanical stiffness, surface topography and zeta potential. Compared 100

heat

Contact angle

1

a

Monomer solution

OOH OOH OOH

Ozone-activated TCPS dish

Percent of positive cells

Figure 1 Long-term culture of H9 hES cells on PMEDSAH with MEF-conditioned media. (a) Schematic diagram showing graft-polymerization used to synthesize the polymer coatings and their chemical structures. Tissue culture polystyrene dishes were first activated by UV ozone and then methacrylate-based monomers were polymerized from the surface. Table lists contact angle, reduced elastic modulus (GPa) (mean ± s.d.), initial hES cell aggregate adhesion (mean ± s.e.m.) and number of passages achieved on each polymer coating. (b) Fourier transform infrared spectrum of PMEDSAH coating showing distinct bands at 1,732.9 cm−1 and 1,208.4 cm−1, which indicated presence of carbonyl and sulfonate groups, respectively. Table lists elemental composition of PMEDSAH obtained using X-ray photoelectron spectroscopy. Relative composition of these elements was in agreement with the expected chemical composition of PMEDSAH. (c) Percentage (mean ± s.e.m.) of hES cells expressing OCT3/4 and SOX2 at passages 3 (P03) and 20 (P20). (d) Relative transcript levels of NANOG, OCT3/4 and SOX2 from hES cells cultured on PMEDSAH and Matrigel. (e,f) After 25 passages, hES cells cultured on PMEDSAH and Matrigel (Supplementary Fig. 1) (e) maintained a normal karyotype and (f) retained pluripotency as demonstrated by teratoma formation in immunosuppressed mice. H&E-stained paraffin sections indicating endoderm (goblet-like cells at arrow), ectoderm (neuroepithelial aggregates at arrow; and cells expressing neuron-restricted protein β-III tubulin in inset) and mesodermal derivatives (cartilage, connective tissue and muscle at arrow). Scale bar, 200 μm.

Cell-aggregate adhesion (%)

© 2010 Nature America, Inc. All rights reserved.

b r i e f c o m m u n i c at i o n s

Figure 2 PMEDSAH supports culture of hES cells in defined medium. (a) Percentage (mean ± s.e.m.) of cell aggregate adhesion (number of aggregates attached with respect to total aggregates passaged) and population doubling time (twofold increase in colony area) for H9 hES cells cultured on PMEDSAH in MEF-conditioned medium, human cell–conditioned medium and defined medium. P-values calculated using unpaired t-test. (b) Fluorescence micrographs of colonies of H9 cells cultured on PMEDSAH in Endoderm Ectoderm Mesoderm StemPro medium showing expression of hES cell markers: OCT3/4, SOX2, SSEA-4, TRA1-60 and TRA-1-81; and a phase-contrast image. Scale bars, 200 μm. (c) RT-PCR analysis of RNA from embryoid bodies showing expression of endoderm (GATA4), ectoderm (KRT18) and mesoderm derivatives (VE-cadherin; also known as CDH5). β-Actin (also known as ACTB) was used as positive control, and for each primer set tested, a reaction lacking RNA was assessed in parallel as a negative control. (d) Micrographs showing immunoreactivity for α-fetoprotein (endoderm), β-III tubulin (ectoderm) and smooth muscle actin (mesoderm) demonstrating the pluripotent state of H9 cells cultured on PMEDSAH in StemPro medium. Scale bars, 200 μm.

582

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

b r i e f c o m m u n i c at i o n s represents a significant step in the development of a fully defined, reproducible culture system for hES cell expansion. Note: Supplementary information is available on the Nature Biotechnology website.

© 2010 Nature America, Inc. All rights reserved.

Acknowledgments H.N. acknowledges funding from the University of Michigan Rackham Predoctoral Fellowship. J.L. gratefully acknowledges support from the National Science Foundation (NSF) in the form of a CAREER grant and funding from the NSF under the MRI program. We would like to thank J.H. Elisseeff and N. Hwang for their insightful comments throughout this study. We further thank J. Garcia-Perez, M. Morell, L.S.D. Emmett, M. Dzaman, M. Bormann, C. Pacut, N.Leff and J. MacDonald for their valuable assistance. We would also like to thank C. Smith, M. Yoshida and S. Brown for their invaluable comments on the manuscript. This research was supported by US National Institutes of Health grants P20 GM-069985, R01 DE016530 and the National Institute of Dental and Craniofacial Research T32 Tissue Engineering and Regeneration Training Program DE 07057. AUTHOR CONTRIBUTIONS L.G.V.-D. and H.N. contributed equally to this work and were involved in experimental design, performing hES cell culture experiments, data analysis and manuscript preparation. L.G.V.-D. carried out cell analysis experiments, and H.N. fabricated the polymer coatings and performed surface analysis. J.D. was involved

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

in immunocytochemistry and RT-PCR, while N.C.N.-d.-S. conducted microarray analysis and qPCR. P.H.K. participated in manuscript preparation. K.S.O. participated in manuscript preparation and performed teratoma assays. J.L. and G.D.S. were involved in conceptual and experimental design, as well as in manuscript preparation. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Ludwig, T.E. et al. Nat. Biotechnol. 24, 185–187 (2006). 2. Wang, L. et al. Blood 110, 4111–4119 (2007). 3. Braam, S.R. et al. Stem Cells 26, 2257–2265 (2008). 4. Brafman, D.A. et al. Stem Cells Dev. 18, 1141–1154 (2009). 5. Mallon, B.S. et al. Int. J. Biochem. Cell Biol. 38, 1063–1075 (2006). 6. Martin, M.J. et al. Nat. Med. 11, 228–232 (2005). 7. Li, Y.J. et al. J. Biomed. Mater. Res. A 79, 1–5 (2006). 8. Derda, R. et al. ACS Chem. Biol. 2, 347–355 (2007). 9. Anderson, D.G. et al. Nat. Biotechnol. 22, 863–866 (2004). 10. Hakala, H. et al. Tissue Eng Part A 15, 1775–1785 (2009). 11. Xu, C. et al. Nat. Biotechnol. 19, 971–974 (2001). 12. Prowse, A.B. et al. Stem Cells Dev. 18, 1135–1140 (2009).

583

Articles

Assessing therapeutic responses in Kras mutant cancers using genetically engineered mouse models

© 2010 Nature America, Inc. All rights reserved.

Mallika Singh1, Anthony Lima1, Rafael Molina1, Patricia Hamilton1, Anne C Clermont1, Vidusha Devasthali1, Jennifer D Thompson1, Jason H Cheng1, Hani Bou Reslan2, Calvin C K Ho2, Timothy C Cao2, Chingwei V Lee3, Michelle A Nannini4, Germaine Fuh3, Richard A D Carano2, Hartmut Koeppen5, Ron X Yu6, William F Forrest6, Gregory D Plowman7 & Leisa Johnson1 The low rate of approval of novel anti-cancer agents underscores the need for better preclinical models of therapeutic response as neither xenografts nor early-generation genetically engineered mouse models (GEMMs) reliably predict human clinical outcomes. Whereas recent, sporadic GEMMs emulate many aspects of their human disease counterpart more closely, their ability to predict clinical therapeutic responses has never been tested systematically. We evaluated the utility of two state-of-the-art, mutant Krasdriven GEMMs—one of non-small-cell lung carcinoma and another of pancreatic adenocarcinoma—by assessing responses to existing standard-of-care chemotherapeutics, and subsequently in combination with EGFR and VEGF inhibitors. Standard clinical endpoints were modeled to evaluate efficacy, including overall survival and progression-free survival using noninvasive imaging modalities. Comparisons with corresponding clinical trials indicate that these GEMMs model human responses well, and lay the foundation for the use of validated GEMMs in predicting outcome and interrogating mechanisms of therapeutic response and resistance. Genetically engineered mouse models (GEMMs) of cancer faithfully recapitulate various aspects of the corresponding human disease1, indicating that they might serve as valuable preclinical platforms for testing new therapeutic agents. Indeed, there are several successful examples of their use in investigating the mechanisms of action and therapeutic resistance to molecularly targeted agents in GEMMs2–4. Despite the widely held view that GEMMs should be better predictors of clinical outcome than xenograft models, they are not usually employed in oncology drug development5,6. Hurdles to the consistent adoption of GEMMs in the pharmaceutical setting include the lack of infrastructure and resources to consistently generate and evaluate the large numbers of genetically complex mice needed for preclinical experiments as well as an absence of a consensus framework for modeling human therapeutic responses in clinical trials in these mice (reviewed in ref. 6). One major obstacle in the latter category is a dearth of studies that adequately simulate clinical trials using standard-of-care chemotherapeutics, either alone or in combination with targeted agents. Such data are required to validate GEMMs as predictive preclinical platforms for evaluating novel therapies that are routinely tested in the clinic in combination with standard-of-care agents. In this study, we use two previously established and characterized GEMMs that are driven by the same conditional mutant Kras allele to examine standard-of-care agents both alone and in combination with therapeutics targeting epidermal growth factor receptor (EGFR) and vascular endothelial growth factor (VEGF): KrasLSL−G12D; p53frt/frt

mice to model non-small cell lung cancer (NSCLC, adenocarcinoma subtype), and KrasLSL−G12D; p16/p19fl/fl; Pdx1-Cre to simulate pancreatic ductal adenocarcinoma (PDAC)7,8. KRAS is mutated in a significant proportion of PDAC, colorectal and non-small-cell lung cancer (NSCLC) patients9 and retrospective analyses have recently indicated that KRAS mutations can differentially influence the response of colorectal cancer and NSCLC patients treated with various EGFR inhibitors (reviewed in ref. 10). The molecular basis of this interference is unknown but is clearly influenced by tumor type, disease setting (naive versus refractory), chemotherapeutic regimen and type of EGFR inhibitor (tyrosine kinase inhibitor (TKI) versus antibody). In KRASmutated NSCLC or colorectal tumors, concurrent administration of an EGFR inhibitor and standard-of-care chemotherapy is inferior to chemotherapy alone11–13. However, this antagonism can be mitigated in first-line NSCLC treatment by administering chemotherapy upfront, followed by maintenance use of an EGFR-TKI 14. In PDAC, where a majority (~70–90%) of patients carry KRAS mutations9,15, comprehensive analyses of KRAS mutations in clinical trial samples are rarely available. Conventional xenograft models have not been able to faithfully predict or recapitulate the influence of KRAS mutations on response to treatment with EGFR inhibitors and chemotherapy (reviewed in ref. 16). The primary mode of action of anti-angiogenic therapies is to indirectly affect tumor viability by targeting the tumor vasculature and depleting its nutrient supply. Bevacizumab (Avastin)—a humanized,

1Department of Molecular Biology, Genentech, Inc., South San Francisco, California, USA. 2Department of Biomedical Imaging, Genentech, Inc., South San Francisco, California, USA. 3Department of Antibody Engineering, Genentech, Inc., South San Francisco, California, USA. 4Department of Cancer Signaling & Translational Oncology, Genentech, Inc., South San Francisco, California, USA. 5Department of Pathology, Genentech, Inc., South San Francisco, California, USA. 6Department of Biostatistics, Genentech, Inc., South San Francisco, California, USA. 7Department of Tumor Biology & Angiogenesis, Genentech, Inc., South San Francisco, California, USA. Correspondence should be addressed to M.S. ([email protected]) and L.J. ([email protected]).

Received 28 December 2009; accepted 28 April 2010; published online 23 May 2010; doi:10.1038/nbt.1640

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

585

© 2010 Nature America, Inc. All rights reserved.

Articles monoclonal antibody that selectively inhibits human VEGF-A (hereafter called VEGF)—has now been approved by the US Food and Drug Administration for use in lung, breast, colon and renal cell cancer in combination with chemotherapy and/or immunotherapy and as a single agent in glioblastoma. In contrast, bevacizumab in combination with gemcitabine (Gemzar) failed to benefit patients with pancreatic cancer, thus highlighting that differences in disease context can influence response to this targeted therapy17. Whereas there are markers to guide the use of EGFR inhibitors, there are currently no known markers that are predictive of response to anti-VEGF therapy. Modeling the therapeutic activity of anti-VEGF antibodies in human xenografts is challenging for two main reasons: first, because much of the VEGF-supporting tumor vasculature is derived from the host, antibodies need to cross-react with both human and mouse VEGF18; second, the vasculature supporting tumor xenografts and/or orthotopics is structurally distinct from that which develops in concert with endogenous neoplasia. The first obstacle can be circumvented by using phage-derived monoclonal antibodies that functionally block both mouse and human VEGF with similar binding efficiencies19. The problem of anomalous vasculature begs the use of tumor models like GEMMs. Recently, the combination of EGFR inhibition along with bevacizumab has shown additive efficacy in xenograft models and in a small NSCLC patient cohort in the second-line setting (OSI 2950 (ref. 20), Supplementary Table 1, reviewed in ref. 21). This combination is currently being clinically evaluated in several solid tumor types, including NSCLC and PDAC. Initial results from these trials (BeTA22, ATLAS23, and AViTA24, see Supplementary Table 1) show that the combination of erlotinib (Tarceva) and bevacizumab can provide some benefit to patients. Moreover, whereas EGFR mutation status is the strongest predictor of response to EGFR kinase inhibitors, even patients with KRAS mutant tumors can benefit from this combination therapy25. Another major challenge in preclinical modeling is the selection of appropriate endpoints that simulate those used in human clinical trials, that is, overall survival (OS), progression-free survival (PFS), time to progression and overall response rate26. It is difficult to draw statistical parallels between these clinical endpoints and those commonly used in standard xenograft models such as tumor growth delay, optimal median treated-tumor mass/median control-tumor mass or net log cell kill27. Whereas the gold standard in measuring therapeutic efficacy in clinical trials is OS, this endpoint is infrequently used in xenografts and would be difficult to interpret because subcutaneous transplants do not recapitulate the systemic and metabolic effects of spontaneous cancers. Orthotopic transplants model the tumor microenvironment better but typically fail to capture the entirety of tumor evolution or the contribution of the immune system, save in syngeneic systems. Genetically engineered mouse tumor models may not recapitulate every aspect of the human disease (e.g., frequency and site of metastasis), but the spontaneous and unconstrained way in which such tumors evolve more closely mirrors both the tumor cell–intrinsic and microenvironmental features of naturally occurring cancers. Access to noninvasive imaging techniques for small animals has facilitated the tracking of tumor progression in murine tumor models. Here, we use two of these modalities, X-ray micro-computed tomo graphy (micro-CT)28 and high-resolution micro-ultrasound29, to follow tumor progression and define PFS endpoints in the NSCLC and PDAC GEMMs, respectively. By comparing OS and PFS in each preclinical model with available corresponding human clinical trial data, we have tested the utility of GEMMs as effective platforms to model therapeutic response and resistance. 586

RESULTS Preclinical modeling in Kras-driven GEMMs We first established a workflow and infrastructure for preclinical studies using GEMMs (Supplementary Fig. 1). To analyze the effect of tumor context on therapeutic response to mutant Krasdriven disease, we exploited the KrasLSL−G12D; p53frt/frt NSCLC and KrasLSL−G12D; p16/p19fl/fl; Pdx1-Cre PDAC GEMMs. Both models carry genetic alterations frequently found in their corresponding human indication: KRAS is mutated in ~20–30% and TP53 in ~35–60% of human NSCLC (adenocarcinoma subtype)9,30, whereas KRAS and P16/CDKN2A are each mutated in ~70–90% of human pancreatic carcinomas9,15 and frequently overlap31. In addition, both GEMMs faithfully model multi-stage tumor progression from early to advanced disease7,8. Gross, histological and biomedical imaging analyses were used in conjunction with OS to establish timelines for disease onset and tumor progression (Supplementary Figs. 2 and 3)7,8,29. Mice exhibiting advanced/late-stage disease (~16 weeks after adenovirus infection for NSCLC and ~7 weeks of age for PDAC) were imaged by either micro-CT or micro-ultrasound and stratified to preclinical trial cohorts. To validate both GEMMs as preclinical models, we employed chemotherapeutics currently used as the standard of care for each cancer type: carboplatin (Paraplatin) for NSCLC and gemcitabine for PDAC. We also evaluated the ability of both GEMMs to model human response in clinical trials of the first-line therapies anti-EGFR smallmolecule erlotinib and anti-VEGF monoclonal antibody bevacizumab (Supplementary Table 1). As pharmacokinetic parameters are important in assessing drug candidates in preclinical development, particularly in combinations where drug-drug interactions may alter the metabolism of various agents, we first determined the maximum tolerated doses and pharmacokinetic profiles for each therapeutic agent in strain-matched C57BL/6N mice (Supplementary Table 2; data not shown). These data then guided subsequent treatment regimens (Supplementary Fig. 4) to best approximate the corresponding clinical trials. Although platinum analogs are routinely used in combination with taxanes in NSCLC patients, we chose to examine carboplatin as a single agent both to mitigate combination toxicity and to facilitate deconvolution of therapeutic combinations (see below). Erlotinib and chemotherapy in KRAS mutant NSCLC The TRIBUTE phase 3 (and TALENT; Supplementary Table 1) investigated the therapeutic efficacy of concurrent dosing of platinumbased chemotherapy and erlotinib as first-line therapy for NSCLC patients. This combination neither met its primary OS endpoint nor improved PFS32 (Supplementary Fig. 5). Furthermore, a retrospective analysis of patients with tumors harboring KRAS mutations indicated that concurrent dosing of erlotinib and chemotherapy is inferior to chemotherapy alone, reducing both OS and PFS 12 (reproduced in Fig. 1a,b). Because EGFR-TKIs block cells in the G1 phase of the cell cycle, these findings may be explained, in part, by the observed reduction in cytotoxic activity of many chemotherapeutics when dosed simultaneously in vitro33. Taken together, these observations led to subsequent clinical studies of first-line therapy in advanced NSCLC patients (Supplementary Table 1; SATURN14 and ATLAS23) where erlotinib was examined in a maintenance setting following chemotherapy. Initial results from SATURN showed a 41% improvement in PFS compared to placebo, with a markedly increased response in patients whose tumors possessed an activating EGFR mutation34. Additional biomarker analysis revealed that whereas KRAS mutation is a negative prognostic VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

Articles

1.0 CP (30) CPE (25)

Survival rate

0.8

b

Median P HR – 13.5 – 4.4 0.02* 2.1

Progression free rate

a

0.6 0.4 0.2 0

1.0 CP (30) CPE (25)

0.8 0.6

Control

0.4 0.2

5

10

15

20

0

25

1.0 0.8

Median P HR 4.9 7.9 – – 6.0 0.07 2.8

0.6 0.4 0.2 0

4

8

12

16

20

d Proportion progression free

Proportion surviving

© 2010 Nature America, Inc. All rights reserved.

Control (32) C (9) CE (12)

5

10

15

20

71.52

35.98

48.88

31.51

47.19

25

Progression-free survival (months)

Overall survival (months)

0

40.72

0 0

c

e

Median P HR 6.0 – – 3.4 0.03* 1.9

Control (32) C (9) CE (12)

1.0 0.8

C

Median P HR 3.9 4.0 – – 3.9 0.26 1.7

0.6 CE

0.4 0.2 0

0

2

Weeks on study

4

6

8

10

12

14

Weeks on study

Figure 1 Influence of KRAS mutations in the first-line treatment of NSCLC with chemotherapy versus chemotherapy plus erlotinib. (a–d) Kaplan-Meier plots from TRIBUTE12 showing the OS (a) and PFS (b) data from NSCLC patients with KRAS mutant tumors treated with chemotherapy alone (CP) or chemotherapy plus concurrent erlotinib (CPE). These figures reprinted with permission, ©2008 American Society of Clinical Oncology. All rights reserved. Corresponding OS (c) and PFS (d) data in the Kras mutant GEMM for NSCLC; note that the mice were dosed with erlotinib maintenance therapy. Control vehicle-treated cohorts are also shown in the case of the GEMM for comparison. The number of patients or mice in each treatment cohort is indicated in parentheses. Relevant P values (log-rank test) and hazard ratios are depicted for each graph, and the comparator group (denominator) is denoted with a hyphen. *, P values ≤ 0.05. CI95 for each hazard ratio shown are (a) 1.11–3.79, (b) 1.05–3.6, (c) 0.92–8.54 and (d) 0.67–4.34. (e) Representative micro-CT images from mice in each of the GEMM cohorts pre-treatment (left) and after ~4 weeks on study (right). A visible tumor example is marked with yellow arrows in the control vehicle-treated cohort and mean tumor burden cross-products for each cohort are shown at the bottom left of each image. C, carboplatin; P, paclitaxel; E, erlotinib; HR, hazard ratio; micro-CT, micro-computed tomography.

marker (P = 0.017; hazard ratio (HR) = 1.50, CI95 = 1.06–2.12), no significant difference in PFS was found in patients with KRAS wild-type versus mutant tumors after erlotinib maintenance therapy34. We sought to ascertain if the Kras mutant NSCLC GEMM could model these observed differences in outcome based on the erlotinib dosing regimen. As simultaneous dosing of erlotinib with either carboplatin or taxanes was poorly tolerated in mice (data not shown), we were not able to model the concurrent treatment regimen used in TRIBUTE. Instead, we administered sequential dosing in KrasLSL−G12D; p53frt/frt mice, treating first with the maximum tolerated dose of carboplatin for 5 d, followed by erlotinib maintenance therapy (Supplementary Fig. 4a). This dosing schedule was well tolerated (Supplementary Fig. 6a,b) and best mirrors the clinical regimen pursued in SATURN. As shown, similar profiles of rapid disease progression were observed in both mice and humans34 (Fig. 1) harboring KRAS mutant NSCLC. However, there was no significant impact on PFS when erlotinib was dosed sequentially in the Kras mutant GEMM (P = 0.26; Fig. 1d), similar to that observed in patients (n = 90) from SATURN with KRAS mutant tumors (P = 0.22; HR = 0.77, CI95 = 0.50–1.19)34. In the Kras mutant GEMM, this translated into an insignificant though notable decrease in OS relative to carboplatin treatment alone (P = 0.07; HR = 2.8, CI95 = 0.92–8.54; Fig. 1c), closely reflecting that observed with larger cohorts of patients in TRIBUTE (Fig. 1a). Xenograft models offered little insight into erlotinib response as both concurrent (albeit under the maximum tolerated dose) and sequential erlotinib dosing regimens demonstrated enhanced efficacy in KRAS mutant NSCLC lines35,36 (Supplementary Fig. 7a,b). Overall, these data indicate that the NSCLC GEMM provides insight into the erlotinib response in the context of mutant Kras and wild-type Egfr. The encouraging human data suggest that patients with KRAS mutant nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

NSCLC are likely to harbor additional alterations in the EGFR pathway that render them sensitive to erlotinib (see Discussion). Erlotinib and chemotherapy in KRAS mutant PDAC The therapeutic impact of erlotinib in pancreatic cancer is different from that in NSCLC. The NCIC CTG PA.3 Phase 3 clinical trial of first-line therapy for pancreatic cancer (Supplementary Table 1) showed modest but significant OS (P = 0.038; HR = 0.82, CI95 = 0.69– 0.99) and PFS (P = 0.004; HR = 0.77, CI95 = 0.64–0.92) benefits for patients treated with gemcitabine plus erlotinib versus those treated with gemcitabine alone37 (reproduced in Fig. 2a,b). A retrospective survival analysis of a small subset of patients (n = 84) whose tumor KRAS mutation status was available showed no significant difference in OS (P = 0.34; HR = 1.24, CI95 = 0.79–1.94) or PFS (P = 0.77; HR = 0.93, CI95 = 0.59–1.48) between the effects of gemcitabine plus erlotinib versus gemcitabine alone38. Although most pancreatic cancer patient tumors are thought to harbor KRAS mutations9, the differences between these results can be attributed to the small size of this patient sub-population and the modest effect that was observed in the larger total population. We examined dosing and pharmacokinetic profiles for gemcitabine and erlotinib in C57BL/6N mice (Supplementary Table 2) and found both agents to be well tole rated in a concurrent dosing scheme that simulated the NCIC CTG PA.3 trial in the KrasLSL−G12D; p16/p19fl/fl; Pdx1-Cre PDAC GEMM (Supplementary Figs. 4b and 6c,d). We observed that gemcitabine treatment resulted in a significant (P < 0.0001) survival advantage as compared to vehicle control (Fig. 2c). The response to gemcitabine in the PDAC GEMM seen here is somewhat greater than that typically observed in the human patient population and may reflect greater heterogeneity among human versus mouse tumors. As observed in humans harboring KRAS mutant PDAC, we saw no significant 587

Articles

80

40 20

Proportion surviving

0

6

12

18

1.0

Control (33) G (32) GE (9)

0.8

Median P HR 2.3 4.0 – – 4.4 0.46 0.75

0.6 0.4 0.2 0

2

4

6

8

10

d

G (284) GE (285)

80

Median P HR 3.6 – – 3.8 0.004*` 0.77

60

e Control

40 20 0

24

Time (months)

0

© 2010 Nature America, Inc. All rights reserved.

b 100

Median P HR 5.9 – – 6.2 0.038* 0.82

60

0

c

G (284) GE (285)

Survival probability (%)

100

Proportion progression free

Survival probability (%)

a

0

5

10

Control (32) G (29) GE (9)

0.8 0.6

149.03

19.74

67.31

28.10

66.05

15

Time (months) 1.0

22.19

Median P HR 1.4 1.6 – – 1.4 0.31 1.48

G

0.4 GE

0.2 0 0

1

Weeks on study

2

3

4

Weeks on study

Figure 2 First-line treatment of PDAC patients and KrasLSL−G12D; p16/p19fl/fl; Pdx1-Cre mice with gemcitabine versus gemcitabine plus erlotinib. (a–d) Kaplan-Meier plots from NCIC CTG PA.3 (ref. 37) showing the OS (a) and PFS (b) data from PDAC patients treated with gemcitabine alone (G) or gemcitabine plus erlotinib (GE). These figures reprinted with permission, © 2008 American Society of Clinical Oncology. All rights reserved. Corresponding OS (c) and PFS (d) data in the Kras mutant GEMM for PDAC are shown. Vehicle control-treated cohorts are also shown in the case of the GEMM for comparison. The number of patients or mice in each treatment cohort is indicated in parentheses. Relevant P values (log-rank test) and hazard ratios are depicted for each graph, and the comparator group (denominator) is denoted with a hyphen. CI 95 for each hazard ratio shown are (a) 0.69–0.99, (b) 0.64–0.92, (c) 0.35–1.6 and (d) 0.69–3.2. (e) Representative high-resolution ultrasound images from mice in each of the GEMM cohorts pre-treatment (left) and after ~11–15 d on study (right). Visible lesions are outlined in yellow and mean tumor burden for each cohort are shown at the bottom left of each image. G, gemcitabine; E, erlotinib; HR, hazard ratio.

difference in OS (P = 0.46; HR = 0.75, CI95 = 0.35–1.6; Fig. 2c) or PFS (P = 0.31; HR = 1.48, CI95 = 0.69–3.2; Fig. 2d) between the gemcitabine plus erlotinib and gemcitabine treatment groups in the PDAC GEMM. However, the OS hazard ratio suggests that the combination group may be nominally advantageous, trending in the same direction as the overall human outcome (Fig. 2a). We did not detect any negative impact on OS when combining erlotinib and chemotherapy in the PDAC GEMM, unlike what was observed in the NSCLC GEMM. Hence, context is important in predicting the response to targeted EGFR agents. Anti-VEGF and chemotherapy in KRAS mutant NSCLC We next addressed the therapeutic impact of agents directed against the tumor vasculature by targeting VEGF. Because bevacizumab does not bind mouse VEGF, we used a phage-derived monoclonal antibody, B20-4.1.1, that functionally blocks both mouse and human VEGF 19. Both B20-4.1.1 and bevacizumab have similar binding affinities for human VEGF (data not shown) and have been extensively studied in preclinical models. To examine the validity of modeling anti-angiogenic therapies in the NSCLC GEMM, we simulated the pivotal ECOG 4599 trial, which demonstrated significant OS and PFS benefits for patients receiving platin-based chemotherapy in combination with bevacizumab versus chemotherapy alone39 (reproduced in Fig. 3a,b). As with human NSCLC patients, KrasLSL−G12D; p53frt/frt mice treated with carboplatin and anti-VEGF showed significant improvement over chemotherapy alone in both OS (P = 0.0007; HR = 0.11, CI95 = 0.03–0.40) and PFS (P = 0.0003; HR = 0.10, CI 95 = 0.03–0.34) (Fig. 3c,d). In this case, the GEMM data are consistent with the results obtained with antiVEGF both alone and in combination with paclitaxel (Taxol) in the KRAS mutant A549 NSCLC xenograft (Supplementary Figs. 7c,d). The results shown here with the GEMM suggest that patients 588

with NSCLC harboring KRAS mutations (~20–30% of all NSCLC patients) would show a positive response to anti-VEGF combination therapy. Indeed analysis of such patients from the BeTA trial (Supplementary Table 1) confirm that regimens containing antiVEGF show a favorable PFS response that is independent of their tumor’s KRAS mutational status25. Anti-VEGF and chemotherapy in PDAC In contrast to the success of anti-VEGF combination therapy in NSCLC, the CALGB 80303 phase 3 trial gauging the impact of combined gemcitabine and bevacizumab in pancreatic cancer patients (Supplementary Table 1) was discontinued because there was no survival benefit17 (Fig. 4a,b). We examined a similar combination regimen in KrasLSL−G12D; p16/p19fl/fl; Pdx1-Cre mice and monitored PFS and OS as before. In contrast to the human data, we observed a significant OS benefit (P = 0.017; HR = 0.43, CI95 = 0.22–0.86; Fig. 4c) in the PDAC GEMM treated with anti-VEGF plus gemcitabine compared to gemcitabine alone. However, closer examination of the survival curves revealed that 50% of the mice performed similarly to the human clinical data, whereas a subset of animals that survived beyond 4 weeks were responsible for this improved response. We also detected a similar trend in PFS (P = 0.06; HR = 0.54, CI95 = 0.28–1.03; Fig. 4d). As the mice comprise a relatively isogenic population, these dichotomous results suggest that additional factors must be responsible for differentiating the response to anti-VEGF plus gemcitabine, providing a unique opportunity for investigating the molecular differences therein. Combining anti-VEGF, erlotinib and chemotherapy in KRAS mutant NSCLC and PDAC Next, we investigated the therapeutic impact of combining anti-VEGF and erlotinib, both with and without various chemotherapeutics, in the VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

Articles

1.0 CP (444) CPA (434)

0.8

b

HR Median P – 10.3 – 12.3 0.003* 0.79

Proportion w/o event

Proportion surviving

a

0.6 0.4 0.2

1.0 0.8 0.6 0.4

Control

0.2 0

0 0

10

20

30

40

0

50

10

0.8 0.6 0.4 0.2 0

4

8

12

16

d Proportion progression free

Proportion surviving

1.0

0

30

40

40.72

71.52

35.98

48.88

43.34

43.74

50

1.0

C

Median P HR 3.9 4.0 – – 10.4 0.0003* 0.10

Control (32) C (9) CA (17)

0.8 CA

0.6 0.4 0.2 0

20

0

2

4

6

8

10

12

14

Weeks on study

Weeks on study

Figure 3 Anti-VEGF provides significant benefit when combined with chemotherapy as first-line therapy in human patients and KrasLSL−G12D; p53frt/frt mice with late-stage NSCLC. (a–d) Kaplan-Meier plots from ECOG 4599 (ref. 39) showing the OS (a) and PFS (b) data from NSCLC patients treated with chemotherapy alone (CP) or chemotherapy plus bevacizumab (CPA). Corresponding OS (c) and PFS (d) data in the Kras mutant GEMM for NSCLC are shown; the control vehicle- and carboplatin-treated cohorts are reproduced from Figure 1 for comparison. The number of patients or mice in each treatment cohort is indicated in parentheses. Relevant P values (log-rank test) and hazard ratios are depicted for each graph, and the comparator group (denominator) is denoted with a hyphen. *, P values ≤ 0.05. CI95 for each hazard ratio shown are (a) 0.69–0.93, (b) 0.56–0.76, (c) 0.032–0.399 and (d) 0.027–0.341. Panels a and b reproduced with the kind permission of Alan Sandler. (e) Representative micro-CT images from mice in each of the GEMM cohorts pretreatment (left) and after ~4 weeks on study (right). A visible tumor example is marked with yellow arrows in the control-treated cohort and mean tumor burden cross-products for each cohort are shown at the bottom left of each image. Control and carboplatin treatment images are reproduced from Figure 1 for comparison. C, carboplatin; P, paclitaxel; A, anti-VEGF; HR, hazard ratio; micro-CT, micro-computed tomography.

maintenance setting. Thus far, this study has successfully met its primary PFS endpoint and future analyses of the data sets will evaluate the role of KRAS status in mediating patient response and outcome. In the Kras mutant NSCLC GEMM, the most significant anti-tumor response

two GEMMs. Such studies mirror several ongoing human clinical trials aimed at exploring combination therapy with erlotinib and bevacizumab (Supplementary Table 1). ATLAS23 is a phase 3 NSCLC clinical trial assessing the utility of erlotinib plus bevacizumab in a first-line 1.0 0.8

Median P 6.1 – 5.8 0.78

G (300) GA (302)

0.6

b

HR – 1.03

Proportion surviving

Proportion surviving

a

0.4 0.2 0 5

10

15

20

1.0

G (300) GA (302)

0.6

Median P 4.7 – 4.9 0.99

HR – 1.00

Control

0.4 0.2

0

Median P HR 2.3 4.0 – – 4.1 0.017* 0.43

Control (33) G (32) GA (19)

0.8 0.6 0.4 0.2 0

0.8

25

Months from study entry

0

2

4

6

8

10

d

Proportion progression free

c

e

1.0

22.19

149.03

19.74

67.31

40.52

96.89

0 0

Proportion surviving

© 2010 Nature America, Inc. All rights reserved.

Median P HR 4.9 7.9 – – 10.4 0.0007* 0.11

Control (32) C (9) CA (17)

20

Progression-free survival (months)

Duration of survival (months)

c

e

Median P HR 4.5 – – 6.2 0.001* 0.66

CP (444) CPA (434)

5

10

15

20

25

Months from study entry 1.0

Control (32) G (29) GA (18)

0.8

Median P 1.4 1.6 – 1.9 0.06

G

HR – 0.54

0.6 0.4 GA

0.2 0

0

1

Weeks on study

2

3

4

5

6

Weeks on study

Figure 4 First-line treatment of PDAC patients and KrasLSL−G12D; p16/p19fl/fl; Pdx1-Cre mice with gemcitabine versus gemcitabine plus anti-VEGF. (a–d) KaplanMeier plots from CALGB 80303 (ref. 17) showing the OS (a) and PFS (b) data from PDAC patients treated with gemcitabine alone (G) or gemcitabine plus bevacizumab (GA). Corresponding OS (c) and PFS (d) data in the Kras mutant GEMM for PDAC are shown; the control vehicle- and gemcitabine-treated cohorts are reproduced from Figure 2 for comparison. The number of patients or mice in each treatment cohort is indicated in parentheses. Relevant P values (log-rank test) and hazard ratios are depicted for each graph, and the comparator group (denominator) is denoted with a hyphen. *, P values ≤ 0.05. CI95 for each GEMM hazard ratio shown are (c) 0.22– 0.86 and (d) 0.28–1.03. Panels a and b reproduced with the kind permission of Hedy Kindler. (e) Representative high-resolution ultrasound images from mice in each of the GEMM cohorts pre-treatment (left) and after ~11–15 d on study (right). Visible lesions are outlined in yellow and mean tumor burden for each cohort are shown at the bottom left of each image. Control and gemcitabine treatment images are reproduced from Figure 2 for comparison. G, gemcitabine; A, anti-VEGF; HR, hazard ratio.

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

589

Articles b

Proportion progression free

c

d

© 2010 Nature America, Inc. All rights reserved.

Proportion surviving

Proportion progression free

a

Proportion surviving

Figure 5 Anti-VEGF is a primary driver of NSCLC GEMM PFS Median NSCLC GEMM OS response in the Kras mutant NSCLC GEMM. Median 1.0 3.9 CE (12) 1.0 (a,b) Anti-VEGF overcomes the negative 7.0 EA (9) CE (12) 6.0 10.4 CA (17) 0.8 interaction observed with carboplatin followed 0.8 EA (9) 8.1 9.9 CEA (10) CA (17) 10.4 by erlotinib maintenance on both OS and PFS. 0.6 0.6 CEA (10) 9.9 Kaplan-Meier plots showing OS (a) and PFS (b) 0.4 0.4 following different dual- and triple-combination regimens using chemotherapy, erlotinib and 0.2 0.2 anti-VEGF are shown. (c,d) Anti-VEGF confers 0 0 a significant OS and PFS benefit relative to 0 4 8 12 16 20 0 2 4 6 8 10 12 14 vehicle control, both as a single agent and in Weeks on study Weeks on study combination with erlotinib. Single-agent NSCLC GEMM OS NSCLC GEMM PFS effects on OS (c) and PFS (d) in comparison Median Median 1.0 1.0 Control (32) 3.9 with the combination of the two targeted agents, 4.9 Control (32) 3.9 E (11) 0.8 anti-VEGF and erlotinib. Significant P values 5.6 E (11) 0.8 A (27) 6.9 6.9 A (27) C (9) 4.0 (log-rank test) and hazard ratios for each graph 7.9 C (9) 0.6 0.6 EA (9) 7.0 are as follows: (a) CA versus CE (P = 0.0003, 8.1 EA (9) 0.4 HR = 0.20, CI95 = 0.08–0.48); EA versus CE 0.4 (P = 0.0003, HR = 0.15, CI95 = 0.05–0.42); 0.2 0.2 CEA versus CE (P = 0.0012, HR = 0.19, CI95 = 0 0.07–0.52). (b) CA versus CE (P = 0.0002, 0 0 2 4 6 8 10 12 14 20 0 4 8 12 16 HR = 0.17, CI95 = 0.07–0.43); EA versus CE Weeks on study Weeks on study (P = 0.0011, HR = 0.17, CI95 = 0.06–0.49); CEA versus CE (P = 0.0004, HR = 0.15, CI95 = Pre-treatment Post-treatment Pre-treatment Post-treatment 0.05–0.43). (c) A versus control (P = 0.0001, HR = 0.32, CI95 = 0.18–0.57); A versus E CE Control (P = 0.0017, HR = 0.28, CI95 = 0.12–0.62); A versus C (P = 0.024, HR = 0.38, CI95 = 31.51 47.19 40.72 71.52 0.16–0.88); EA versus control (P = 0.0052, HR = 0.32, CI95 = 0.15–0.71); EA versus E EA E (P = 0.0084, HR = 0.28, CI95 = 0.11–0.72); EA versus C (P = 0.057, HR = 0.39, CI95 = 18.72 33.98 35.10 38.24 0.14–1.03). (d) A versus control (P = 6.3e–08, HR = 0.17, CI95 = 0.09–0.32); A versus E CA A (P = 3e–06, HR = 0.14, CI95 = 0.06–0.32); 51.99 56.67 A versus C (P = 0.0011, HR = 0.25, CI95 = 43.34 43.74 0.11–0.58); EA versus control (P = 0.0004, HR = 0.22, CI95 = 0.10–0.51); EA versus E C CEA (P = 0.0008, HR = 0.19, CI95 = 0.07–0.50); 48.54 35.12 35.98 48.88 EA versus C (P = 0.03, HR = 0.34, CI95 = 0.13–0.89). (e) Representative micro-CT images from mice in each of the cohorts pre-treatment (left) and after ~4 weeks on study (right). Mean tumor burden cross-products for each cohort are shown at the bottom left of each image. The control, C, CE and CA cohorts are reproduced from previous figures. C, carboplatin; A, anti-VEGF; E, erlotinib; HR, hazard ratio; CI95, 95% confidence interval; micro-CT, micro-computed tomography.

e

was observed with the triple regimen of carboplatin (C), erlotinib (E), and anti-VEGF (A) at 2 weeks (Supplementary Fig. 8c). However, this effect was not durable (compare Supplementary Fig. 8c,f) and did not provide any significant improvement in PFS or OS over antiVEGF combined with either carboplatin (CA) or erlotinib (EA). These results can be compared with ATLAS once biomarker data are available. Notably, none of these anti-VEGF–containing combinations elicited any of the deleterious effects observed with carboplatin followed by erlotinib maintenance in this model; indeed, the addition of anti-VEGF appears to alleviate such effects (Fig. 5a, OS: CEA versus CE, P = 0.0012, HR = 0.19, CI95 = 0.07–0.52 and EA versus CE, P = 0.0003, HR = 0.15, CI95 = 0.05–0.42; Fig. 5b, PFS: CEA versus CE, P = 0.0004, HR = 0.15, CI95 = 0.05–0.43 and EA versus CE P = 0.0011, HR = 0.17, CI95 = 0.06–0.49). To dissect the individual contributions made by each agent in the above combination studies, we assessed the effects of each single treatment in the NSCLC GEMM (Fig. 5c,d). Single agent anti-VEGF showed significant improvements in both OS (P = 0.0001; HR = 0.32, CI95 = 0.18–0.57) and PFS (P = 6.3e−08; HR = 0.17, CI95 = 0.09–0.32) compared to vehicle control and appears to be the primary driver of a response in mice with mutant Kras and wildtype Egfr (Supplementary Fig. 8). 590

Recently, the AViTA phase 3 trial examining the addition of bevacizumab to the combination of gemcitabine and erlotinib in pancreatic cancer patients (Supplementary Table 1) demonstrated a significant improvement in PFS (P = 0.0002; HR = 0.73, CI95 = 0.61–0.86) and a positive trend (albeit not significant) in OS (P = 0.2087; HR = 0.89, CI95 = 0.74–1.07) in the triple regimen arm24. We simulated the treatment arms from AViTA in PDAC-bearing mice (Fig. 6) and found the triple regimen of anti-VEGF, gemcitabine and erlotinib resulted in a PFS with an improved hazard ratio compared to gemcitabine plus erlotinib (P = 0.17; HR = 0.50, CI95 = 0.19–1.34). However, the rapidity of tumor progression in this model, combined with the imaging intervals and much smaller cohort sizes examined here, most likely precluded statistical significance by log-rank analysis. As in human patients, a modest improvement in OS was observed (P = 0.82; HR = 0.90, CI95 = 0.34–2.34), though neither attained statistical significance. These data indicate that incremental benefits can indeed be achieved by combinatorial strategies and that the data from the GEMM resemble those from the significantly larger clinical cohorts in AViTA. We also examined the role of each single agent in the PDAC model (Fig. 6c,d). In contrast to the NSCLC GEMM, neither anti-VEGF nor erlotinib had any impact on PFS and OS. The lack of any anti-VEGF response is in contrast to the apparent efficacy of anti-VEGF antibodies as single agents VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

Articles b

Proportion progression free

c

d

© 2010 Nature America, Inc. All rights reserved.

Proportion surviving

Proportion progression free

a

Proportion surviving

Figure 6 Gemcitabine is a primary driver of PDAC GEMM OS PDAC GEMM PFS a survival benefit, with incremental benefit 1.0 1.0 Median conferred by the addition of targeted agents Median GE (9) 1.4 GE (9) 4.4 0.8 in a Kras mutant PDAC GEMM. (a,b) Kaplan0.8 GA (18) 1.9 GA (19) 4.1 Meier plots showing OS (a) and PFS (b) EA (9) 1.3 EA (9) 4.1 0.6 0.6 following different dual- and triple-combination GEA (9) 3.4 GEA (9) 5.1 0.4 0.4 regimens using chemotherapy, erlotinib and anti-VEGF are shown. (c,d) Targeted agents 0.2 0.2 do not show single-agent effects, but erlotinib 0 0 plus anti-VEGF is comparable to conventional 0 2 4 6 8 10 0 1 2 3 4 5 6 chemotherapy with gemcitabine. Single-agent Weeks on study Weeks on study effects on OS (c) and PFS (d) in comparison PDAC GEMM PFS PDAC GEMM OS with the combination of the two targeted Median 1.0 1.0 Median agents, anti-VEGF and erlotinib. Significant Control (33) 2.3 Control (32) 1.4 P values (log-rank test) and hazard ratios for 0.8 0.8 A (10) 2.4 A (10) 1.4 E (8) 2.6 E (8) 1.6 each graph are as follows: (a) GA versus GE G (32) 4.0 0.6 0.6 G (29) 1.6 (P = 0.20, HR = 0.57, CI 95 = 0.24–1.35; note, EA (9) 4.1 EA (9) 1.3 only the HR is notable in this case), 0.4 0.4 (b) GA versus GE (P = 0.04, HR = 0.40, 0.2 0.2 CI95 = 0.17–0.94), GEA versus GE (P = 0.17, HR = 0.50, CI95 = 0.19–1.34; note, only the 0 0 0 1 2 3 4 5 6 4 0 2 6 8 10 HR is notable in this case), (c) G versus control Weeks on study Weeks on study (P = 3.4e−07, HR = 0.22, CI95 = 0.12–0.39), G versus E (P = 0.0011, HR = 0.25, CI95 = 0.11–0.57), G versus A (P = 0.0012, HR = 0.28, Control GE CI95 = 0.13–0.61), EA versus control (P = 9e−05, HR = 0.18, CI95 = 0.07–0.42), 149.03 22.19 66.05 28.10 EA versus E (P = 0.0028, HR = 0.20, CI95 = 0.07–0.59), EA versus A (P = 0.0037, HR = 0.23, CI95 = 0.08–0.63) and (d) G versus control A GA (P = 0.018, HR = 0.53, CI95 = 0.32–0.90), G versus E (P = 0.43, HR = 0.72, CI95 = 203.29 36.53 40.52 96.89 0.32–1.61), G versus A (P = 0.066, HR = 0.49, CI95 = 0.22–1.05), EA versus control EA E (P = 0.066, HR = 0.47, CI95 = 0.21–1.05), EA versus E (P = 0.39, HR = 0.64, CI95 = 14.35 139.87 34.61 74.38 0.23–1.79), EA versus A (P = 0.09, HR = 0.43, CI95 = 0.16–1.15). Unlike the OS data, only the HRs are notable for each of the G GEA PFS analyses in d except for G versus control. (e) Representative high-resolution ultrasound 19.74 67.31 8.38 16.13 images from mice in each of the cohorts pretreatment (left) and after ~11–15 d on study (right). Visible lesions are outlined in yellow and mean tumor burden for each cohort are shown at the bottom left of each image. The control, G, GE and GA cohorts are reproduced from previous figures. G, gemcitabine; A, anti-VEGF; E, erlotinib; HR, hazard ratio; CI95, 95% confidence interval.

e

in human pancreatic cancer xenografts18 (Supplementary Fig. 7g,h). In the PDAC GEMM, gemcitabine is the primary driver of response as evidenced by the impact on PFS, OS and tumor burden (Supplementary Fig. 9), and the addition of anti-VEGF improved durability. Interestingly, the targeted agent combination (EA) offered marked improvement in both endpoints (as well as tumor burden, Supplementary Fig. 9) versus either agent alone (Fig. 6c, OS: EA versus E (P = 0.0028, HR = 0.20, CI95 = 0.07–0.59), EA versus A (P = 0.0037, HR = 0.23, CI95 = 0.08–0.63); Fig. 6d, PFS: EA versus E (P = 0.39, HR = 0.64, CI95 = 0.23–1.79), EA versus A (P = 0.091, HR = 0.43, CI95 = 0.16–1.15)), closely mirroring that observed with gemcitabine alone. DISCUSSION Despite a clear need for more predictive models in oncology drug development, and over two decades of translational research in GEMMs, they have yet to be widely adopted by the pharmaceutical sector. The work described here constitutes a rigorous attempt to test the feasibility of using GEMMs to interrogate clinically relevant agents and combinations. This study demonstrates that such experiments are doable and establishes a framework for modeling in GEMMs human response to nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

first-line therapies during clinical trials. The experiments in this study are not focused on dissecting the mechanistic events mediating therapeutic response or acquired resistance, but rather serve as retrospective and prospective tests of the ability of GEMMs to reflect therapeutic outcomes in humans. The outcomes observed in the GEMMs show similarities to their clinical counterparts and, compared to conventional xenograft models, better inform of the subtleties that are observed in human clinical trials. To our knowledge, estimating PFS from the noninvasive imaging data has not been attempted previously in small animal models. Our data are largely consistent with the observed OS trends in both GEMMs and indicate that the efficacy and durability observed with noninvasive imaging can be predictive of survival outcomes. The ability to detect minor differences in PFS (in certain instances) was likely reduced by the rapid tumor growth and multiplicity observed in these models, and may be resolved in future analyses employing more frequent imaging intervals and larger cohorts. The response of KRAS mutant tumors to EGFR inhibition has been an active area of debate and investigation in both NSCLC and metastatic colorectal cancer, particularly in light of the approval of erlotinib as a first-line therapy in pancreatic cancer, where ~70–90% 591

© 2010 Nature America, Inc. All rights reserved.

Articles of patient tumors harbor KRAS mutations10,40–44. Indeed, data showing that patients with KRAS mutated colorectal cancers do not benefit from EGFR inhibitory antibodies in combination with chemotherapy have recently prompted the European Medicines Evaluation Agency to restrict the use of panitumamab in colorectal cancer to patients with KRAS wild type tumors. The NSCLC GEMM experiments recapitulate the lack of OS benefit previously observed in small subsets of patients with tumors that harbored KRAS mutations and underwent first- or second-line therapy12,44. It will be interesting to see if data from the SATURN trial validate the observations reported here from the NSCLC GEMM model. However, it will likely be challenging to make such comparisons as well as to demonstrate correlations between OS and PFS in ongoing and future trials as NSCLC patients today have more options and frequently go on to subsequent therapies when their disease progresses. In contrast, the mice in these studies remain on one treatment regimen until the end of study, providing unique opportunities to interrogate the mechanism(s) by which Kras mutations may influence the response to EGFR inhibition. One hypothesis for the difference in response to chemotherapy plus erlotinib between KRAS mutant NSCLC and PDAC is greater co-deregulation of Ras and EGFR signaling in the latter45. Thus, it will be important to further investigate how simultaneous deregulation of both EGFR (either overexpression or activating mutations) and KRAS influences response to EGFRtargeted therapies in the context of lung and pancreatic cancer. Mouse models uniquely afford the ability to genetically dissect how simultaneous alteration of these two oncogenes can influence outcome, particularly given the complications that can arise in interpreting human NSCLC patient datasets due to therapeutic crossover. Anti-VEGF treatment significantly improves the outcome of KrasLSL−G12D; p53frt/frt NSCLC-bearing mice both alone and in combination with chemotherapy. We observed a negative interaction in this GEMM when combining platin-based chemotherapy with erlotinib in the context of mutant Kras and wild type Egfr. However, anti-VEGF was able to overcome this effect, underscoring that in this model, anti-VEGF is a primary driver of response and that any additional benefit observed with erlotinib is likely dependent on EGFR, ErbB3, cMet, ALK and PI3K pathway activation45–48 (data not shown). This may have clinical implications, particularly because it reflects early, though limited, Ph II and Ph III clinical observations20–22. Although biomarker analysis of the ATLAS trial has not been reported, we predict that the additional benefit that erlotinib adds to anti-VEGF would either be in patients with KRAS wild type tumors or in tumors with EGFR pathway activation. We observed incremental benefits when anti-VEGF was added to gemcitabine plus erlotinib in the PDAC GEMM that were similar to the clinical responses attained in AViTA24, indicating that we now have a relevant model to mechanistically dissect the effects of both EGFR and VEGF inhibition. Additionally, the subset of PDAC mice that benefit from anti-VEGF plus gemcitabine treatment could provide insight into the factors mediating the positive patient responses that were observed in smaller phase II PDAC clinical trials with bevacizumab plus gemcitabine49,50. Possible mediators of response include additional genetic changes as well as the mix of histological sub-types observed in both human and murine PDAC. Taken together, the patterns observed in both the human and GEMM trials argue that one or more factors modify the response to EGFR inhibition in KRAS mutant tumors, including the tumor microenvironment (lung versus pancreas), disease stage and histological subtype, the chemotherapeutic and/or regimen used in combination (e.g., platin-based or presence of anti-VEGF), and additional co-existing 592

molecular alterations (e.g., in EGFR itself and/or in the tumor suppressors TP53 and P16/P19/CDKN2A)51,52. Moreover, our findings as well as those recently obtained from clinical trials in NSCLC20,21,23 and PDAC24 imply that anti-VEGF can improve the benefit from erlotinib treatment. Although clinical studies demonstrate that KRAS mutations are clearly associated with worse prognosis in NSCLC34,53,54, the data shown here (along with recent clinical data) strongly indicate that these mutations alone should not be used to define patient eligibility for therapies containing EGFR inhibitors. In summary, GEMMs provide a unique opportunity to study the influence of KRAS mutational status on differences in therapeutic responses in humans and the underlying molecular alterations that mediate these effects. Further model development is required to examine additional alterations that frequently arise in human tumors, both pre- and post-therapeutic intervention. The similarities in responses observed between these GEMM studies and available human clinical trial data validate their preclinical and predictive utility and corroborate their use to model complex therapeutic strategies, discover predictive markers, and identify beneficial drug combinations to evaluate in the clinic. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Note: Supplementary information is available on the Nature Biotechnology website. Acknowledgments We would like to thank S. Kelsey, J. Hambleton, O. Rosen, S. Erickson, F. Borellini, D. Colburn and G. Evan for critically evaluating the manuscript as well as F. de Sauvage, B. Mass and M. Benyunes for invaluable input. J. Bower, V. Javinal, A. Arrazate, L. Nguyen and A. Wong provided excellent technical assistance. We also received extensive and able technical support from the in-house genotyping and murine reproductive technology core groups. A special note of gratitude to H. Wong, L. Salphati, B. Liederer and L. Damico for pharmacokinetic support and analyses. L. Berry and B. Hollister supervised the xenograft studies shown here. B. Tong, J. Yi and J. Wacker provided statistical information and feedback. The graphics and layout were ably provided by J. Wood and D. Wood. AUTHOR CONTRIBUTIONS M.S. and L.J. designed, planned and performed the experiments, analyzed data and wrote the manuscript. A.L., R.M., P.H., A.C.C., V.D., J.D.T., J.H.C., H.B.R., C.C.K.H. and T.C.C. performed experiments and analyzed data. C.V.L. and G.F. developed and provided the B20-4.1.1 anti-VEGF antibody. M.A.N. and R.A.D.C. provided design input and supervised animal dosing and imaging experiments, respectively. G.D.P. provided design input and contributed to manuscript preparation. H.K. carried out histopathological analyses, and R.X.Y. and W.F.F. performed all the statistical analyses and contributed to the writing of the manuscript. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturebiotechnology/. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Van Dyke, T. & Jacks, T. Cancer modeling in the modern era: progress and challenges. Cell 108, 135–144 (2002). 2. Abate-Shen, C. A new generation of mouse models of cancer for translational research. Clin. Cancer Res. 12, 5274–5276 (2006). 3. Engelman, J.A. et al. Effective use of PI3K and MEK inhibitors to treat mutant Kras G12D and PIK3CA H1047R murine lung cancers. Nat. Med. 14, 1351–1356 (2008). 4. Varmus, H., Pao, W., Politi, K., Podsypanina, K. & Du, Y.C. Oncogenes come of age. Cold Spring Harb. Symp. Quant. Biol. 70, 1–9 (2005). 5. Becher, O.J. & Holland, E.C. Genetically engineered models have advantages over xenografts for preclinical studies. Cancer Res. 66, 3355–3359 (2006).

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

Articles 6. Singh, M. & Johnson, L. Using genetically engineered mouse models of cancer to aid drug development: an industry perspective. Clin. Cancer Res. 12, 5312–5328 (2006). 7. Jackson, E.L. et al. The differential effects of mutant p53 alleles on advanced murine lung cancer. Cancer Res. 65, 10280–10288 (2005). 8. Aguirre, A.J. et al. Activated Kras and Ink4a/Arf deficiency cooperate to produce metastatic pancreatic ductal adenocarcinoma. Genes Dev. 17, 3112–3126 (2003). 9. Forbes, S.A. et al. The catalogue of somatic mutations in cancer (COSMIC). Curr. Protoc. Hum. Genet. 57, 10.11.1–10.11.26 (2008). 10. Laurent-Puig, P., Lievre, A. & Blons, H. Mutations and response to epidermal growth factor receptor inhibitors. Clin. Cancer Res. 15, 1133–1139 (2009). 11. Bokemeyer, C. et al. Fluorouracil, leucovorin, and oxaliplatin with and without cetuximab in the first-line treatment of metastatic colorectal cancer. J. Clin. Oncol. 27, 663–671 (2009). 12. Eberhard, D.A. et al. Mutations in the epidermal growth factor receptor and in KRAS are predictive and prognostic indicators in patients with non-small-cell lung cancer treated with chemotherapy alone and in combination with erlotinib. J. Clin. Oncol. 23, 5900–5909 (2005). 13. Tol, J. et al. Chemotherapy, bevacizumab, and cetuximab in metastatic colorectal cancer. N. Engl. J. Med. 360, 563–572 (2009). 14. Cappuzo, F. et al. SATURN: A double-blind, randomized, phase III study of maintenance erlotinib versus placebo following nonprogression with first-line platinum-based chemotherapy in patients with advanced NSCLC. J. Clin. Oncol. 27, 15s suppl, abstr 8001 (2009). 15. Jones, S. et al. Core signaling pathways in human pancreatic cancers revealed by global genomic analyses. Science 321, 1801–1806 (2008). 16. Troiani, T. et al. The use of xenograft models for the selection of cancer treatments with the EGFR as an example. Crit. Rev. Oncol. Hematol. 65, 200–211 (2008). 17. Kindler, H.L. et al. A double-blind, placebo-controlled, randomized phase III trial of gemcitabine (G) plus bevacizumab (B) versus gemcitabine plus placebo (P) in patients (pts) with advanced pancreatic cancer (PC): A preliminary analysis of Cancer and Leukemia Group B (CALGB 80303). J. Clin. Oncol. ASCO. Annu. Meet. Proc. 25 Part I, 4508 (2007). 18. Liang, W.C. et al. Cross-species vascular endothelial growth factor (VEGF)-blocking antibodies completely inhibit the growth of human tumor xenografts and measure the contribution of stromal VEGF. J. Biol. Chem. 281, 951–961 (2006). 19. Fuh, G. et al. Structure-function studies of two synthetic anti-vascular endothelial growth factor Fabs and comparison with the Avastin Fab. J. Biol. Chem. 281, 6625–6631 (2006). 20. Herbst, R.S. et al. Phase II study of efficacy and safety of bevacizumab in combination with chemotherapy or erlotinib compared with chemotherapy alone for treatment of recurrent or refractory non small-cell lung cancer. J. Clin. Oncol. 25, 4743–4750 (2007). 21. Tortora, G., Ciardiello, F. & Gasparini, G. Combined targeting of EGFR-dependent and VEGF-dependent pathways: rationale, preclinical studies and clinical applications. Nat. Clin. Pract. Oncol. 5, 521–530 (2008). 22. Hainsworth, J.D. & Herbst, R.S. A phase III, multicenter, placebo-controlled, doubleblind, randomized clinical trial to evaluate the efficacy of bevacizumab (Avastin) in combination with erlotinib (Tarceva) compared with erlotinib alone for treatment of advanced non-small cell lung cancer after failure of standard first-line chemotherapy (BeTA). J. Thoracic Oncol. 3, S302 (2008). 23. Miller, V.A., O’Conner, P., Soh, C. & Kabbinavar, F. A randomized, double-blind, placebo-controlled, phase IIIb trial (ATLAS) comparing bevacizumab therapy with or without erlotinib after completion of chemotherapy with B for first-line treatment of locally advanced, recurrent, or metastatic non-small cell lung cancer. J. Clin. Oncol. 27, 15s suppl, abstr LBA8002 (2009). 24. Van Cutsem, E. et al. Phase III trial of bevacizumab in combination with gemcitabine and erlotinib in patients with metastatic pancreatic cancer. J. Clin. Oncol. 27, 2231–2237 (2009). 25. Herbst, R.S. et al. Biomarker evaluation in the phase III, placebo-controlled, randomized BeTa trial of bevacizumab and erlotinib for patients with advanced non-small cell lung cancer after failure of standard 1st-line chemotherapy: correlation with treatment outcomes. in 100th Annual Meeting of the American Association for Cancer Research, vol. LB-131 (AACR, Denver, 2009). 26. Pazdur, R. Endpoints for assessing drug activity in clinical trials. Oncologist 13 Suppl 2, 19–21 (2008). 27. Alley, M.C., Hollingshead, M.G., Dykes, D.J. & Waud, W.R. Human tumour xenograft models in NCI drug development in Anticancer Drug Development Guide: Preclinical Screening, Clinical Trials, and Approval edn. 2 (eds. Teicher, B.A. & Andrews, P.A.) 125–152 (Humana Press Inc., 2004). 28. De Clerck, N.M. et al. High-resolution X-ray microtomography for the detection of lung tumors in living mice. Neoplasia 6, 374–379 (2004).

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

29. Cook, N., Olive, K.P., Frese, K. & Tuveson, D.A. K-Ras-driven pancreatic cancer mouse model for anticancer inhibitor analyses. Methods Enzymol. 439, 73–85 (2008). 30. Ding, L. et al. Somatic mutations affect key pathways in lung adenocarcinoma. Nature 455, 1069–1075 (2008). 31. Rozenblum, E. et al. Tumor-suppressive pathways in pancreatic carcinoma. Cancer Res. 57, 1731–1734 (1997). 32. Herbst, R.S. et al. TRIBUTE: a phase III trial of erlotinib hydrochloride (OSI-774) combined with carboplatin and paclitaxel chemotherapy in advanced non-small-cell lung cancer. J. Clin. Oncol. 23, 5892–5899 (2005). 33. Davies, A.M. et al. Pharmacodynamic separation of epidermal growth factor receptor tyrosine kinase inhibitors and chemotherapy in non-small-cell lung cancer. Clin. Lung Cancer 7, 385–388 (2006). 34. Brugger, W. et al. Biomarker analyses from the phase III placebo-controlled SATURN study of maintenance erlotinib following first-line chemotherapy for advanced NSCLC. J. Clin. Oncol. 27, 15s (2009). 35. Higgins, B. et al. Antitumor activity of erlotinib (OSI-774, Tarceva) alone or in combination in human non-small cell lung cancer tumor xenograft models. Anticancer Drugs 15, 503–512 (2004). 36. Sirotnak, F.M., Zakowski, M.F., Miller, V.A., Scher, H.I. & Kris, M.G. Efficacy of cytotoxic agents against human tumor xenografts is markedly enhanced by coadministration of ZD1839 (Iressa), an inhibitor of EGFR tyrosine kinase. Clin. Cancer Res. 6, 4885–4892 (2000). 37. Moore, M.J. et al. Erlotinib plus gemcitabine compared with gemcitabine alone in patients with advanced pancreatic cancer: a phase III trial of the National Cancer Institute of Canada Clinical Trials Group. J. Clin. Oncol. 25, 1960–1966 (2007). 38. Santos, G.C. et al. Molecular predictors of outcome in a phase III study of gemcitabine and erlotinib therapy in patients with advanced pancreatic cancer (NCIC CTG PA.3). Cancer (in the press). 39. Sandler, A. et al. Paclitaxel-carboplatin alone or with bevacizumab for non-small-cell lung cancer. N. Engl. J. Med. 355, 2542–2550 (2006). 40. Baselga, J. & Rosen, N. Determinants of RASistance to anti-epidermal growth factor receptor agents. J. Clin. Oncol. 26, 1582–1584 (2008). 41. Ciardiello, F. & Tortora, G. EGFR antagonists in cancer treatment. N. Engl. J. Med. 358, 1160–1174 (2008). 42. Laurent-Puig, P. & Taieb, J. Lessons from Tarceva in pancreatic cancer: where are we now, and how should future trials be designed in pancreatic cancer? Curr. Opin. Oncol. 20, 454–458 (2008). 43. Pirker, R. et al. Cetuximab plus chemotherapy in patients with advanced non-smallcell lung cancer (FLEX): an open-label randomised phase III trial. Lancet 373, 1525–1531 (2009). 44. Zhu, C.Q. et al. Role of KRAS and EGFR as biomarkers of response to erlotinib in National Cancer Institute of Canada Clinical Trials Group Study BR.21. J. Clin. Oncol. 26, 4268–4275 (2008). 45. Fujimoto, N. et al. High expression of ErbB family members and their ligands in lung adenocarcinomas that are sensitive to inhibition of epidermal growth factor receptor. Cancer Res. 65, 11478–11485 (2005). 46. Mitsudomi, T. & Yatabe, Y. Mutations of the epidermal growth factor receptor gene and related genes as determinants of epidermal growth factor receptor tyrosine kinase inhibitors sensitivity in lung cancer. Cancer Sci. 98, 1817–1824 (2007). 47. Yang, Y. et al. Phosphatidylinositol 3-kinase mediates bronchioalveolar stem cell expansion in mouse models of oncogenic K-ras-induced lung cancer. PLoS ONE 3, e2220 (2008). 48. Yang, Y. et al. A selective small molecule inhibitor of c-Met, PHA-665752, reverses lung premalignancy induced by mutant K-ras. Mol. Cancer Ther. 7, 952–960 (2008). 49. Kindler, H.L. et al. Phase II trial of bevacizumab plus gemcitabine in patients with advanced pancreatic cancer. J. Clin. Oncol. 23, 8033–8040 (2005). 50. Ko, A.H. et al. A phase II study evaluating bevacizumab in combination with fixeddose rate gemcitabine and low-dose cisplatin for metastatic pancreatic cancer: is an anti-VEGF strategy still applicable? Invest. New Drugs 26, 463–471 (2008). 51. Ruiz, M.I. et al. Combined assessment of EGFR pathway-related molecular markers and prognosis of NSCLC patients. Brit. J. Cancer 100, 145–152 (2008). 52. Yarden, Y. & Sliwkowski, M.X. Untangling the ErbB signalling network. Nat. Rev. Mol. Cell Biol. 2, 127–137 (2001). 53. Lim, E.H. et al. Using whole genome amplification (WGA) of low-volume biopsies to assess the prognostic role of EGFR, KRAS, p53, and CMET mutations in advanced-stage non-small cell lung cancer (NSCLC). J. Thorac. Oncol. 4, 12–21 (2009). 54. Marks, J.L. et al. Prognostic and therapeutic implications of EGFR and KRAS mutations in resected lung adenocarcinoma. J. Thorac. Oncol. 3, 111–116 (2008).

593

ONLINE METHODS

Mouse strains. We licensed mice from the following institutions: KrasLSL−G12D from Tyler Jacks (Massachusetts Institute of Technology), p53frt/frt from Exelixis, Inc., p16/p19fl/fl from Anton Berns (NKI, Netherlands) and Pdx1Cre from Andy Lowy (University of Ohio). Animals were housed and cared for according to guidelines from the Institutional Animal Care and Use Committee (IACUC) at Genentech, Inc. In vitro fertilization was used to generate large cohorts of mice with similar birth dates to facilitate these experiments.

© 2010 Nature America, Inc. All rights reserved.

Adeno-FLPe/IRES/Cre infection. KrasLSL−G12D; p53frt/frt mice were infected at 7–9 weeks of age with recombinant, replication-deficient adenovirus serotype 5 expressing the FLPe and Cre recombinases (Ad-FIC; Galapagos) at 5 × 106 infectious units (i.u.) per mouse. Ad-FIC:CaPi coprecipitates were prepared as described55. Mice were anesthetized with avertin and Ad-FIC:CaPi coprecipitates were delivered intranasally in two 31.25 μl instillations, one per nostril.

micro-imaging system employing a single-element ultrasound transducer with a 40 MHz center frequency, focal length of 6 mm, field of view (FOV) of 14.6 mm, an axial resolution of 40 μm, and a lateral resolution of 100 μm. At the final imaging time point, a subset of animals was imaged with a second transducer (30 MHz center frequency, focal length = 12.7 mm, FOV = 20 mm, axial resolution = 55 μm, lateral resolution = 115 μm) because the tumor extended beyond the FOV of the original transducer. During imaging, mice were anesthetized with 2% isoflurane and temperature was maintained at 37 °C with a heated imaging platform. Three-dimensional (3D) images of the pancreas were acquired with a motorized drive mechanism that traverses a third dimension (z) while acquiring two-dimensional (2D) B-mode images at regular spatial intervals (50 μm). The 3D image data set consisted of a FOV of 10×10×z mm (for second transducer: 17×17×z mm), where z ranged from 6 to 18 mm based on tumor length. Tumor volume estimates were made by defining a 3D region of interest (ROI) using the Visualsonics image analysis software package. Each mouse was imaged before study enrollment; the calculated tumor burden, body weights and gender were then used to randomize mice into treatment cohorts using a random number generator in JMP statistical discovery software package version 6.0.3 (SAS).

Treatment regimens and dosing. All dosing regimens were carried out according to IACUC guidelines. Study animals were monitored daily and body weights were measured at least twice weekly. Animals were censored for survival based either on mortality or pre-determined morbidity criteria for euthanasia (in consultation with the veterinary staff), which included hunching, ruffled fur, belabored breathing, low body temperature, lack of mobility and/or >20% body weight loss from the time of study start. Dosing regimens are depicted in Supplementary Figure 4. All chosen regimens were well tolerated in the GEMMs, with no significant body weight loss or overt signs of toxicity other than those attributable to the disease itself (Supplementary Fig. 4). Chemotherapeutics used: carboplatin (Bristol-Myers Squibb or generic from Pliva d.d.) was dosed at 25 mg/kg via intraperitoneal (i.p.) injection for the first 5 d of study (qdx5), paclitaxel (Natural Pharmaceuticals) was dosed at 30 mg/kg by intravenous (i.v.) injection every other day for five cycles (q2dx5) at study start, and gemcitabine (Eli Lilly) was dosed at either 100 mg/kg (GEMMs) or 160 mg/kg (BxPC3) q3dx4 via i.p. injection at study start; all chemotherapeutics were formulated as per the manufacturer’s guidelines. B20-4.1.1 was prepared and purified in-house as previously described19 and dosed at 5 mg/kg twice weekly by i.p. injection throughout the course of a study for GEMMs, starting at day 15 for A549 xenografts, or 5 mg/kg once weekly by i.p. injection starting on day 1 for BxPC3 xenografts. Erlotinib (OSI Pharmaceuticals) was formulated in either 0.5% methyl-cellulose/0.2%Tween-80 (MCT) or CAPTISOL (CyDex) and dosed daily (qd) at 100 mg/kg orally (p.o.), starting at day 1 (PDAC), day 6 (NSCLC) or day 15 (A549 xenografts) through the duration of a study. Both formulations were shown to have equivalent pharmacokinetic properties (data not shown).

Preclinical trial endpoints. We measured overall survival (OS), as assessed by mortality or by pre-determined, overt morbidity criteria for each model (see above). Progression-free survival (PFS) was calculated using noninvasive imaging endpoints similar to humans. Tumor progression in humans is defined by RECIST as a 20% or greater increase in the sum of the longest uni-dimensional tumor measurements based upon serial imaging time points throughout a clinical trial59. We collected imaging data at three sequential time points in each model for this purpose. All mice were imaged before study enrollment. NSCLC mice were imaged again on or around day 15 or day 28 post-study initiation, whereas PDAC mice received serial images on or around day 11 and day 21 or 28 (if alive). Consolidated imaging profiles for individual mice are shown in Supplementary Figures 8 and 9. To account for the rapid growth of mouse tumors, we defined disease progression as a doubling in tumor burden for each GEMM and used that criterion to build PFS Kaplan-Meier curves as shown in Figures 1–6. Control and select treatment cohorts were repeated across independent studies over a period of ~2 years and were reproducible (that is, no statistical differences). Hence, like cohorts were pooled in the OS and PFS analyses shown in figures to increase statis tical power. Repeated cohorts in this study are NSCLC controls (four repeats), A (4), CA (2), PDAC controls (3), G (3) and GA (2). Median survival times, P values calculated from log-rank tests, and hazard ratios are presented for each Kaplan-Meier plot to aid comparison with corresponding human data.

Noninvasive biomedical imaging. We used X-ray micro-CT for the NSCLC GEMM and high-resolution micro-ultrasound for the PDAC GEMM. MicroCT was performed as described in ref. 56. Briefly, serial lung imaging was performed with two in-vivo micro-CT systems (vivaCT 40 and vivaCT 75, Scanco Medical). Animals were randomized between systems and serial scans were performed on the same system that was used for baseline imaging. Data were acquired at 38 mm isotropic voxel size (vivaCT 75: 50 mm), 1,000 projections, 250 ms integration time (vivaCT 75: 200 ms), 45 keV photon energy and 177 mA current. During in vivo imaging, the animals were anesthetized with 2% isoflurane in air and kept at a constant 37 °C temperature by regulated warm airflow. The imaging time at each time point was ~25 min per animal and the estimated radiation dose was 0.2 Gy. Image data were evaluated using Analyze (AnalyzeDirect), an image analysis software package. The lung was viewed in the coronal plane to identify tumors. The largest cross-sectional plane of each tumor was determined by the user, from which estimates of maximal tumor diameter (d1) and the largest perpendicular diameter (d2) were determined by placing a ruler on the screen. The total tumor burden was determined to be the sum of the cross-product of the directional estimates (d1 × d2) of all tumors. This in vivo micro-CT tumor analysis method was validated on a set of 12 animals; the sum of the cross-products as determined from the in vivo data correlated strongly (R = 0.99) with the total tumor volume as determined by ex vivo micro-CT analysis. Pancreatic tumor volumes were estimated using in vivo high-resolution microultrasound imaging57,58. Imaging was performed with a Visualsonics Vevo 770

Statistical analyses. In each figure showing Kaplan-Meier survival curve estimates for GEMMs, a Cox proportional hazards model60 was fit to the data in the R statistical language61 using the ‘survival’ library62. For each such model, study baseline tumor burden (logarithmically scaled, with “1” added to each scan’s value to specify a baseline of “0” on the logarithmic scale) and treatment group were included for each mouse as explanatory factors. The endpoint for mice was survival time (overall survival, OS) or the minimum of survival and time to doubling in tumor size (progression-free survival, PFS). Ties in event times were resolved with Efron’s approximation63. All P values reported for hazard ratios are two-sided, and test the null hypothesis that the hazard ratio is 1 between the two groups specified. PDAC mice in the following groups were not imaged by ultrasound and hence were included in the OS analyses but excluded from PFS analyses and the imaging data in Supplementary Figure 9: 1 in control, 3 in gemcitabine, and 1 in gemcitabine plus anti-VEGF. For each of the imaging data sets shown in Supplementary Figures 8 and 9, the fold-change observed from the first scan for each mouse was logarithmically transformed. A one-way analysis of variance (ANOVA) was fit to these data and pair-wise treatment comparisons examined to determine whether differential treatment effects were evident. The Tukey-Kramer method64 was applied to correct for the multiple tests performed using the ‘multcomp’ package in R65. In Supplementary Figure 9c, regimen-specific growth rates in ultrasound tumor burden were estimated as the slope of the longitudinal growth curves within each regimen via a linear mixed-effects model66 implemented with the ‘nlme’ package in R67. Pair-wise differences in slopes across groups

nature biotechnology

doi:10.1038/nbt.1640

were evaluated by t-tests of contrasted slope estimates and significant differences between regimens noted. In this case, the Holm procedure68 was used to correct for multiple testing.

© 2010 Nature America, Inc. All rights reserved.

55. Fasbender, A. et al. Incorporation of adenovirus in calcium phosphate precipitates enhances gene transfer to airway epithelia in vitro and in vivo. J. Clin. Invest. 102, 184–193 (1998). 56. Caunt, M. et al. Blocking neuropilin-2 function inhibits tumor cell metastasis. Cancer Cell 13, 331–342 (2008). 57. Wirtzfeld, L.A. et al. A new three-dimensional ultrasound microimaging technology for preclinical studies using a transgenic prostate cancer mouse model. Cancer Res. 65, 6337–6345 (2005). 58. Graham, K.C. et al. Three-dimensional high-frequency ultrasound imaging for longitudinal evaluation of liver metastases in preclinical models. Cancer Res. 65, 5231–5237 (2005). 59. Therasse, P. et al. New guidelines to evaluate the response to treatment in solid tumors. European Organization for Research and Treatment of Cancer, National Cancer Institute of the United States, National Cancer Institute of Canada. J. Natl. Cancer Inst. 92, 205–216 (2000).

60. Collett, D. Modelling Survival Data in Medical Research (Chapman & Hall, London, 1994). 61. RDevelopmentCoreTeam. A Language and Environment for Statistical Computing (R.F.f.S. Computing, Vienna, 2008). 62. Therneau, T. & Lumley, T. in Survival: Survival Analysis, Including Penalised Likelihood. R Package Version 2.34–1 (R Foundation for Statistical Computing, Vienna, Austria; 2008). . 63. Efron, B. The efficiency of Cox’s likelihood function for censored data. J. Am. Stat. Assoc. 72, 557–565 (1977). 64. Hsu, J.C.. Multiple Comparisons (Chapman & Hall, London, 1996). 65. Hothorn, T., Bretz, F. & Westfall, P. Simultaneous inference in general parametric models. Biom. J. 50, 346–363 (2008). 66. Laird, N.M.J.H. Random effects models for longitudinal data. Biometrics 38, 963–974 (1982). 67. Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D. & The R Core Team. nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1–89 (R Foundation for Statistical Computing, Vienna, Austria; 2008). . 68. Holm, S. A simple sequentially rejective multiple test procedure. Scand. J. Stat. 6, 65–70 (1979).

doi:10.1038/nbt.1640

nature biotechnology

letters

Single-molecule enzyme-linked immunosorbent assay detects serum proteins at subfemtomolar concentrations

© 2010 Nature America, Inc. All rights reserved.

David M Rissin1,3, Cheuk W Kan1,3, Todd G Campbell1, Stuart C Howes1, David R Fournier1, Linan Song1, Tomasz Piech1, Purvish P Patel1, Lei Chang1, Andrew J Rivnak1, Evan P Ferrell1, Jeffrey D Randall1, Gail K Provuncher1, David R Walt2 & David C Duffy1 The ability to detect single protein molecules1,2 in blood could accelerate the discovery and use of more sensitive diagnostic biomarkers. To detect low-abundance proteins in blood, we captured them on microscopic beads decorated with specific antibodies and then labeled the immunocomplexes (one or zero labeled target protein molecules per bead) with an enzymatic reporter capable of generating a fluorescent product. After isolating the beads in 50-fl reaction chambers designed to hold only a single bead, we used fluorescence imaging to detect single protein molecules. Our single-molecule enzyme-linked immunosorbent assay (digital ELISA) approach detected as few as ~10–20 enzyme-labeled complexes in 100 ml of sample (~10−19 M) and routinely allowed detection of clinically relevant proteins in serum at concentrations (<10−15 M) much lower than conventional ELISA3–5. Digital ELISA detected prostate-specific antigen (PSA) in sera from patients who had undergone radical prostatectomy at concentrations as low as 14 fg/ml (0.4 fM). The clinical use of protein biomarkers to differentiate between healthy and disease states, and to monitor disease progression, requires the measurement of low concentrations of proteins in complex samples. Current immunoassays typically measure proteins at concentrations above 10−12 M6. The serum concentrations of the majority of proteins important in cancer7, neurological disorders8,9 and the early stages of infection10, however, are thought to range from 10−16 to 10−12 M. For instance, a 1-mm3 tumor composed of a million cells that each secrete 5,000 proteins into 5 liters of circulating blood translates to a concentration of ~2 × 10−15 M (or 2 fM). Moreover, serum from individuals recently infected with HIV contains 10–3,000 virions per ml, resulting in estimated concentrations of the p24 capsid antigen ranging from 50 × 10−18 M (50 aM) to 15 × 10−15 M (15 fM)10. Attempts to develop methods capable of measuring these concentrations of proteins have focused on the replication of nucleic acid labels on proteins11,12, or on measuring the bulk, ensemble properties of labeled protein molecules13–16. The work of Mirkin et al.12,17 and others18 using labels based on gold nanoparticles and DNA biobarcodes has pushed the detection of proteins into the low femtomolar range; a recent report

using this technology demonstrated the detection of 10 fM of PSA in serum17. Nonetheless, the sensitivities achieved by methods for detecting proteins still lag behind those for nucleic acids, such as PCR, limiting the number of gene products that have been detected in blood6,19. The isolation and detection of single protein molecules provides a promising approach for measuring extremely low concentrations of proteins1,2. For example, Todd et al.2 have developed flow-based methods for serially detecting single fluorescently labeled detection antibodies that have been released from immunocomplexes formed on solid substrates. Here we report an approach for detecting thousands of single protein molecules simultaneously using the same reagents as the gold standard for detecting proteins, namely, the ELISA. This method has been used to detect proteins in serum at subfemtomolar concentrations. Our approach makes use of arrays of femtoliter-sized reaction chambers (Fig. 1)—which we term single-molecule arrays (SiMoAs)—that can isolate and detect single enzyme molecules20–24. This approach builds from the work of Walt et al.20–23, who used these arrays to study the kinetics21 and inhibition20 of single enzymes. Our objective was to exploit the ability of SiMoAs to trap and detect single enzymes to detect single enzyme–labeled proteins. In the first step of this single-molecule immunoassay (Fig. 1a), a sandwich antibody complex is formed on microscopic beads (2.7 μm diameter), and the bound complexes are labeled with an enzyme, as in a conventional bead-based ELISA. When assaying samples containing extremely low concentrations of protein, the ratio of protein molecules (and the resulting enzyme-labeled complex) to beads is small (typically <1:1) and, as such, the percentage of beads that contain a labeled immuno complex follows a Poisson distribution. At low concentrations of protein, the Poisson distribution indicates that beads carry either a single immunocomplex or none. For example, if 50 aM of a protein in 0.1 ml (3,000 molecules) is captured and labeled on 200,000 beads, then 1.5% of the beads will carry one protein molecule and 98.5% will not carry any protein molecules (Fig. 1b)22. It is not possible to detect these low numbers of enzyme labels using standard detection techno logy (for example, a plate reader), because the fluorophores generated by each enzyme diffuse into a large assay volume (typically 0.1–1 ml), and it takes hundreds of thousands of enzyme labels to generate a

1Quanterix

Corporation, Cambridge, Massachusetts, USA. 2Department of Chemistry, Tufts University, Medford, Massachusetts, USA. 3These authors contributed equally to this work. Correspondence should be addressed to D.C.D. ([email protected]). Received 1 February; accepted 29 April; published online 23 May 2010; doi:10.1038/nbt.1641

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

595

letters

© 2010 Nature America, Inc. All rights reserved.

Figure 1 Digital ELISA based on arrays of femtoliter-sized wells. (a,b) Single protein molecules are captured and labeled on beads using standard ELISA reagents (a), and beads with or without a labeled immunoconjugate are loaded into femtoliter-volume well arrays for isolation and detection of single molecules by fluorescence imaging (b). (c) Scanning electron micrograph of a small section of a femtoliter-volume well array after bead loading. Beads (2.7 μm diameter) were loaded into an array of wells with diameters of 4.5 μm and depths of 3.25 μm. (d) Fluorescence image of a small section of the femtoliter-volume well array after signals from single enzymes are generated. Whereas the majority of femtolitervolume chambers contain a bead from the assay, only a fraction of those beads possess catalytic enzyme activity, indicating a single, bound protein molecule. The concentration of protein in bulk solution is correlated to the percentage of beads that carry a protein molecule.

a

1. 2. 3. 10 μm

b 1. Enzyme substrate 2. Seal silicone gasket 3. Fluorescence imaging

Thousands of individual beads loaded into wells

d

100 μm Magnetic beads

Capture antibody

fluorescence signal above background. In contrast, SiMoAs permit the detection of very low concentrations of enzyme labels by confining the fluorophores generated by individual enzymes to extremely small volumes (~50 fl), ensuring a high local concentration of fluorescent product molecules. To achieve this confinement in our assay, beads are loaded into an array of femtoliter-sized wells; we used 2-mm-wide arrays with ~50,000 wells, each with a diameter of 4.5 μm and a depth of 3.25 μm (Fig. 1c). After sealing the loaded arrays against a rubber gasket in the presence of a droplet of fluorogenic enzyme substrate, each bead is isolated in a femtoliter-volume reaction chamber. Beads possessing a single enzyme–labeled immunocomplex generate a high concentration of fluorescent product that is restricted to the 50-fl reaction chamber (Fig. 1d). By acquiring time-lapsed fluorescence images of the array using standard microscope optics, it is possible to distinguish beads associated with a single enzyme molecule (“on” well) from those not associated with an enzyme (“off ” well); Supplementary Figure 1 shows histograms of fluorescence from “on” and “off ” wells. Imaging the arrays allows simultaneous detection of tens to tens of thousands of single immunocomplexes. The protein concentration in the test sample is determined by counting the number of wells containing both a bead and fluorescent product relative to the total number of wells containing beads (Fig. 1d). As SiMoAs enable concentration to be determined digitally rather than by using the total analog signal, we call our approach to detecting single immunocomplexes digital ELISA. We first assessed the intrinsic sensitivity of this strategy by creating populations of beads with well-characterized enzyme-to-bead ratios. We mixed 400,000 biotin-modified beads with a range of concentrations of the enzyme conjugate streptavidin-β-galactosidase (SβG). For convenience, biotinylated beads were provided by hybridizing biotinylated DNA with beads functionalized with complementary DNA. (We note that this experiment should not be construed as a sensitive DNA assay; the sensitivity of such an assay is limited by nonspecific interactions between the enzyme conjugate and surfacebound DNA as shown in Supplementary Fig. 2.) These beads were detected in two different ways. First, we assayed an ensemble of beads in 100 μl using a fluorescence plate reader after 1 h incubation with 100 μM resorufin-β-d-galactopyranoside (RGP), a fluorogenic substrate for β-galactosidase. The detection limit for enzyme on the plate reader was 15 fM of SβG (Fig. 2). Second, we loaded the beads into femtoliter-volume well arrays and, after sealing a solution of RGP 596

c

Formation of single immunocomplexes on beads

Detection antibody with biotin tag

Antigen

Streptavidin-βgalactosidase

into the wells of the array, allowed the signal from single enzymes to accumulate in the reaction chambers for 2.5 min, acquiring fluorescent images every 30 s. At the end of the experiment we used an image of the array that was acquired in white light to identify wells that contained beads (bead-containing wells scatter light differently than empty wells). The fluorescent images were used to determine which of those beads had an associated bound enzyme (from increasing intensity in time-lapsed fluorescent images). Figure 2 shows a log-log plot of the percentage of beads that contained an enzyme as a function of bulk SβG concentration. The lowest concentration of enzyme conjugate detected was 350 zeptomolar (zM) and the calculated limit of detection (LOD)—determined by extrapolating the enzyme concentration at a signal equal to background plus 3 s.d. of the background signal—was 220 zM. The sensitivity of SiMoAs to intrinsic label was, therefore, ~10–20 enzymes in 100 μl, corresponding to an increase in sensitivity over ensemble measurements using a typical ELISA-plate reader of a factor of about 68,000. For comparison, chemiluminescent detection of alkaline phosphatase4, a highly sensitive enzyme reporter system widely used in clinical diagnostics, has an LOD of about 30 aM (see ). The high sensitivity that arises from the thermodynamic and kinetic efficiency of the capture and detection processes in our approach are discussed below and detailed in Supplementary Table 1. The linear dynamic range of digital detection of enzyme labels by SiMoAs is determined by the ability to distinguish “on” and “off ” wells. At ratios of enzyme to beads of less than ~1:10, Poisson statistics show that the only statistically significant populations of beads are those carrying either one enzyme or no enzyme. Single enzymes can be detected provided that sufficient beads are interrogated and the number of active beads rises above the Poisson noise of counting active beads. At ratios of enzyme to beads greater than ~1:10, the fraction of active beads becomes much higher, and Poisson statistics show that there are a significant number of beads with multiple enzymes. To quantify the number of detected enzymes and maintain linearity in the subpopulations of beads with multiple enzymes, we use Poisson statistics to convert the number of active beads to the number of detected enzymes (see Supplementary Table 1). As the percentage of active beads approaches 50% (ratios of enzyme to beads greater than ~1:1.5), however, distinguishing “on” and “off ” wells using image analysis software becomes challenging, and we reach a practical upper limit of the digital dynamic range. For example, the VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters Figure 2 Digitization of enzyme-linked complexes greatly increases sensitivity Average Average 3.4 SiMoA Technology 4 Measurement Poisson compared with bulk, ensemble measurements. [SβG] (aM) number of fraction of Noise CV Plate Reader active beads active beads CV (a) Log-log plot of signal output (% active 0 1 0.0016% 87% 122% 2.9 2 beads for single-molecule array (SiMoA) or 0.35 3 0.0086% 75% 55% relative fluorescence units (r.f.u.) for plate 0.7 5 0.0099% 63% 46% 2.4 3.5 22 0.0413% 10% 21% reader) as a function of the concentration of 0 7 38 0.0713% 15% 16% streptavidin-β-galactosidase (SβG) captured on 1.9 35 237 0.4461% 1% 7% biotinylated beads. SβG concentrations for –2 70 385 0.8183% 5% 5% the ensemble readout ranged from 3 fM to 350 1787 3.3802% 2% 2% 700 4036 7.5865% 5% 2% 300 fM, with a detection limit of 15 × 10−15 M 3500 15634 30.6479% 3% 1% (15 fM; green broken line). For the SiMoA [Streptavidin-β-galactosidase] M 7000 24836 44.5296% 1% 1% assay, SβG concentrations ranged from 350 zM to 7 fM, demonstrating a linear response of ~10,000-fold, with a calculated detection limit of 220 × 10 −21 M (220 zM; red broken line). Error bars are based on the s.d. over three replicates for both technologies. LODs were determined by extrapolating the concentration from the signal equal to background signal plus 3 s.d. of the background signal. (b) The imprecision from SiMoAs is determined by the Poisson noise of counting single events. The intrinsic variation (Poisson noise) of counting single active beads is given by √n. Comparing the Poisson noise–associated coefficient of variation (%CV = √n/n) with the SiMoA %CV over three measurements confirmed that the imprecision of the assay is determined by counting error.

a

b

2

3

–1

–1

10

10

10

4

5

–1

6

–1

10

–1

7

10

10

8

–1

–1

10

1% active beads (see below), this loading results in a background signal of 200–300 active beads detected, corresponding to an acceptable coefficient of variation (CV) from Poisson noise of 6–7%. Third, excessive bead concentrations can lead to both increases in non specific binding that reduces signal-to-background and low ratios of analyte to beads that can result in high CVs from Poisson noise. The balance of these factors means that 200,000 to 1,000,000 beads per 100 μl of test sample is optimal for digital ELISA. The concentrations of detection antibody and enzyme conjugate were also minimized to yield the acceptable background signal (1%) and Poisson noise (Supplementary Discussion). Figure 3 shows data from digital ELISAs for PSA and TNF-α. The human forms of the proteins were spiked into 25% bovine serum to final concentrations representative of clinical test samples. A fourfold dilution factor is typically used to reduce matrix effects in immunoassays4. Using digital ELISA to detect PSA in 25% serum, we obtained an LOD of ~50 aM (1.5 fg/ml), which equates to an LOD in whole serum of ~200 aM (6 fg/ml). The lowest concentration tested and detected was 250 aM in 25% serum, corresponding to 1 fM in whole serum. As LOD is determined by extrapolating the concentration at background plus 3 s.d. of the background, LODs for different runs are dependent on the CV of the background. We obtained subfemtomolar LODs of PSA in whole serum over several experiments with typical background variances. For comparison, a leading commercial PSA assay (ADVIA Centaur, Siemens) reports an LOD of 3 pM (0.1 ng/ml) in human serum, and ultrasensitive assays have been reported with LODs in the range

b 1.0 0.5 0 –16

–15

–14

log [PSA] (M)

–13

2 r = 0.995

3

log (% active beads)

% active beads

1.5

2 1 0

1

2 [PSA] (fM)

3

2

1.5 % active beads

a log (% active beads)

© 2010 Nature America, Inc. All rights reserved.

10

–1

9

SiMoA log (% active)

Plate Reader log (r.f.u.)

signal at 7 fM (~45% active) in Figure 2 deviates from linearity. As a result, the digital, linear dynamic range demonstrated here using 50,000 wells spanned approximately four orders of magnitude (from 3.5 fM down to 350 zM). Provided that proteins are labeled using appropriate enzyme concentrations (see below), this dynamic range is sufficient for many clinical applications. We have developed digital ELISAs for two clinically relevant proteins—PSA and tumor necrosis factor-α (TNF-α)—to determine the sensitivity of the approach for detecting proteins in blood. The critical parameters in developing these assays were the concentrations of the beads and the two labeling reagents (detection antibody and enzyme conjugate). The choice of bead concentration depends on several competing factors. First, a sufficient number of beads must be present to capture most of the target analyte from thermodynamic and kinetic perspectives. Thermodynamically, 200,000 beads in 100 μl, each of which has ~80,000 antibodies25 bound to it, equates to an antibody concentration of about 0.3 nM. The antibody-protein equilibrium at that concentration permits a high capture efficiency (>70%) (D.M.R., E.P.F., D.C.D., unpublished work). Kinetically, for 200,000 beads dispersed in 100 μl, the average distance between beads is about 80 μm. Proteins the size of TNF-α and PSA (17.3 and 30 kDa, respectively) will diffuse 80 μm in <1 min, suggesting that capture of the protein molecules will not be limited kinetically over a 2-h incubation. Second, a sufficient number of beads must be present to be loaded onto the arrays to limit Poisson noise. Loading 200,000 beads into 50,000-well arrays typically results in 20,000–30,000 beads being trapped in femtoliter-sized wells. For a typical background signal of

1.0 0.5 0

2 r = 0 .977

1

–0.5 –15

–14

log [TNF-α] (M)

–13

0

1

2

3

[TNF-α] (fM)

Figure 3 Subfemtomolar detection of proteins in serum using digital ELISA. (a,b) Changes in the percentage of active beads with changes in analyte concentration for human prostate-specific antigen (PSA) spiked into 25% serum (a) and human tumor necrosis factor-α (TNF-α) spiked into 25% serum (b). The concentrations plotted on the x axes refer to the final concentration of spiked protein in the diluted sample. The plots on the left-hand side show the assay response over the concentration range tested in log-log space. The plots on the right-hand side show the assay response in the femtomolar range in linear-linear space to illustrate the limit of detection (LODs) and linearity of response. LODs were determined by extrapolating the concentration from the signal equal to background signal plus 3 s.d. of the background signal. Broken lines, signal at the LOD. Error bars, s.d. over three replicates.

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

597

letters

[PSA] (pg/mL)

© 2010 Nature America, Inc. All rights reserved.

of 10–30 fM17,26. The detection limit determined from the TNF-α digital ELISA was ~150 aM (2.5 fg/ml), corresponding to ~600 aM (10 fg/ml) in whole serum; the highest sensitivity commercially available ELISA for TNF-α has an LOD of 21 fM (0.34 pg/ml) in serum (Supplementary Fig. 3). The zero concentration spike of target protein for both assays provides a useful negative control for these experiments: 25% serum contains millimolar concentrations of a wide variety of proteins, and very low concentrations of target protein can be detected above these high-protein backgrounds. We also developed a proof-of-principle digital assay for DNA to show that low concentrations of nucleic acids can be detected without replication of the target (Supplementary Fig. 2). The ability of digital ELISA to measure much lower concentrations of proteins than conventional ELISA derives from two effects: (i) the high sensitivity of SiMoAs to enzyme label and (ii) the low background signals that can be achieved by digitizing the detection of proteins. The sensitivity of any immunoassay is determined by the sensitivity of the detection technology to the label, the antibody affinity, the assay background and the variance (%CV) of the background measurement27. The subattomolar sensitivity of SiMoAs to enzyme labels (Fig. 2) enables digital ELISA to detect subfemtomolar concentrations of labeled proteins. That said, for antibodies of given affinity, the sensitivity of the immunoassay will be determined by the assay background, and the high label sensitivity of SiMoAs helps reduce this background. Control experiments have shown that backgrounds in digital ELISA arise from nonspecific binding of detection antibody and enzyme conjugate to the capture bead surface (Supplementary Table 2). As SiMoAs provide superior label sensitivity over conventional assays, substantially less detection antibody (~1 nM) and enzyme conjugate (1−50 pM) are needed to detect binding events as compared with conventional assays (labeling reagent concentrations ~10 nM). The decreased label concentration reduces nonspecific binding to the capture surface, resulting in much lower background signals. For example, in the TNF-α and PSA digital ELISAs, the levels of nonspecific binding were equivalent to the signal produced by 1.8 fM and 1.2 fM of target protein, respectively (Fig. 3). The highest sensitivity commercial TNF-α assay has a level of nonspecific binding equivalent to 85 fM of TNF-α (Supplementary Fig. 3), which is 50 times more than we observe using digital ELISA. The ability to reduce backgrounds in digital ELISA by lowering the concentration of labeling reagents enables more sensitive immunoassays than are possible using conventional assays. 598

[PSA] (M)

10,000 Figure 4 Digital detection of prostate-specific –10 10 antigen (PSA) in serum samples of patients Patient [PSA] Dose Patient [PSA] Dose Patient [PSA] Dose 1,000 who had undergone radical prostatectomy. ID (pg/mL) CV ID (pg/mL) CV ID (pg/mL) CV –11 10 Concentrations of PSA in serum samples from S600 9.39 6% S590 0.22 22% S580 0.30 4% 100 radical prostatectomy patients ( ), healthy –12 S599 0.75 10% S589 0.85 17% S579 1.22 15% 10 control samples ( ) and Bio-Rad PSA control S598 2.71 12% S588 2.33 3% S578 0.090 91% 10 –13 samples ( ) determined using digital ELISA. S597 1.79 12% S587 1.06 13% S577 1.92 6% 10 Radical prostatectomy patient samples 1.0 S596 2.46 17% S586 1.29 22% S576 0.014 286% –14 10 S595 0.32 21% S585 0.49 84% S575 0.79 63% (SeraCare Life Sciences) all had undetectable 0.1 S594 1.63 15% S584 0.056 136% S574 1.62 20% PSA levels as measured by a leading clinical –15 10 S593 1.15 12% S583 1.33 26% S573 0.22 32% diagnostic assay (ADVIA Centaur); the green 0.01 S592 3.46 9% S582 4.76 9% S572 1.04 20% broken line represents the detection limit of the –16 10 S591 0.21 25% S581 1.57 31% S571 0.24 21% ADVIA Centaur PSA assay (100 pg/ml or 3 pM). 0.001 All 30 patient samples were above the detection Patient Population limit of the PSA digital ELISA, shown by the red broken line (0.006 pg/ml or ~200 aM), with the lowest patient PSA concentrations measured at 0.014 pg/ml (~400 aM) using digital ELISA. Patient samples with the lowest PSA levels approached the LOD of the assay, resulting in a large imprecision in the concentration determined (high dose %CV). The digital ELISA was validated for specificity to PSA using control standards (Bio-Rad) and serum from healthy individuals (ProMedDx) that had been assayed using the ADVIA Centaur PSA assay (Supplementary Table 3).

To demonstrate the possible diagnostic value of detecting very low concentrations of proteins in blood using digital ELISA, we measured PSA in serum samples from patients who had undergone radical prostatectomy. Levels of serum PSA are used both to screen for prostate cancer and to monitor recurrence of the disease in patients who have undergone radical prostatectomy28. After radical prostatectomy the vast majority of PSA is eliminated, and concentrations fall below the detection limit of standard commercial assays (3 pM or 0.1 ng/ml). Although regular monitoring of these patients for increases in PSA concentrations can detect recurrence of prostate cancer, several years may pass after surgery for biochemical recurrence to be detected by current immunoanalyzers. The ability to accurately quantify PSA levels in patients who have undergone prostatectomy at very low concentrations (<300 fM or 10 pg/ml) should provide early indication of recurrence if PSA levels increase17,29. Figure 4 shows PSA concentrations measured using digital ELISA in the serum of 30 patients (age 60–89) who had undergone radical prostatectomy and whose blood was collected at least 6 weeks after surgery. The PSA concentrations in the sera of all 30 patients were below the detection limit of commercial assays. Whole-serum samples were diluted 1:4 in buffer and measured using the digital ELISA specific for PSA (Fig. 3a). PSA was successfully detected in all 30 patients using digital ELISA, with concentrations ranging from 14 fg/ml to 9.4 pg/ml, with an average of 1.5 pg/ml. Further clinical studies are required to establish the diagnostic benefit of measuring PSA at femtogram per milliliter amounts in patients following radical prostatectomy. By isolating and detecting single immunocomplexes in arrays of femtoliter-volume wells, digital ELISA enables clinically important proteins in serum to be measured at subfemtomolar concentrations. We believe that the improvement in sensitivity shown by digital ELISA over previously reported approaches will translate into diagnostic benefits. For example, PSA concentrations in 9 of the 30 radical prostatectomy samples assayed in this work fell below the LOD of the previously highest sensitivity PSA assay based on nanoparticle labels17. An attractive feature of this approach is the ability to directly use reagents developed for standard ELISA for substantially more sensitive assays. We continue to improve the SiMoA technology in two key areas. First, based on the sensitivity to enzyme label (Fig. 2), it seems that the sensitivity of protein detection could be increased at least 100fold if nonspecific interactions that cause background signals could be minimized. The ability to isolate and interrogate single molecules on individual beads provides avenues for distinguishing antibody-antigen VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters binding events from nonspecifically bound complexes. Second, we are simplifying the logistics of the assays. Even in its present form, however, we believe that digital ELISA has the potential to facilitate earlier diagnosis and treatment of disease. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Note: Supplementary information is available on the Nature Biotechnology website.

© 2010 Nature America, Inc. All rights reserved.

Acknowledgments The project described was supported by Award Number R43CA133987 from the National Cancer Institute. Author Contributions D.M.R., C.W.K., D.R.F., D.R.W. and D.C.D. conceived the approach. D.R.F. built the imaging system. D.M.R., C.W.K., T.G.C., S.C.H., L.S., P.P.P., A.J.R., E.P.F., J.D.R. and G.K.P. conducted the experiments. T.P. wrote the image analysis software. L.C. prepared reagents. D.M.R. and D.C.D. wrote the manuscript. All authors were involved in designing experiments, reviewing and discussing data, and commenting on the manuscript. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturebiotechnology/. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Tessler, L.A., Reifenberger, J.G. & Mitra, R.D. Protein quantification in complex mixtures by solid phase single-molecule counting. Anal. Chem. 81, 7141–7148 (2009). 2. Todd, J. et al. Ultrasensitive flow-based immunoassays using single-molecule counting. Clin. Chem. 53, 1990–1995 (2007). 3. Gosling, J.P.A. Decade of development in immunoassay methodology. Clin. Chem. 36, 1408–1427 (1990). 4. Wild, D. The Immunoassay Handbook 3rd Edn. (Elsevier, 2005). 5. Zhang, H.Q., Zhao, Q., Li, X.F. & Le, X.C. Ultrasensitive assays for proteins. Analyst (Lond.) 132, 724–737 (2007). 6. Giljohann, D.A. & Mirkin, C.A. Drivers of biodiagnostic development. Nature 462, 461–464 (2009). 7. Srinivas, P.R., Kramer, B.S. & Srivastava, S. Trends in biomarker research for cancer detection. Lancet Oncol. 2, 698–704 (2001). 8. Galasko, D. Biomarkers for Alzheimer’s disease—clinical needs and application. J. Alzheimers Dis. 8, 339–346 (2005).

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

9. de Jong, D., Kremer, B.P.H., Olde Rikkert, M.G.M. & Verbeek, M.M. Current state and future directions of neurochemical biomarkers for Alzheimer’s disease. Clin. Chem. Lab. Med. 45, 1421–1434 (2007). 10. Barletta, J.M., Edelman, D.C. & Constantine, N.T. Lowering the detection limits of HIV-1 viral load using real-time immuno-PCR for HIV-1 p24 antigen. Am. J. Clin. Pathol. 122, 20–27 (2004). 11. Adler, M., Wacker, R. & Niemeyer, C.M. Sensitivity by combination: immuno-PCR and related technologies. Analyst (Lond.) 133, 702–718 (2008). 12. Nam, J.M., Thaxton, C.S. & Mirkin, C.A. Nanoparticle-based bio-bar codes for the ultrasensitive detection of proteins. Science 301, 1884–1886 (2003). 13. Armani, A.M., Kulkarni, R.P., Fraser, S.E., Flagan, R.C. & Vahala, K.J. Label-free, single-molecule detection with optical microcavities. Science 317, 783–787 (2007). 14. Cui, Y., Wei, Q.Q., Park, H.K. & Lieber, C.M. Nanowire nanosensors for highly sensitive and selective detection of biological and chemical species. Science 293, 1289–1292 (2001). 15. Gaster, R.S. et al. Matrix-insensitive protein assays push the limits of biosensors in medicine. Nat. Med. 15, 1327–1332 (2009). 16. Fan, R. et al. Integrated barcode chips for rapid, multiplexed analysis of proteins in microliter quantities of blood. Nat. Biotechnol. 26, 1373–1378 (2008). 17. Thaxton, C.S. et al. Nanoparticle-based bio-barcode assay redefines “undetectable” PSA and biochemical recurrence after radical prostatectomy. Proc. Natl. Acad. Sci. USA 106, 18437–18442 (2009). 18. Bao, Y.P. et al. Detection of protein analytes via nanoparticle-based bio bar code technology. Anal. Chem. 78, 2055–2059 (2006). 19. Anderson, N.L. & Anderson, N.G. The human plasma proteome: history, character, and diagnostic prospects. Mol. Cell. Proteomics 1, 845–867 (2002). 20. Gorris, H.H., Rissin, D.M. & Walt, D.R. Stochastic inhibitor release and binding from single-enzyme molecules. Proc. Natl. Acad. Sci. USA 104, 17680–17685 (2007). 21. Rissin, D.M., Gorris, H.H. & Walt, D.R. Distinct and long-lived activity states of single enzyme molecules. J. Am. Chem. Soc. 130, 5349–5353 (2008). 22. Rissin, D.M. & Walt, D.R. Digital readout of target binding with attomole detection limits via enzyme amplification in femtoliter arrays. J. Am. Chem. Soc. 128, 6286–6287 (2006). 23. Rissin, D.M. & Walt, D.R. Digital concentration readout of single enzyme molecules using femtoliter arrays and Poisson statistics. Nano Lett. 6, 520–523 (2006). 24. Tan, W. & Yeung, E.S. Monitoring the reactions of single enzyme molecules and single metal ions. Anal. Chem. 69, 4242–4248 (1997). 25. Rissin, D.M. & Walt, D.R. Duplexed sandwich immunoassays on a fiber-optic microarray. Anal. Chim. Acta 564, 34–39 (2006). 26. Ferguson, R., Yu, H., Kalyvas, M., Zammit, S. & Diamandis, E. Ultrasensitive detection of prostate-specific antigen by a time-resolved immunofluorometric assay and the Immulite immunochemiluminescent third-generation assay: potential applications in prostate and breast cancers. Clin. Chem. 42, 675–684 (1996). 27. Jackson, T.M. & Ekins, R.P. Theoretical limitations on immunoassay sensitivity. Current practice and potential advantages of fluorescent Eu3+ chelates as nonradioisotopic tracers. J. Immunol. Methods 87, 13–20 (1986). 28. Bock, J.L. & Klee, G.G. How sensitive is a prostate-specific antigen measurement? How sensitive does it need to be? Arch. Pathol. Lab. Med. 128, 341–343 (2004). 29. Trock, B.J. et al. Prostate cancer-specific survival following salvage radiotherapy vs observation in men with biochemical recurrence after radical prostatectomy. Jama-Journal of the American Medical Association 299, 2760–2769 (2008).

599

ONLINE METHODS

© 2010 Nature America, Inc. All rights reserved.

Materials. Optical fiber bundles were purchased from Schott North America. Nonreinforced gloss silicone sheeting was obtained from Specialty Manufacturing. Hydrochloric acid, anhydrous ethanol and molecular biology– grade Tween-20 were all from Sigma-Aldrich. Carboxyl-terminated magnetic beads (2.7-μm diameter) were purchased from Varian. Monoclonal capture antibody to human TNF-α, polyclonal detection antibody to human TNF-α and recombinant human TNF-α were purchased from R&D Systems. Monoclonal capture antibody to PSA, monoclonal detection antibody to PSA and purified PSA were purchased from BiosPacific; the detection antibody was biotinylated using standard methods. 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride, N-hydroxysulfosuccinimide and SuperBlock T-20 Blocking Buffer were purchased from Thermo Scientific. Purified DNA was purchased from Integrated DNA Technologies. SβG was conjugated in the laboratory using standard protocols. RGP was purchased from Invitrogen. The fiber polisher and polishing consumables were purchased from Allied High Tech Products. Preparation of magnetic beads presenting biotin-labeled DNA and capture of enzyme conjugate (Fig. 2). Beads functionalized with DNA capture probe (5′-NH2/C12-GTT GTC AAG ATG CTA CCG TTC AGA G-3′) were prepared according to the manufacturer’s instructions. These beads were incubated with 1 μM of biotinylated complementary DNA (5′-biotin-C TCT GAA CGG TAG CAT CTT GAC AAC-3′) overnight (16 h) in TE buffer containing 0.5 M NaCl and 0.01% Tween-20. After incubation the beads were washed three times in PBS buffer containing 0.1% Tween-20. The bead stock was distributed into a microtiter plate giving 400,000 beads per well in 100 μl. The buffer was aspirated from the microtiter plate wells; the beads were resuspended and incubated with various concentrations of SβG in Superblock containing 0.05% Tween-20 for 4 h. The beads were then separated and washed six times with 5× PBS buffer containing 0.1% Tween-20. For detection of enzyme, the beads were either (i) resuspended in 20 μl of PBS containing 0.1% Tween-20, and 10-μl aliquots were loaded onto two femtoliter-volume well arrays for SiMoA detection; or (ii) resuspended in 100 μl of 100 μM RGP in PBS, incubated for 1 h at 23 °C and read on a fluorescence plate reader (Infinite M200, Tecan). Capture of proteins on magnetic beads and formation of enzyme-labeled immunocomplex (Figs. 3 and 4). Beads functionalized with an antibody to the target protein were prepared according to the manufacturer’s instructions. Test solutions (100 μl) containing the protein of interest were incubated with suspensions of 200,000 magnetic beads for 2 h at 23 °C. The beads were then separated and washed three times in PBS and 0.1% Tween-20. The beads were resuspended and incubated with solutions containing detection antibody (typically ~1 nM) for 45 min at 23 °C. The beads were then separated and washed three times in PBS and 0.1% Tween-20. The beads were incubated with solutions containing SβG (1–50 pM) for 30 min at 23 °C, separated and washed six times in PBS and 0.1% Tween-20. The beads were then resuspended in 10 μl of PBS and loaded onto a femtoliter-volume well array. The total time of the assay was ~6 h. Capture of DNA on magnetic beads and formation of enzyme-labeled complex (Supplementary Fig. 2). A total of 200,000 beads functionalized with DNA capture probe were incubated with 100-μl solutions containing the target DNA (5′-TT GAC GGC GAA GAC CTG GAT GTA TTG CTC C TCT GAA CGG TAG CAT CTT GAC AAC-3′) for 2 h. After incubation, the DNA target solution was removed and the beads were washed three times in 0.2× SSC buffer containing 0.1% Tween-20. The beads were then resuspended and mixed with 10 nM biotinylated signal probe (5′-TAC ATC CAG GTC TTC GCC GTC AA/Biotin/-3′) that is also specific to the target DNA for 1 h. The beads were then washed three times in 0.2× SSC buffer containing 0.1% Tween-20 after removing the signal probe. A solution of 10 pM SβG was then added to the bead pellet, resuspended and mixed for 1 h. The beads were separated and washed six times in 5× PBS buffer containing 0.1% Tween-20. The beads were then resuspended in 10 μl of PBS and loaded onto a femtolitervolume well array.

nature biotechnology

Preparation of femtoliter-volume well arrays. Optical fiber bundles approximately 5 cm long were sequentially polished on a polishing machine using 30-, 9- and 1-μm-sized diamond lapping films. The polished fiber bundles were chemically etched in a 0.025 M HCl solution for 130 s and then immediately submerged into water to quench the reaction. The etched fibers were sonicated for 5 s in water, washed in water for 5 min and dried under vacuum. The differential etch rate of the core and cladding glass of the fiber bundle arrays caused 4.5-μm-diameter wells to be formed in the core fibers30. Different etch times resulting in different well depths were initially investigated. If wells were too deep, then multiple beads were deposited in each well and sealing was disrupted; if wells were too shallow, then the beads were not retained in the wells and poor loading efficiencies were observed. Well depths of 3.25 ± 0.5 μm were optimal for retaining single beads in wells while maintaining good seals. Loading of beads into femtoliter-volume well arrays. A short length of PVC tubing was placed on the etched end of a fiber bundle to create a reservoir to hold the bead solution. Ten microliters of the concentrated bead solution were pipetted into this reservoir. The fiber bundle was then centrifuged at 1,300g for 10 min to force the beads into the etched wells. The PVC tubing was removed after centrifugation. The fiber bundle was dipped in PBS solution to wash off excess bead solution, and the surface was swabbed with deionized water. In addition to well depth (see above), bead concentration was an important parameter for maximizing bead loading efficiencies. Above concentrations of 200,000 beads per 10 μl loaded, typically 40–60% of wells in a 50,000-well array were occupied by a single bead, resulting in percentage active beads with acceptable Poisson noise. At concentrations below 200,000 beads per 10 μl loaded, bead loading efficiency dropped, resulting in fewer active beads and higher Poisson noise. In these experiments, therefore, at least 200,000 beads per reaction were used and loaded onto the arrays. Detection of beads and enzyme-labeled beads in femtoliter-volume well arrays. A custom-built imaging system containing a mercury light source, filter cubes, objectives and a CCD camera was used for acquiring fluorescence images21,23. Fiber bundles were mounted on the microscope stage using a custom fixture. A droplet of β-galactosidase substrate (RGP) was placed on the silicone gasket material and placed in contact with the well arrays. A precision mechanical platform moved the silicone sheet into contact with the end of the fiber bundle, creating an array of isolated femtoliter-volume reaction vessels. Fluorescence images were acquired (558 nm excitation; 577 nm emission) with an exposure time of 1,011 ms. Five frames (at 30-s intervals) were taken for each femtoliter-volume well array. The product of the enzymatic reaction used in these studies—resorufin—has high photostability with a low photobleaching rate (rate of photobleaching, kph = 0.0013 s−1)21, making multiple exposures possible. We performed time-course fluorescence measurements (i) to allow stable fluorescent artifacts to be removed from images and (ii) to ensure that the signal from a beaded well was from an enzyme. For (i), the first fluorescent image was subtracted from fluorescent images acquired at each subsequent time point. This process removed light intensity that did not change with time; for example, fluorescence from dust and scattered light. For (ii), a positive or “on” well was identified only where fluorescence intensity in a beaded well increased in every frame and by at least 20% over four frames. This process removed false positives from random changes in fluorescence during image acquisition. Supplementary Figure 1 shows histograms of fluorescence from wells with and without enzymes, showing the good distinction between “on” and “off ” wells. Arrays were also imaged with white light to identify those wells that contain beads. After acquiring the fluorescence images, the arrays were illuminated with white light and imaged on the CCD camera. Due to scattering of light by the beads, those wells that contained a bead appeared brighter in the image than wells without beads. The fluorescence and whitelight images were analyzed using customized software.

30. Pantano, P. & Walt, D.R. Ordered nanowell arrays. Chem. Mater. 8, 2832–2835 (1996).

doi:10.1038/nbt.1641

letters

Identification of influenza A nucleoprotein as an antiviral target

© 2010 Nature America, Inc. All rights reserved.

Richard Y Kao1–3, Dan Yang4, Lai-Shan Lau1, Wayne H W Tsui1, Lihong Hu4, Jun Dai1,2, Mei-Po Chan1, Che-Man Chan1, Pui Wang1, Bo-Jian Zheng1–3, Jian Sun4, Jian-Dong Huang5, Jason Madar6, Guanhua Chen4, Honglin Chen1–3, Yi Guan1–3 & Kwok-Yung Yuen1–3 Influenza A remains a significant public health challenge because of the emergence of antigenically shifted or highly virulent strains1–5. Antiviral resistance to available drugs such as adamantanes or neuraminidase inhibitors has appeared rapidly6–9, creating a need for new antiviral targets and new drugs for influenza virus infections. Using forward chemical genetics, we have identified influenza A nucleoprotein (NP) as a druggable target and found a small-molecule compound, nucleozin, that triggers the aggregation of NP and inhibits its nuclear accumulation. Nucleozin impeded influenza A virus replication in vitro with a nanomolar median effective concentration (EC50) and protected mice challenged with lethal doses of avian influenza A H5N1. Our results demonstrate that viral NP is a valid target for the development of small-molecule therapies. The propensity of influenza virus to develop resistance to commonly used drugs requires continued development of new therapeutics. In the 2008–2009 flu season, almost 100% of the seasonal influenza H1N1 viruses circulating in the United States were resistant to the neuraminidase inhibitor oseltamivir (Tamiflu), and all isolates of the H3N2 viruses were resistant to adamantanes1–6. Over half of the indi viduals infected by the H5N1 subtype died irrespective of treatment with both classes of drug7–9. In our previous studies on SARS corona virus, we demonstrated that a forward chemical genetics approach using a chemical library of 50,240 compounds with diverse structures could interrogate most known targets for viral infection10,11. Here we screened the same library using Madin-Darby canine kidney (MDCK) cell–based influenza A infection assays and identified 950 compounds that showed protective effects as primary hits. We evaluated the selected compounds in a secondary screen using a cytopathic effect assay and selected 39 compounds for further studies based on their potency. To investigate the modes of action of these bioactive compounds, we focused on processes crucial for successful establishment of influenza infection. Influenza NP is the most abundantly expressed protein during the course of infection with multiple functionalities12. NP accumulates

in the nucleus in the early phases of infection and is exclusively dis tributed in cytoplasm later during viral assembly and maturation12–15. We examined the effects of the 39 compounds on NP nuclear traf ficking by fluorescence microscopy and identified 5 compounds that blocked the nuclear accumulation of NP. Compound FA-1 showed the best potency with EC50 < 1 μM in a plaque reduction assay (PRA) on MDCK cells infected with influenza A/WSN/33 (H1N1) virus. The schematic representation of the procedures and results of the primary, secondary and subsequent fluorescence microscopy screens are sum marized in Supplementary Figure 1. Based on the structural information of compound FA-1, four struc turally similar analogs (Fig. 1a) obtained from commercial sources were shown to have EC50 against influenza A/WSN/33 virus at submicromolar levels in PRA. We selected compound FA-4 (nucleozin) for further characterization based on its better solubility in aqueous solutions (unpublished observations) and potent antiviral activities. Nucleozin inhibited infection of MDCK cells by the viruses influenza A/WSN/33, H3N2 (clinical isolate) and Vietnam/1194/04 (H5N1) with an EC50 of 0.069 ± 0.003 μM, 0.16 ± 0.01 μM and 0.33 ± 0.04 μM in PRA, respectively (Fig. 1b), severely suppressed viral growth at 0.1 μM and totally inhibited virus production at 1 μM in multicycle growth assays (Fig. 1c). The compound has a TC50 (50% toxic concentration) >250 μM as demonstrated by MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide) assay, suggesting a wide therapeutic window (Supplementary Fig. 2). Furthermore, nucleozin effectively inhibited viral growth even when added within 6 h after inoculation of the MDCK cells with the virus (Fig. 1d), indicating that the antiviral activities of nucleozin reside on post-entry and post-nuclear events, suggesting that multiple processes involving NP may be affected, although only the nuclear import process of NP can be readily observed. Detailed fluorescence microscopy studies using human alveolar basal epithelial (A549) cells as the host for influenza A/WSN/33 virus infection showed that nucleozin is a potent antagonist of NP accumulation in the nucleus, leading to the formation of a halo of dense NP surrounding the perinuclear region in the cytoplasm at 3 h after infection (Fig. 1e). Because NP failed to enter the nucleus in the

1Department

of Microbiology, The University of Hong Kong, Hong Kong. 2Research Center of Infection and Immunology, The University of Hong Kong, Hong Kong. Key Laboratory of Emerging Infectious Diseases, The University of Hong Kong, Hong Kong. 4Department of Chemistry, The University of Hong Kong, Hong Kong. 5Department of Biochemistry, The University of Hong Kong, Hong Kong. 6Department of Computing Sciences, Capilano University, British Columbia, Canada. Correspondence should be addressed to R.Y.K. ([email protected]) or K.-Y.Y. ([email protected]). 3State

Received 23 November 2009; accepted 27 April 2010; published online 30 May 2010; doi:10.1038/nbt.1638

600

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters a

Compound FA-2

d 1 μM nucleozin 0.1 μM nucleozin No nucleozin

108 107 106 105 104 103 102 101

120

PFU relative to controls in the absence of compounds (%)

© 2010 Nature America, Inc. All rights reserved.

b

100 80 60 40

8.0 × 104 6.0 × 104

2.0 × 104 20 40 60 Hours post-infection

0

80

–1 0 1 2 4 6 8 (+) Hours relative to infection

Compound FA-4 (nucleozin) H1N1 (A/WSN/33) with nucleozin H3N2 (clinical isolate) with nucleozin H5N1 (A/Vietnam/ 1194/04) with nucleozin H1N1 (A/WSN/33) with oseltamivir

e

1 μM nucleozin NP

DAPI

No nucleozin Merged with DAPI

NP

Merged with DAPI

3h

24 h

20 0

1.0 × 105

4.0 × 104

0

Compound FA-3

1.4 × 105 1.2 × 105

PFU/ml

Compound FA-1

Viral titre (PFU/ml)

c

0.01 10 100 0.1 1 Concentration of compounds (μM)

Figure 1 Chemical structures and biological activities of nucleozin and related compounds. (a) Chemical structures of compound FA-1, FA-2, FA-3 and FA-4. (b) Nucleozin is effective against human H1N1, H3N2 and H5N1 influenza viruses. MDCK cells were infected with different strains of virus and antiviral activities determined by PRA. Oseltamivir (curve in red) was included for comparisons of in vitro efficacies. (c) Antiviral activity of nucleozin in multicycle growth assays. MDCK cells were infected with A/WSN/33 virus at 0.001 MOI in the presence or absence of nucleozin (0.1 or 1 μM). Zanamivir at 1 μM has stopped viral growth but data were omitted for clarity. Viral titers were determined by plaque assay at the time indicated. (d) Efficacies of nucleozin added at various time points. MDCK cells were infected at an MOI of 2 and nucleozin (1 μM) was added before infection (−1 h), at the time of infection (0 h) and at 1, 2, 4, 6 and 8 h after infection as indicated. +, controls with no nucleozin added. Viral titers were determined at 12 h after infection by plaque assay. The experiments were carried out in triplicate and repeated twice for confirmation. The mean value is shown with s.d. (e) Nucleozin blocked nuclear accumulation of influenza A NP in virus-infected A549 cells. Cells were infected with A/WSN/33 virus (10 MOI) in the presence or absence of 1 μM nucleozin. Influenza A NP accumulated in the nucleus at early-infection stage and was distributed exclusively in the cytoplasm at late-infection stage in the absence of nucleozin. At the indicated time point, cells were fixed and DAPI staining and mouse anti-influenza A NP antibodies were used to define the locations of the nucleus and viral NP, respectively. PFU, plaque forming unit.

resence of nucleozin, NP trapped in the cytoplasm was seen scattered p randomly in host cells at 24 h after infection (Fig. 1e). Nucleozin also inhibited the nuclear accumulation of NP in virus-infected MDCK cells (Supplementary Fig. 3), suggesting that the mechanism of action of nucleozin is not restricted to particular cell types. Using the published crystal structure of influenza A/WSN/33 NP16, we carried out molecular docking studies trying to see if NP is a direct molecular target of nucleozin. For unbiased predictive virtual dockings, we set the docking potential gradient box to cover the entire NP mono mer to probe all potential binding sites. We identified three potential binding sites of nucleozin (Fig. 2a): a previously unidentified small groove (binding site 1, Supplementary Fig. 4) in the back of the body of NP, the arginine-rich groove (binding site 2, Supplementary Fig. 5) proposed to be the RNA binding domain and the proposed tail loop groove (binding site 3, Supplementary Fig. 6). As the docking method treats the nucleoprotein as a rigid molecule and thus might not perfectly resemble the binding of nucleozin to nucleo protein in vivo, the docking results should be considered as preliminary indications of such potential binding sites, to be subjected to further val idations by biochemical and mutational studies. To localize the binding site of nucleozin to NP, escape virus mutants resistant to nucleozin were selected with increasing concentrations of nucleozin. We were not able to obtain escape mutants by using nucleozin as the selecting agent but an escape mutant showing cross-resistance to nucleozin was obtained after five passages of selection using compound FA-1. The resistant viral clone was plaque purified and selected for further studies. It is noteworthy that in the absence of nucleozin, the viral titer of the resistant mutant was nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

only slightly lower than the parental virus under low (0.001) multiplic ity of infection (MOI) conditions (Supplementary Fig. 7) but about an order of magnitude lower than the wild-type virus when a high MOI of 7 was used (unpublished observations). We speculate that the escape mutant virus may produce more defective interfering particles at high MOI but this observation may be applicable only to this Y289H mutant and not be generalized to other strains of influenza virus. Furthermore, although the mutant virus seemed to be less fit for infection in MDCK cells when compared with wild-type WSN virus, we caution that the fitness of a laboratory-induced mutant virus could only be meaningfully evaluated in the context of relevant animal models. Sequencing of all eight gene segments of the escape virus revealed only a single T-to-C mutation at nucleotide position 865 of the NP gene in segment 5 of the influenza A genome; this mutation translates into a single amino acid substitution of tyrosine to histidine in residue 289 of NP. To further confirm that the Y289H substitution was indeed the only mutation contributing to the resistant phenotype of the selected mutant virus, we used reverse genetics to create an influenza A/WSN/33 recombinant virus with a single nucleotide substitution of T with C at position 865 of the NP gene. After co-transfection of eight pHW2000based plasmids (replacing the pHW2000-NP with the mutant plasmid) encoding the gene segments of the virus into co-cultured 293T/MDCK cells17, we plaque purified and isolated the resulting recombinant virus that had a single Y289H substitution in NP and demonstrated that the recombinant virus was resistant to high concentrations of nucleozin and had a resistance profile indistinguishable from the originally isolated resistant escape viral clone (Fig. 2b). 601

letters Figure 2 Influenza A NP is the molecular target 120 Binding site 1 of nucleozin. (a) Three potential binding sites of nucleozin on the influenza A NP crystal structure 100 as predicted by molecular docking models. Electrostatic surface representation of influenza 80 A NP is color-coded (red, negative; blue, positive; Binding 60 light gray, neutral). Potential binding sites of site 2 Y289H virus nucleozin are highlighted by yellow circles. by reverse genetics 40 Y289H escape (b) Escape mutant virus and recombinant virus Binding site 3 mutant virus carrying the Y289H substitution in influenza 20 Parental A/WSN/33 A NP confer resistance to high concentrations H1N1 virus 0 of nucleozin. MDCK cells were infected with 90° 0.01 0.1 1 10 100 1,000 A/WSN/33 virus, Y289H escape mutant virus Concentration of nucleozin (μM) or Y289H variant virus generated by reverse genetics. Antiviral activities determined by PRA. The highest concentration of nucleozin used was 1 μM nucleozin No nucleozin Merged with DAPl NP DAPl Merged with DAPl NP 0.5% DMSO limited to 125 μM as fine precipitates appeared 1 μM nucleozin at higher concentration that interfered with the 5 μM nucleozin 3h determination of plaques. All virus strains were 25 μM nucleozin 6 tested in the same experiments for comparisons 20 × 10 of in vitro resistance profiles. (c) MDCK cells 15 × 106 were infected with the Y289H escape mutant 24 h virus (MOI = 5) in the presence or absence of 1 μM 10 × 106 nucleozin. The addition of nucleozin did not block the nuclear accumulation of the viral NP. No nucleozin 5 × 106 10 μM nucleozin At the indicated time points, cells were fixed NP DAPl Merged NP DAPl Merged and DAPI staining and mouse anti-influenza A 0 A/WSN/33 Y289H WSN NP antibodies were used to define the locations mutant NP of the nucleus and viral NP, respectively. (d) Nucleozin inhibits the parental virus NP Y289H activity but not theY289H variant virus NP in NP a luciferase reporter assay. 0, 1, 5 or 25 μM of nucleozin was added to 293T cells transfected with minigenomes containing A/WSN/33 virus NP or Y289H variant NP. Luciferase activities were measured 24 h post-transfection. The experiments were carried out in triplicate and repeated twice. The mean value is shown with s.d. (e) Nucleozin inhibits the nuclear import of exogenously added A/WSN/33 NP but not the Y289H variant NP. Purified recombinant NP or Y289H variant NP at 25 μM was added to digitonin-treated MDCK cells in the presence or absence of 10 μM nucleozin. Nuclear import of proteins was allowed for 30 min and followed by cell fixation and immunostaining for the presence of NP in the nucleus. DAPI was used to indicate the location of the nucleus. Images were visualized by confocal microscopy.

b

PFU relative to controls in the absence of nucleozin (%)

a

e

Fluorescence microscopy studies showed that NP of the escape mutant accumulated in the nucleus despite the presence of nucleozin (Fig. 2c), indicating that the Y289H mutation overcame the observed antagonistic effects of nucleozin against the nuclear accumulation of NP and supporting our notion that nucleozin inhibits viral infection by halting the nuclear accumulation of viral NP. Results from the luciferase reporter–based mini-genome assay18 further showed that concentrations of nucleozin that effectively abolished the replication of the virus had little effect on the Y289H variant NP (Fig. 2d). The results from mini-genome assays also indicated that the replication activity of the Y289H was about 20% lower than the wild-type NP, which might explain the lower NP signal in Y289H immunofluores cence studies (Fig. 2c). When we used purified recombinant wild-type NP or Y289H NP in a nuclear import assay19, the nuclear import of the wild-type NP was abolished in the presence of nucleozin (Fig. 2e), indicating that nucleozin acts on NP in the absence of other viral components. The inability of nucleozin to inhibit Y289H variant NP in nuclear import showed again that Y289H is a crucial mutation for nucleozin resistance. Influenza A NP has six tryptophan residues scattered through out the protein. Binding of exogenous molecules is likely to induce a change in the microenvironments of tryptophan molecules leading to a change in intrinsic fluorescence. Fluorescence spectroscopy showed that incubation of nucleozin with purified recombinant NP elicited a clear dose-dependent fluorescence-quenching effect, indicating that nucleozin binds to influenza A NP and that this binding may induce 602

d

Luciferase activity (AU)

© 2010 Nature America, Inc. All rights reserved.

c

conformational changes in the protein. Fluorescence titration using the Y289H variant NP resulted in a much lower fluorescence-quenching effect, suggesting that the binding of nucleozin to the Y289H variant NP is much weakened but not totally abolished (Fig. 3a). The docking model (Supplementary Fig. 4) suggests that residue N309 of NP forms a hydrogen bond with nucleozin, and residue Y289 forms hydrophobic interactions with nucleozin by aromatic ring stacking. Replacement of tyrosine by histidine at position 289 presumably disrupts the ringstacking effect and destabilizes the binding of the nucleozin to NP. As the experimental results with the Y289H mutant virus agreed with the predicted binding of nucleozin to NP, the small groove behind the body (Fig. 2a and Supplementary Fig. 4) of NP is likely to be critical to the binding of nucleozin. Sequencing the NP gene of ten independent nucleozin-resistant clones from our mutantraising experiment has shown that nine clones are Y289H mutants. One clone carried a N309K instead of the Y289H mutation. This second resistance mutation reconciles well with the original predictions of the docking model (Fig. 2a and Supplementary Fig. 4), suggesting that N309 may stabilize the binding of nucleozin through hydrogen bonds. Y289 is not in close proximity to the identified nuclear localiza tion signal (residues 3–13 and 198–216) of NP making, it unlikely that direct interactions of nucleozin with elements related to the nuclear localization signal are responsible for the inhibitory effect. We speculate that the binding of nucleozin to influenza NP may induce conformational changes in the protein that render NP unsuitable VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters

Relative fluorescence quenching

a

b 1.0

WSN NP Y289H variant NP

0.8 0.6 0.4

1 – + –

2 – + +

3 Wt + –

4 5 6 Wt Wt Wt + + + + + + 0.2 1 5

7 8 Mu Mu + + – + 0.2

9 Mu + + 1

10 Mu + + 5

Lane NP (8 μM) RNA (2 μM) Nucleozin Nucleozin/NP ratio

NP-RNA

NP-RNA

complex

complex

1 – + –

2 – + +

3 4 5 6 7 8 9 10 Wt Wt Wt Wt Mu Mu Mu Mu + + + + + + + + + + – + + + – + 0.2 1 5 0.2 1 5

0.2 0 0

5 10 15 20 25 Concentration of nucleozin (μM)

d

Lane 1 2 3 4 5 6 7 NP (8 μM) WT WT WT WT WT Mu Mu Nucleozin – + + + + – + Nucleozin/NP ratio 0.2 1 5 25 0.2

© 2010 Nature America, Inc. All rights reserved.

c

Lane NP (8 μM) RNA (2 μM) Nucleozin Nucleozin/NP ratio

Free RNA

8 9 10 Mu Mu Mu + + + 1 5 25

e

10 μM nucleozin

No nucleozin NP

DAPI

Merged

NP

DAPI

Merged

WSN NP

Y289H NP NP

Figure 3 Nucleozin interacts with NP and causes NP aggregation. (a) Binding of nucleozin to NP induced fluorescence quenching. We used 4 μM wild-type WSN NP or Y289H variant NP and 0–25 μM of nucleozin for fluorescence titrations. Internal fluorescence-quenching effect was due to micro-conformational changes upon the binding of nucleozin to the NP. The experiments were carried out in duplicate and repeated three times for confirmation. (b–d) Nucleozin does not inhibit RNA binding but mediates formation of large NP-RNA complexes not resolvable by gradient gel electrophoresis. Free RNA oligomer and NP-RNA complex were visualized by ethidium bromide staining in b and NP was visualized by Coomassie brilliant blue G-250 staining in c. RNA incubated with 40 μM of nucleozin in the absence of NP (lane 2) was used as a control. (d) Visualization and comparison of the effects of nucleozin on recombinant wild-type and Y289H NP (in the absence of RNA); native gradient gel conditions, stained by Coomassie brilliant blue G-250. WT, purified recombinant A/WSN/33 NP; Mu, purified recombinant Y289H variant NP; +, in the presence; −, in the absence. Positions of free RNA, NP-RNA complex and NP are indicated. (e) Visualization of nucleozin-induced NP aggregates in cells. MDCK cells transfected with plasmids expressing wild-type NP or Y289H variant NP were treated with nucleozin for 4 h. DAPI staining and mouse anti-influenza A NP antibodies were used to define the locations of the nucleus and NP respectively.

for nuclear trafficking. Further evidence supporting our proposal that nucleozin alters the conformations of NP comes from the RNA binding experiment showing that nucleozin did not inhibit NP-RNA binding (Fig. 3b) but caused a dose-dependent reduction of NP-RNA complex that could be run into the 4–12% polyacrylamide gradient gel (Fig. 3b,c; lanes 3–6). The result suggested that nucleozin induced the formation of very large NP-RNA aggregates that were too big to get into the gradient gel during electrophoresis. The NP in the assay was not lost on account of degradation; this was apparent when we separated the reaction content by denaturing SDS-PAGE under reducing conditions and detected intact NP (Supplementary Fig. 8), indicating the presence of NP aggregates that could be dissolved in denaturing conditions. In the absence of RNA, the nucleozin treat ment could also reduce the amount of NP running into the native gradient gel in a dose-dependent manner as judged by the intensi ties of Coomassie blue–stained NP in each lane (Fig. 3d, lanes 1–5), presumably due to formation of very large NP complexes. Separating the reaction content by denaturing SDS-PAGE demonstrated that the NP was present at comparable levels in treated and untreated samples and detectable under denaturing conditions (Supplementary Fig. 9). The data suggest that although nucleozin alone induced NP aggre gate formation, the presence of RNA in the system greatly enhanced the formation of very large NP-RNA complexes nonresolvable by native electrophoresis. The Y289H variant NP, on the other hand, was inert to the addition of nucleozin, in the absence or presence of RNA (Fig. 3c, lanes 7–10; Fig. 3d, lanes 6–10). To investigate if nucleozin also induces NP aggregate formation in cells, we expressed NP in MDCK cells until sufficient amounts of NP nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

were present for immunofluorescence microscopy detection. We then treated the cells with cycloheximide to stop further protein translation, added nucleozin to the cells and monitored NP aggregate formation by immunofluorescence microscopy. The formation of nucleozininduced NP aggregates could be detected 1 h after the addition of nucleozin (unpublished observation) and was readily observed 4 h after the addition of nucleozin (Fig. 3e). We speculate that the halo of dense NP surrounding the perinuclear region of infected cells were nucleozin-induced NP aggregates (Fig. 1e). Based on the immunofluorescence and biochemical data, we pro pose that nucleozin inactivates NP by inducing the formation of very large NP complexes that aggregate with RNA and possibly other cel lular components yet to be identified, leading to a complete halt in NP nuclear import. As NP is involved in different steps crucial for viral replication in many phases of the replication cycle20,21, we speculate that other processes involving NP will also be disrupted by the nucleozin-mediated formation of NP complexes. With respect to in vivo antiviral efficacy, mice treated with nucleo zin had a considerably higher survival rate after inoculation by influenza A virus H5N1 strain A/Vietnam/1194/04 than untreated controls. The protection against this highly pathogenic strain high lights the efficacy of this compound, as the H5N1 virus rapidly dis seminates to multiple organs, killing the host soon after infection 22. Without any treatment, all mice had died 7 d after inoculation. All mice survived when treated with the potent neuraminidase inhibitor zanamivir (Relenza). In the nucleozin-treated group, 50% of those receiving two doses of nucleozin (100 μl of 2.3 mg/ml nucleozin) per day for 7 d survived for more than 21 d (Fig. 4a). Three mice 603

letters a

viral replication. Investigation of other nucleoproteins of other viruses as direct druggable targets for small-molecule therapeutics is warranted.

b 10

6

105

80 PFU/lung

60 40 Zanamivir Nucleozin Untreated

20

103

0 5

10 15 Days post-challenge

20

Note: Supplementary information is available on the Nature Biotechnology website.

102

Za

0

Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/.

104

na m iv U ir nt re at ed N uc le oz in

Mice survival rate (%)

100

© 2010 Nature America, Inc. All rights reserved.

Mice groups

Figure 4 Efficacies of nucleozin in a mice H5N1 virus infection model. (a) Nucleozin protected the mice infected with highly pathogenic influenza H5N1 virus. Mice (nine per group) infected with 5 LD 50 (median lethal dose) of A/Vietnam/1194/04 H5N1 virus received 100 μl of 20 mg/ml zanamivir, 2.3 mg/ml nucleozin or PBS twice daily intraperitoneally. Treatments stopped at day 7 after infection. Conditions of the mice were monitored for 21 d. (b) Zanamivir and nucleozin reduced the viral load in the lungs of infected mice when compared to the control (untreated mice). Three mice from each group were euthanized at day 6 and lungs removed for viral load determination by standard plaque assay. Shown are the mean values with s.d.

were euthanized from each treated and untreated group on the 6th day after infection and their lungs tested for the presence of live virus by plaque assay. There was about a tenfold reduction of viral load in the lungs of the nucleozin-treated mice when compared to the untreated control group (Fig. 4b). The animal study results show that nucleozin protected mice against hypervirulent influenza A H5N1 virus in vivo and thus has the potential to be developed into useful anti-influenza therapeutics. While this manuscript was in preparation, the novel swine-origin influenza A (S-OIV) H1N1 virus appeared. Analyzing all the publicly available NP sequence of human H1, H2, H3, H5, H7, H9 influenza A subtypes revealed that although Y289H is a rare polymorphism in influenza A viruses, the newly emerged S-OIV is a natural Y289H variant. Among 525 Y289H variants detected in H1N1 viruses, 512 belong to the S-OIV (Supplementary Tables 1–3). The S-OIV has high resistance to nucleozin (EC50 > 50 μM; data not shown) in MDCK cell infection models but a close analog, compound FA-10, of nucleozin has shown promising antiviral activity against the Y289H variant and the S-OIV, with EC50 = 11.3 ± 0.9 μM and 5.0 ± 0.5 μM, respectively. FA-10, however, has reduced activity toward the wild-type WSN virus, with EC50 = 15.3 ± 2 μM (Supplementary Fig. 10). Molecular modeling suggests that FA-10 may bind favo rably into binding site 1 of nucleozin (Supplementary Fig. 11) but the hydrogen bond formation, with NP residue N309 observed in the original nucleozin-NP binding model (Supplementary Fig. 4), is abolished, presumably leading to the reduced activity of FA-10 toward wild-type WSN virus when compared with nucleozin. Further mutational studies of the binding site of nucleozin and derivatives in conjunction with structure-activity relationship are warranted to generate useful small-molecule compounds targeting the NP of a variety of viral strains. In summary, using a forward chemical genetics approach23,24, we have identified the influenza A NP as a druggable target. We also describe a lead compound, nucleozin, with efficacy in vitro and in animal studies. Nucleozin induces the formation of NP aggregates and antagonizes its nuclear accumulation, leading to cessation of 604

Acknowledgments This study was supported in part by the Carol Yu Center for Infection Seed Fund for Basic Research from the University of Hong Kong, the Research Fund for the Control of Infectious Diseases and the Area of Excellence Scheme of the University Grant Council (Grant AoE/M-12/06). The Beckman Coulter Core system is a generous gift from the Hong Kong Sanatorium Hospital Doctors’ Donation Fund by Y.-C. Tsao, C.-M. Chan, G. Lo, K.-M. Lai, R.K.Y. Lo, M. Tsao, B.S.S. Tse, T.-F. Tse, S.W.C. Wu, D.Y.C. Yu, R.Y.H. Yu and Y.-K. Tsao. We are grateful to R. Webster for gifts of the pHW2000 plasmids and E. Hoffmann for luciferase reporter system. We thank V. Poon, C. Chan and Q. Zhang for mice studies and K.H. Chan for virus strains. The use of Confocal Systems Core Facility provided by the LKS Faculty of Medicine, HKU, is acknowledged. Author contributions R.Y.K. and K.-Y.Y conceived the study. R.Y.K. designed and performed experiments and analyzed data. D.Y. gave conceptual advice and technical support on chemistry. L.-S.L., W.H.W.T., J.D., M.-P.C., C.-M.C. and P.W. performed experiments. J.S., L.H., and G.C. performed molecular dockings. B.-J.Z. provided animal study data. J.-D.H. gave conceptual advice on protein trafficking. J.M. constructed database and performed HTS data normalization. H.C. and Y.G. provided reverse genetics system. K.-Y.Y. did troubleshooting and provided the grant support. R.Y.K. and K.-Y.Y. supervised the study and wrote the paper. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Webster, R.G. & Govorkova, E.A. H5N1 influenza – continuing evolution and spread. N. Engl. J. Med. 355, 2174–2177 (2006). 2. Regoes, R.R. & Bonhoeffer, S. Emergence of drug-resistant influenza virus: population dynamical considerations. Science 312, 389–391 (2006). 3. Moscona, A. Global transmission of oseltamivir-resistant influenza. N. Engl. J. Med. 360, 953–956 (2009). 4. Dharan, N.J. et al. Oseltamivir-Resistance Working Group. Infections with oseltamivirresistant influenza A(H1N1) virus in the United States. J. Am. Med. Assoc. 301, 1034–1041 (2009). 5. Layne, S.P., Monto, A.S. & Taubenberger, J.K. Pandemic influenza: an inconvenient mutation. Science 323, 1560–1561 (2009). 6. Lackenby, A., Thompson, C.I. & Democratis, J. The potential impact of neuraminidase inhibitor resistant influenza. Curr. Opin. Infect. Dis. 21, 626–638 (2008). 7. Yuen, K.Y. et al. Clinical features and rapid viral diagnosis of human disease associated with avian influenza A H5N1 virus. Lancet 351, 467–471 (1998). 8. Cumulative Number of Confirmed Human Cases of Avian Influenza A (H5N1) Reported to WHO (http://www.who.int/csr/disease/avian_influenza/country/cases_table_2009_09_ 24/en/index.html). 9. Wong, S.S. & Yuen, K.Y. Avian influenza virus infections in humans. Chest 129, 156–168 (2006). 10. Kao, R.Y. et al. Identification of novel small-molecule inhibitors of severe acute respiratory syndrome-associated coronavirus by chemical genetics. Chem. Biol. 11, 1293–1299 (2004). 11. Kao, R.Y. et al. Characterization of SARS-CoV main protease and identification of biologically active small molecule inhibitors using a continuous fluorescence-based assay. FEBS Lett. 576, 325–330 (2004). 12. Portela, A. & Digard, P. The influenza virus nucleoprotein: a multifunctional, RNA-binding protein pivotal to virus replication. J. Gen. Virol. 83, 723–734 (2002). 13. Davey, J., Dimmock, N.J. & Colman, A. Identification of the Sequence Responsible for the Nuclear Accumulation of the Influenza Virus. Cell 40, 667–675 (1985). 14. Boulo, S., Akarsu, H., Ruigrok, R.W.H. & Baudin, F. Nuclear traffic of influenza virus proteins and ribonucleoprotein complexes. Virus Res. 124, 12–21 (2007). 15. Ozawa, M. et al. Contributions of two nuclear localization signals of influenza A virus nucleoprotein to viral replication. J. Virol. 81, 30–41 (2007). 16. Ye, Q., Krug, R.M. & Tao, Y.J. The mechanism by which influenza A virus nucleoprotein forms oligomers and binds RNA. Nature 444, 1078–1082 (2006).

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters 21. Elton, D., Medcalf, E., Bishop, K., Harrison, D. & Digard, P. Identification of amino acid residues of influenza virus nucleoprotein essential for RNA binding. J. Virol. 73, 7357–7367 (1999). 22. Zheng, B.J. et al. Delayed antiviral plus immunomodulator treatment still reduces mortality in mice infected by high inoculum of influenza A/H5N1 virus. Proc. Natl. Acad. Sci. USA 105, 8091–8096 (2008). 23. Stockwell, B.R. Chemical genetics: ligand-based discovery of gene function. Nat. Rev. Genet. 1, 116–125 (2000). 24. Strausberg, R.L. & Schreiber, S.L. From knowing to controlling: a path from genomics to drugs using small molecule probes. Science 300, 294–295 (2003).

© 2010 Nature America, Inc. All rights reserved.

17. Hoffmann, E., Neumann, G., Kawaoka, Y., Hobom, G. & Webster, R.G. A DNA transfection system for generation of influenza A virus from eight plasmids. Proc. Natl. Acad. Sci. USA 97, 6108–6113 (2000). 18. Wang, P. et al. Nuclear factor 90 negatively regulates influenza virus replication by interacting with viral nucleoprotein. J. Virol. 83, 7850–7861 (2009). 19. Wu, W.W., Sun, Y.H. & Panté, N. Nuclear import of influenza A viral ribonucleoprotein complexes is mediated by two nuclear localization sequences on viral nucleoprotein. Virol. J. 4, 49 (2007). 20. Digard, P. et al. Modulation of nuclear localization of the influenza virus nucleoprotein through interaction with actin filaments. J. Virol. 73, 2222–2231 (1999).

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

605

ONLINE METHODS

© 2010 Nature America, Inc. All rights reserved.

Virus and chemical reagents. Influenza A/WSN/33, H3N2 and swineorigin influenza A (H1N1) virus S-OIV (A/HK/415742/09) were propa gated in MDCK cells. After full cytopathic effects developed in cultures in infected MDCK cell cultures, the viral particles were harvested and stored in −70 °C freezers for further studies. The influenza A virus strain A/Vietnam/1194/04 was grown in embryonated eggs and the virus-containing allantoic fluid was harvested and stored in aliquots at −70 °C. A total of 50,240 structurally diverse small-molecule compounds (ChemBridge) was screened. MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazo lium bromide) was purchased from Sigma-Aldrich. RNA oligomer (5′UUUGUUACACACACACACGCUGUG-3′) used for RNA binding assays was synthesized by IDT (Integrated DNA Technologies). Cell-based high-throughput screening (HTS) in 384-well microtiter plates. The primary HTS was carried out in a fully automated Beckman Coulter Core System (Fullerton) integrated with a Kendro robotics CO2 incubator (Thermo Fisher Scientific) at Chemical Genetics Unit, Department of Microbiology, Research Center of Infection and Immunology, LKS Faculty of Medicine, the University of Hong Kong. Compounds were arrayed in 384-well microtiter plates (Greiner Bio-One) in triplicate with a final concentration of 20 μg/ml and 5,000 MDCK cells per well in 50 μl complete Eagle’s minimal essential medium (EMEM) supplemented with 1% heat-inactivated FBS. Cells were then infected with influenza virus (A/WSN/33) at an MOI of 0.01. After infec tion, plates were incubated at 37 °C with 5% CO2. At 3 d post-infection, 20 μl of 0.625 mg/ml of MTT was added into each well followed by an additional incubation time of 8 h at 37 °C with 5% CO2. At the end of the incubation, 30 μl SDS with 0.01 M HCl was added to solubilize the formazan, and after overnight incubation, MTT readings were recorded in a DTX 880 multimode detector (Beckman Coulter) at 570 nm with 640 nm as the reference wavelength. HTS data analysis. The HTS data was transferred to a Dell Precision 500 Workstation and normalized by custom designed data analysis software in two stages. For stage one normalization, each well was divided by the median of each plate and a normalized value was obtained for each well. As the HTS was carried out in triplicate, the second stage normalization took into account the variation between each well and the final reading was calculated by aver aging the two closest readings. This two-stage normalization method mini mized potential experimental errors, which are usually random and sporadic in nature. Secondary screening. Secondary screening was carried out in triplicate in 96-well tissue culture plate (TPP) at 10 μg/ml. Selected compounds were first dispensed in the wells, followed by the addition of 20,000 MDCK cells and 200 PFU of influenza A/WSN/33 (H1N1) virus into each well. The plates were incubated at 37 °C with 5% CO2 and monitored daily using a Leica DM inverted light microscope for virus-induced cytopathic effect. Compounds that gave full protection of MDCK cells (no cytopathic effect) were selected for further studies. The cytotoxicity of selected compounds was determined by MTT assay according to manufacturer’s instructions. Plaque reduction assay. The PRA assay was performed in triplicate in 24-well tissue culture plates (TPP). The MDCK cells were seeded at 1 × 105 cells/well in EMEM (Invitrogen) with 10% FBS on the day before carrying out the assay. After 16–24 h, 40–50 PFU of influenza virus were added to the cell monolayer with or without the addition of compounds and the plates further incubated for 2 h at 37 °C with 5% CO2 before removal of unbound viral particles by aspiration. The cell monolayer was washed once with EMEM before being overlaid with 1% low melting agarose (Cambrex) in EMEM containing 1% FBS, 1 μg/ml TPCK trypsin (Invitrogen) and appropriate amounts of compounds. The plates were incubated at 37 °C with 5% CO2 for 72 h. At 72 h after infec tion, the wells were fixed with 10% formaldehyde (BDH). After removal of the agarose plugs, the monolayers were stained with 0.7% crystal violet (BDH) and the plaques counted. The percentage of plaque inhibition relative to the control (without the addition of compound) plates were determined for each compound concentration and the EC50 was calculated using Sigma plot (SPSS). The PRA were carried out in triplicate and repeated twice for confirmation.

nature biotechnology

For multicycle growth experiments for the evaluation of antiviral activities of compounds, 0.001 MOI was used accordingly25 and viral yield determined by plaque assay. Immunofluorescence microscopy. A549 and MDCK cells were grown to 70–80% confluency on coverslips. Cells were infected for 2 h at MOI = 10 and MOI = 5 for A549 and MDCK cells, respectively in the presence or absence of 1 μM nucleozin and washed. Nucleozin was maintained in culture throughout the experiment. Infections were stopped at indicated time points by fixation in 4% paraformaldehyde (Electron Microscopy Sciences) for 15 min. Cells were permeabilized in 0.1% Triton-X100 for 5 min and then were incubated for 1 h with primary antibodies against NP (Abcam) in PBS containing 5% goat serum (dilution 1:1,000), washed and stained with fluo rescein isothiocyanate (FITC)-conjugated secondary antibodies (Invitrogen) (dilution 1:150) for 0.5 h. Coverslips were then washed and counterstained with 4′,6-diamidino-2-phenylindole, dihydrochloride (DAPI) (Invitrogen) for nucleus localization and mounted on slides using Prolong Gold antifade mounting medium (Invitrogen) before image analysis by fluorescence micro scopy (SPOT Diagnostic Instrument). Cloning and purification of recombinant influenza A/WSN/33 NP and Y289H variant NP. The full influenza A/WSN/33 (H1N1) NP or Y289H NP gene was cloned into pET28a vector (Novagen) with C-terminal His-tag pro vided by the vector, expressed in Escherichia coli BL21 (DE3) cells, and the recombinant NPs purified to homogeneity by HisTrap HP, HiTrap heparin HP and Superdex 200 gel filtration columns (Amersham Biosciences). The purified protein was 95% pure as determined by SDS-PAGE. Fluorescence spectroscopy. Fluorescence-quenching method26 was used to examine if nucleozin interacts with purified influenza A NP in vitro. Briefly, a Hitachi F-4800 fluorescence spectrophotometer was used for fluorescence titrations in a 2 ml quartz cuvette at 25 °C. The excitation and emission wavelengths were set at 295 nm and 333 nm, respectively. Samples con tained 4 μM purified influenza NP in 20 mM Tris, 150 mM NaCl, pH 7.3. Nucleozin was used at concentrations between 0.03 and 25 μM. For the titrations, additives were not allowed to exceed 1.5% of the total volume of the solution. The experiments were carried out in duplicate and repeated three times for confirmation. Molecular docking of nucleozin to influenza A NP. To postulate the potential binding site(s) of nucleozin in the viral NP, molecular modeling for NP-nucleozin interaction was performed using Autodock3.0.5 (ref. 27) and CHARMM force field28. As amino acid residues 73–91, 397–401 and 429–437 do not have defined tertiary structure in the available NP crystal structure of H1N1 virus (Protein Data Bank code: 2IQH), SWISS-MODEL was used to repair the structure of the NP. The three-dimensional structure of nucleozin was then generated by Chem3D (CambridgeSoft). Hydrogen atoms were assigned and Sybyl Mol2 files for Autodock were prepared by InsightII (Accelrys), and partial charge and potential were assigned by CHARMM force field. Autodock3.0.5 Lamarckian Genetic Algorithm and its default parameters were used for docking. After carrying out docking, models of the complexes of nucleozin interacting with influenza NP were obtained. CHARMM force field was used to optimize the docked complexes by removing bad contacts between the ligand and the protein. PyMol (Delano) was used to plot the modeling figures. Generation of escape mutant influenza A virus resistant to nucleozin. One method of mutant generation29 was followed to raise mutant viruses resistant to compound FA-1 and nucleozin. Briefly, the influenza A/WSN/33 (H1N1) was passaged in MDCK cells in the presence of increasing concentrations of the compounds and the desired resistant viral clone purified by plaque isola tion on MDCK cell monolayers. It took five passages to obtain the escape mutant. After plaque purification, we carried out whole genome sequencing for one clone. The escape mutant viral RNA was extracted by TRIZOL Reagent (Invitrogen) and complementary DNA (cDNA) of all eight segments was obtained by reverse transcription using Superscript III reverse transcriptase (Invitrogen) and cDNA amplified by PCR according to standard procedures.

doi:10.1038/nbt.1638

© 2010 Nature America, Inc. All rights reserved.

After identification of NP as the potential target of nucleozin, we further sequenced the NP genes of ten independent resistant clones to examine other possible mutations leading to nucleozin resistance. All DNA sequencing was carried out in the Genome Research Center (University of Hong Kong).

paraformaldehyde. Cells were immunostained as described above. The images were visualized using confocal microscropy (Carl Zeiss).

Generation of recombinant influenza virus by reverse genetics. The pHW2000 eight-plasmid system was used to generate desired recombinant viruses17. The eight plasmids contain the cDNA of the eight segments of the influenza A/WSN/33 (H1N1) genome: pHW2000-PB2, pHW2000-PB1, pHW2000-PA, pHW2000-HA, pHW2000-NP, pHW2000-NA, pHW2000-M and pHW2000-NS. Standard PCR-based techniques were used to clone the mutation in the NP gene of the escape mutant into the BsmI sites of pHW2000-NP. On the day before transfection, confluent 293T and MDCK cells were trypsinized and co-cultured in a six-well tissue culture plate (TPP). TransIT -Oligo Tranfection Reagent (Mirrus) was used to transfect the cocultured 293T and MDCK cells according to manufacturer’s instructions. The infectious particles from the supernatants were harvested 48 and 72 h post-transfection, and the recombinant virus titer enumerated in MDCK cells by plaque assay.

Electrophoretic mobility shift assay. Electrophoretic mobility shift assay was used to examine the effect of nucleozin on the RNA binding activity of the NP proteins. Purified recombinant wild-type WSN or Y289H variant NP proteins were incubated with nucleozin at 25° C for 30 min, then a 24-nucleotide RNA oligomer16 was added and incubated for another 30 min. Nuclease-free water was added, up to 10 μl. Final concentration of small RNA oligomer was 2 μM and molar ratio of NP/RNA was kept at 4:1. After incubation, the samples were mixed with 3 μl 6× DNA loading dye (0.25% bromophenol blue, 0.25% xylene xyanol, 40% sucrose) and loaded into sample wells of nondenaturing 4–12% gradient Bis-Tris NuPAGE gel (Invitrogen) equilibrated by pre-electrophoresis at 50 V in 1× TBE. Samples were separated by electrophoresis at a constant voltage of 150 V for 35 min at 25° C in 1× TBE. The gel was first visualized by ethidium bromide staining for RNA shift patterns followed by staining with Coomassie brilliant blue G-250 for NP shift patterns. For examining the effects of nucleozin-NP interac tions in vitro in the absence of RNA, we used NativePAGE 4–16% Bis-Tris gradient gel (Invitrogen) for the separation of NP under native conditions accordingly30.

Luciferase reporter assay for polymerase complex activity. The luciferase reporter assay was performed as previously reported18. Full-length genomic segments of NP, PA, PB1 and PB2 derived from different virus strains were cloned into pHW2000 vector. RNP complex comprised of WSN wild-type NP or Y289H variant NP, PA, PB1 and PB2 were co-transfected into 293T cells with a luciferase reporter plasmid, pYH-Luci, which contains noncoding sequence from the M segment of influenza A virus and firefly luciferase gene driven by polI. Plasmid phRL-TK (Promega), which expresses Renilla luci ferase, was also co-transfected as an internal control for data normalization. At 2 h after transfection, different amounts (final concentrations: 1–25 μM) of nucleozin (in DMSO) were added to the transfected cells. As a solvent control, some volume of DMSO was added to a final concentration of 0.5%. At 24 h after transfection, the luciferase activities were measured using Dual Luciferase Assay System kit (Promega E1910) and Victor3 multilabel plate reader (Perkin Elmer).

Animal experiments. As previously described22, we kept the 5- to 7-week-old BALB/c female mice in biosafety level 3 housing and gave the mice access to standard pellet feed and water ad libitum. All experimental protocols followed the standard operating procedures of the approved biosafety level 3 animal facilities and were approved by the Animal Ethics Committee. One group (nine mice) was injected intraperitoneally (i.p.) with 100 μl of 20 mg/ml of zanamivir (GlaxoSmithKline), a second group (nine mice) was injected with 2.3 mg/ml of nucleozin, and the untreated group (nine mice) was injected with PBS 1 h before inoculating the mice intranasally with 5 LD50 (100 PFU) of the A/Vietnam/1194/04 H5N1 virus in 20 μl 2 mg/ml zanamivir, 0.23 mg/ml of nucleozin or PBS. Two doses per day of i.p. 100 μl of 20 mg/ml zanamivir, 2.3 mg/ml of nucleozin or PBS were given for 7 d. Animal survival and general conditions were monitored for 21 d or until death. Three mice in each group were euthanized randomly on day 6 post-inoculation and lungs were removed for determination of viral titers by plaque assay.

NP nuclear import assay in MDCK cells. The assay was done as described19. MDCK cells were seeded on glass coverslips and were washed with import buffer (20 mM HEPES, pH 7.4, 110 mM potassium acetate, 1 mM EGTA, 5 mM sodium acetate, 2 mM magnesium acetate and 2 mM dithiothreitol) before being permeabilized with 20 μg/ml of digitonin (Sigma) for 5 min at 25° C. Meanwhile, 25 μM of NP with or without 10 μM of nucleozin were preincubated for 5 min at 25° C. After washing off the digitonin with import buffer, the NP-nucleozin mixture was added to the permeabilized cells and incubated for 30 min. The energy-generating system (0.4 mM ATP, 0.45 mM GTP, 4.5 mM phosphocrea tine and 18 U/ml phosphocreatine kinase (Merck)), 15% rabbit reticulocyte lysate (Promega), 1.6 mg/ml BSA (Sigma) and protease inhibitors (Roche) were subsequently added to the cells and incubated at 37 °C CO2 incubator for another 30 min. The cells were then rinsed with import buffer and fixed with 4%

25. Min, J.Y. & Krug, R.M. The primary function of RNA binding by the influenza A virus NS1 protein in infected cells: Inhibiting the 2′-5′ oligo (A) synthetase/RNase L pathway. Proc. Natl. Acad. Sci. USA 103, 7100–7105 (2006). 26. Campanacci, V. et al. Moth chemosensory protein exhibits drastic conformational changes and cooperativity on ligand binding. Proc. Natl. Acad. Sci. USA 100, 5069–5074 (2003). 27. Morris, G.M. et al. Automated docking using a Lamarckian genetic algorithm and an empirical binding free energy function. J. Comput. Chem. 19, 1639–1662 (1998). 28. Brooks, R. et al. CHARMM: a program for macromolecular energy, minimization, and dynamics calculations. J. Comput. Chem. 4, 187–217 (1983). 29. Gubareva, L.V. et al. Characterization of mutants of influenza A virus selected with the neuraminidase inhibitor 4-guanidino-Neu5Ac2en. J. Virol. 70, 1818–1827 (1996). 30. Niepmann, M. & Zheng, J. Discontinuous native protein gel electrophoresis. Electrophoresis 27, 3949–3951 (2006).

doi:10.1038/nbt.1638

nature biotechnology

letters

© 2010 Nature America, Inc. All rights reserved.

Synthetic peptide-acrylate surfaces for long-term self-renewal and cardiomyocyte differentiation of human embryonic stem cells Zara Melkoumian1,3, Jennifer L Weber1,3, David M Weber1, Andrei G Fadeev1, Yue Zhou1, Paula Dolley-Sonneville1, Jiwei Yang2, Liqun Qiu2, Catherine A Priest2, Christopher Shogbon1, Arthur W Martin1, Jodelle Nelson1, Peter West1, James P Beltzer1, Santona Pal1 & Ralph Brandenberger2 Human embryonic stem cells (hESCs) have two properties of interest for the development of cell therapies: self-renewal and the potential to differentiate into all major lineages of somatic cells in the human body. Widespread clinical application of hESC-derived cells will require culture methods that are low-cost, robust, scalable and use chemically defined raw materials. Here we describe synthetic peptideacrylate surfaces (PAS) that support self-renewal of hESCs in chemically defined, xeno-free medium. H1 and H7 hESCs were successfully maintained on PAS for over ten passages. Cell morphology and phenotypic marker expression were similar for cells cultured on PAS or Matrigel. Cells on PAS retained normal karyotype and pluripotency and were able to differentiate to functional cardiomyocytes on PAS. Finally, PAS were scaled up to large culture-vessel formats. Synthetic, xeno-free, scalable surfaces that support the self-renewal and differentiation of hESCs will be useful for both research purposes and development of cell therapies. Since the derivation of the first hESC lines1,2, hundreds of new lines have been derived and propagated under various culture conditions3. Maintenance of hESCs in an undifferentiated state originally required complex culture systems comprising mouse or human feeder cell layers, medium containing fetal bovine serum (FBS) or serum replacement to provide an extracellular matrix (ECM)-rich environment for cell adhesion, and soluble growth factors1,4,5. Feeder cells can be avoided by culturing hESCs on Matrigel, a basement membrane material prepared from mouse sarcoma6, human serum7 or purified ECM proteins6,8–10. However, most of these biological materials are expensive to manufacture, have limited scalability and may have high batch-to-batch variability. In addition, animal-derived materials must be subject to costly testing to ensure freedom from pathogens. Here we describe the development of synthetic surfaces made of acrylate conjugated to biologically active peptides (PAS) for the culture of undifferentiated hESCs and their differentiated derivatives. Acrylates are commonly used biomaterials with tunable physical properties.

Biologically active peptides provide a synthetic, scalable alternative to complex extracellular matrix proteins. PAS were prepared by depositing carboxylic acid containing acrylate onto culture vessel surfaces and conjugating amine-containing peptides using 1-Ethyl-3-(3-dimethyl aminopropyl)-carbodiimide/N-hydroxysuccinimide (EDC/NHS) chemistry (Fig. 1a). Peptides derived from active domains of extra cellular matrix proteins were conjugated to PAS: bone sialoprotein (BSP-PAS)11, vitronectin (VN-PAS)12, long or short fibronectin (sFN-PAS or lFN-PAS)13 and laminin (LM-PAS)14 (Table 1). H1 and H7 hESCs were seeded into 96-well plates coated with the different PAS and cultured for 48 h in the defined medium X-VIVO 10 supplemented with 80 ng/ml hrbFGF and 0.5 ng/ml hrTGF-β1 (X-VIVO 10 + GF). Matrigel-coated 96-well plates were used as positive controls. Cell attachment was assessed after 48 h by measuring the activity of alkaline phosphatase, an enzyme highly expressed on the surface of undifferentiated hESCs15. BSP-PAS and VN-PAS, but not sFN-PAS, lFN-PAS or LM-PAS, supported adhesion of H1 and H7 hESCs to a similar extent as Matrigel (Fig. 1b). Incorporation of a spacer such as poly(ethylene oxide) may reduce steric hindrance and therefore have beneficial effects for peptide concentration and cell adhesion response on peptide modified surfaces. However, a poly(ethylene oxide) linker added to the N-terminus of the BSP peptide did not significantly improve cell adhesion compared to BSP-PAS (Fig. 1b). To further investigate the utility of PAS, we cultured H7 and H1 hESCs in 6-well plates coated with different PAS for several days. Typical hESC colony morphology (tightly packed colonies with a high cell nucleus/cytoplasm ratio) and cell numbers similar to hESC cultures on Matrigel were observed only on BSP-PAS and VN-PAS (Fig. 1c,d). Although some hESC colonies attached to sFN-PAS and lFN-PAS, they appeared as poorly spread clusters of cells surrounded by multiple areas of differentiated cells (Fig. 1d). Notably, all peptides used contain the RGD cell adhesion motif, but only the BSP- and VN-derived peptides supported hESC adhesion and morphology, suggesting that the RGD sequence alone is not sufficient for optimal PAS–cell interaction.

1Corning Life Sciences, Corning Inc., Corning, New York, USA. 2Geron Corporation, Menlo Park, California, USA. 3These authors contributed equally to this work. Correspondence should be addressed to Z.M. ([email protected]) and R.B. ([email protected]).

Received 19 October 2009; accepted 1 April 2010; published online 30 May 2010; doi:10.1038/nbt.1629

606

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters

Cell number (×10∧6)

60 40 20

© 2010 Nature America, Inc. All rights reserved.

H7 hESC

6.0 5.0 4.0 3.0 2.0 1.0

Mock-PAS

IF N M at rig el

N sF

P

LM M at rig el

N

IF N

sF

BS

f

BSP-PAS

5 0. 06 3 M at rig el

7.0 6.0 5.0 4.0 3.0 2.0 1.0 0

0. 12

0. 75

0. 5 0. 25

P

Li nk Per VN

Fluorescent peptide H7 cell number

14 12 10 8 6 4 2 0

VN

0

BS

e

H1 hESC

7.0

0

1

BSP-PAS

sFN-PAS

80

8.0

Cell number × 10∧6)

Matrigel

Peptide-acrylate on surface

100

BS

d

Acrylate on surface

c

H7 hESC

120

Relative fluorescence unit × 1,000

Uncoated surface

H1 hESC 140

AttoPhos RFU (% of matrigel)

C O C 2H O 2H C O 2 C H O 2H C O 2H

Deposition of carboxylic acid containing acrylate coating

b

EDC/NHS conjugation of amine– containing peptide

C O C NH O – N p C H– ept O pe id N p e C H– t i d O pe e N p C H– t i d O pe e N H ptid –p ep e tid e

a

Peptide concentration mM Figure 1 Development of the PAS surface. (a) Schematic representation Matrigel 1 mM BSP 0.0625 mM BSP 0 mM BSP of the peptide-acrylate surface coating process. Vessel surfaces are coated with carboxylic acid containing acrylate, followed by EDC/NHS conjugation of amine-containing peptide. (b) hESC attachment to PAS in defined medium, X-VIVO 10 + GF in 96-well plates. H1 (white bars) and H7 (black bars) hESCs attach to BSP- or VN-, but not FN- or LM-PAS. AttoPhos relative fluorescent units (RFU) represent alkaline phosphatase activity of attached cells 48 h after seeding. Values are normalized to cell attachment to Matrigel (absolute values from 5,000–55,000). (c) H7 and H1 hESC growth on PAS in X-VIVO 10 + GF. One million cells were seeded in each well of a 6-well plate and cultured for 4 d. hESC cell number was comparable to Matrigel for BSP-PAS and VN-PAS but not FN-PAS. (d) H7 hESC colony morphology on day 3 on Matrigel, BSP-PAS and sFN-PAS. Scale bars, 200 μm. (e) Peptide concentration–dependent H7 hESC attachment and growth. BSP-PAS 6-well plates were prepared with serial dilutions of BSP peptide spiked with 0.25% rhodamine-labeled peptide. 1 × 10 6 H7 hESCs were seeded per well and cultured for 5 d in X-VIVO + GF. RFU correspond to fluorescent intensity of 0.25% rhodamine-labeled peptide. (f) ELISA staining with BCIP/NBT using anti-BSP antibodies shows uniform peptide distribution within a well of a 6-well plate. (g) Peptide surface density–dependent hESC attachment. H7 hESCs were seeded at 1 × 10 6 cells/well of a BSP-PAS 6-well plate and cultured in X-VIVO 10 + GF for 5 d. Crystal violet staining of cells shows uniform cell distribution and BSP concentrationdependent confluency.

g

Earlier studies in the field16,17 suggested that functional peptide density on the plating surface and uniform peptide distribution are important attributes for optimal cell response. To investigate these surface characteristics in our system, we used a dilution series of BSP peptide solution to generate different densities of BSP peptide on the plating surface. We observed good correlation between conjugated peptide density and H7 cell number (Fig. 1e) after 5 d in culture on BSP-PAS in defined medium, as well as uniform distribution of BSP peptide on the plating surface (Fig. 1f). BSP peptide solution concentration >0.5 mM was sufficient to yield cell numbers similar to those of Matrigel control cultures (Fig. 1e,g; see Supplementary Fig. 1a for calculation of BSP surface peptide density). Our results confirm that high surface density of the supportive peptide on BSP-PAS is required to achieve hESC expansion similar to that on Matrigel. Similar results were obtained for H7 on VN-PAS (Supplementary Fig. 1b). The primary goal of this study was to develop a synthetic surface for sustained self-renewal of undifferentiated hESCs in medium that is free from nonhuman, animal-origin raw materials (xeno-free) and that is chemically defined. To this end, H7 hESCs were cultured on BSP-PAS and VN-PAS for more than ten serial passages in a defined medium consisting of X-VIVO 10 supplemented with hrbFGF and hrTGF-β1 (X-VIVO 10 + GF)9. Cell doubling time, viability, colony morphology and hESC marker expression were assessed at the end of each passage. Doubling times for cells grown on either PAS were similar to those of Matrigel cultures (Fig. 2a and Supplementary Table 1), with typical hESC colony morphology (Fig. 2b). Expression of hESC-specific markers, such as OCT4, TRA-1-60 and SSEA-4, was similar between cells cultured on BSP- or VN-PAS and Matrigel control cultures nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

(Fig. 2c and Supplementary Fig. 2a). Marker expression levels were quantified throughout the ten-passage study (Supplementary Table 1 and Supplementary Fig. 2b). No karyotypic abnormalities were associated with hESC propagation on PAS surfaces (data not shown). Similar data were obtained for the H1 cell line (Supplementary Table 1 and Supplementary Fig. 2c–e), and for BG01V hESCs from Invitrogen cultured on VN-PAS (data not shown). To evaluate the compatibility of PAS with different culture systems, we tested the ability of PAS to support self-renewal of hESCs in several commercially available media: the chemically defined medium mTeSR1, KnockOut SR-supplemented medium (Supplementary Table 1 and Supplementary Fig. 3a–d) and the chemically defined, animal protein–free medium TeSR2 (data not shown). PAS effectively supported self-renewal of hESCs in all three media, with the cells retaining important hESC characteristics, such as doubling time and predicted phenotypic marker expression.

Table 1 List of peptide sequences used in this study Sequence

Abbreviation

Reference

KYGAASIKVAVSADR Ac-KGGNGEPRGDTYRAY Ac-KGGPQVTRGDVFTMP PEO4-NGEPRGDTYRAY GRGDSPK Ac-KGGAVTGRGDSPASS

LM BSP VN BSP-Linker sFN lFN

14 11 12 14 13 13

LM, laminin; BSP, bone sialoprotein; VN, vitronectin; BSP-Linker, BSP peptide sequence with polyethylene oxide linker at the N terminus; sFN, short fibronectin sequence; lFN, long fibronectin sequence.

607

letters

Doubling time (h)

100 80

VN Matrigel

60 40 20 0

© 2010 Nature America, Inc. All rights reserved.

BSP

0 1 2 3 4 5 6 7 8 9 10 11 12 13 Passage number

b

Matrigel

BSP-PAS

VN-PAS

Figure 2 PAS supports long-term culture and pluripotency of H7 hESCs in chemically defined medium X-VIVO 10 + GF. (a) Doubling time of H7 cells over the course of 12 consecutive passages on BSP-PAS, VN-PAS and Matrigel. Seeding density at each passage was 1 × 106 cells/well of a 6-well plate. Cells were harvested for passaging after 4–5 d. (b) Phase contrast images of cell colony morphology for H7 cells at passage 2 (top panel) and 10 (bottom panel) on Matrigel and PAS, as indicated. Scale bars, 200 μm. (c) Indirect immunofluorescence staining of H7 hESCs after ten consecutive passages on BSP-PAS, VN-PAS and Matrigel. Scale bars, 100 μm. (d) H7 cells retain pluripotency after long-term culture on VN-PAS or BSP-PAS. H7 hESCs were thawed and cultured for eight passages on BSP-PAS, VN-PAS and Matrigel (positive control); cells were harvested and 1 × 10 7 cells were injected by intramuscular injection into the flank of SCID/bg mice. Tissues from all three germ layers, represented as secretory epithelium (endoderm), cartilage (mesoderm) and neuroepithelium (ectoderm), were identified in teratomas formed by H7 hESCs cultured on VN-PAS, BSP-PAS or Matrigel. Scale bar, 50 μm.

Maintenance of pluripotency is a critical parameter when evaluating new surfaces for hESC culture. The pluripotency of H7 hESCs maintained on PAS was assessed by in vivo teratoma formation. Cells that had been maintained for nine passages on BSP-PAS or VN-PAS were cryopreserved, then thawed back onto PAS and cultured for an additional eight passages before injection. Cells were harvested, and 1 × 107 cells were injected intramuscularly into the flanks of severe combined immunodeficient/beige (SCID/bg) mice. Cells cultured on both BSP-PAS and VN-PAS formed teratomas containing tissues from all three germ layers (Fig. 2d). Pluripotency of H7 hESCs maintained on BSP-PAS or VN-PAS was confirmed by differentiation to embryoid bodies (Supplementary Fig. 4). Several studies, as well as our own data, suggest that cell culture surface, medium and passaging method could contribute to variations in hESC differentiation capacity18,19. To examine the cardiac differentiation potential of H7 hESCs after long-term culture on PAS, we sequentially treated confluent cell monolayers on BSP-PAS with activin A and BMP4 as described previously20. About 2 weeks after the induction of differentiation, we observed spontaneously beating structures that expressed the cardiomyocyte-specific markers Nkx2.5 and α-actinin (Fig. 3a); flow cytometric analysis revealed that ~60% of the cells were α-actinin and Nkx2.5 positive (Fig. 3b). H7 hESCs maintained and differentiated to cardiomyocytes on VN-PAS were submitted for electrophysiological characterization to ChanTest Corporation. Single-cell action potentials were recorded with perforated patch current clamp technology. Action potential duration at 90% repolarization (APD90) was used as the primary parameter to categorize action potential waveforms into ventricular, atrial or nodal cell types. An APD90 value of 150 ms was used to discriminate atrial from ventricular cardiomyocytes. The majority of the cells (88%) were of the ventricular type, and 12% were designated atrial type (Supplementary Table 2 and Supplementary Fig. 5a,b). No nodaltype cells were found. These results confirmed the capacity of H7 hESCs to differentiate into cardiomyocytes after long-term culture on BSP- and VN-PAS, showing that these surfaces are suitable for both the expansion and differentiation phases of hESC culture. 608

c

OCT4

SSEA-4

Tra-1-60

Hoechst

Hoechst

Hoechst

Passage 10 (96 h) Passage 2 (96 h)

a

d

Secretory epithelium

Cartilage

Neuroepithelium

Matrigel

BSP-PAS

VN-PAS

Next, we tested whether cells maintained on PAS could be cryo preserved and thawed back onto PAS. Cell banks were generated from H7 hESCs that were maintained on BSP- or VN-PAS for nine passages. Cells were thawed from these banks and plated onto the same surface that had been used for culturing the cells before cryopreservation, and cultured for multiple passages. Recovery was similar to that of cell banks produced from cells cultured on Matrigel and thawed onto Matrigel (Supplementary Fig. 6a–c). Cultures grown on BSP-PAS or VN-PAS for up to 16 passages after thaw maintained undifferentiated hESC phenotype and normal karyotype (data not shown). To determine whether the PAS technology is scalable, we coated T75 flasks with PAS to produce VN-PAS T75 flasks (Fig. 3c–e). Uniform distribution of VN peptide was confirmed by enzyme-linked immunosorbent assay (ELISA) BCIP (5-bromo-4-chloro-3-indolyl phosphate)/NBT (nitro blue tetrazolium) staining of the T75 flask surface using anti-VN peptide antibodies (Fig. 3c). H7 hESCs were seeded into the VN-PAS T75 flasks, cultured for 4 d, and the distribution of hESC colonies was assessed by alkaline phosphatase activity (Fig. 3d). Uniform distribution of hESC colonies was obtained, demonstrating uniformity of the VN-PAS surface on T75 flasks. H7 hESCs cultured on VN-PAS T75 flasks exhibited typical hESC morphology (data not shown) and expressed the hESC marker OCT4 (Fig. 3e). In this report we describe the development and application of synthetic PAS for self-renewal of hESCs. We demonstrated successful maintenance of H1 and H7 hESCs on PAS for more than ten serial passages in xeno-free, defined medium. H7 and H1 hESCs cultured on PAS retain a stable doubling time, typical hESC morphology, expression of stem cell markers, in vitro and in vivo pluripotency and normal karyotype. Other studies have reported defined surfaces for hESC culture16,21. Although both of these studies demonstrated cell adhesion to a peptide-coated surface, interpretation of the specificity of the peptidecell interaction and clinical relevance are limited owing to the use of mouse embryonic fibroblast–conditioned medium or coating of the peptide surface with bovine serum. Furthermore, these studies VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

letters a

b

c

250 60% α-actinin +

Counts

200 150 100 50 0 100

101

250

d

did not demonstrate long-term self-renewal, pluripotency and karyotypic stability of hESCs in their culture systems. A defined medium with a matrix of four nonrecombinant proteins has been described10. The proteins used were highly complex chemical entities purified from sources such as human plasma or human placenta, which would limit the use of this matrix for large-scale clinical production of hESC-derived cells. More recently, a comparison of hESC long-term culture on several ECM proteins in defined and feeder cell–conditioned media22 showed that recombinant VN protein was the only defined, functional alternative to Matrigel for sustained self-renewal of three independent hESC lines. Our results demonstrate that a short peptide sequence of VN is sufficient to support long-term self-renewal of hESCs in a defined medium. The fully synthetic nature of PAS greatly reduces the risk of cellular contamination with animal-derived pathogens and provides a scalable, robust platform for the culture of hESCs. We demonstrated maintenance of H7 cells in several commercially available media suitable for hESC culture: the chemically defined media X-VIVO 10 + GF and mTeSR1, and Knockout SR-supplemented medium. In addition, we adapted the PAS coating to a variety of vessel formats, including T75 flasks, 96-well and 6-well plates. Notably, we showed that hESCs cultured on PAS can be cryopreserved and successfully thawed back onto PAS for further expansion or differentiation. This property is important for the production of cell banks, which will be required for the development of hESC-based therapeutics. Finally, we showed successful directed differentiation of H7 cells into functional cardiomyocytes on the same PAS, demonstrating the suitability of PAS for both expansion and differentiation of hESCs. In conclusion, we have developed fully defined, synthetic surfaces for culture vessels and shown their utility for sustained self-renewal of hESCs and directed differentiation of hESCs to cardiomyocytes in a defined medium. We believe that these surfaces will be useful for both research purposes and the production of hESC-derived cells for cellular therapies. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

102 FL1-H

103

104

60% Nkx 2.5 +

200 Counts

Figure 3 Direct differentiation of H7 hESCs into cardiomyocytes on PAS and scalability of PAS production. (a) Cardiomyocyte differentiation. H7 hESCs were maintained on BSP-PAS in defined X-VIVO 10 + GF medium for ten passages followed by a directed differentiation into cardiomyocytes on the same PAS. Confocal indirect immunofluorescence for cardiomyocyte-specific markers, α-actinin (green) and Nkx2.5 (in red). Scale bar, 50 μm. (b) Flow cytometry analysis for α-actinin and Nkx2.5 markers. (c) Uniform peptide distribution in VN-PAS T75 flasks. Surface-bound peptide was visualized by light-purple ELISA BCIP/NBT staining using anti VN-peptide antibodies. (d) Uniform distribution of alkaline phosphatase–positive H7 hESC colonies after 4 d in culture. H7 hESCs were cultured in X-VIVO 10 + GF in VN-PAS T75 flasks. After 4 d in culture cells were fixed and stained with BCIP/NBT to detect alkaline phosphatase activity. (e) H7 hESC colonies cultured in X-VIVO 10 + GF in VN-PAS T75 flasks express the nuclear hESC marker OCT4. Scale bar, 200 μm.

150 100 50 0 100

101

102 FL1-H

103

104

e

Note: Supplementary information is available on the Nature Biotechnology website. Acknowledgments We would like to thank our sponsors from Corning Incorporated, J. Mooney and M. McFarland, and from Geron Corporation, J. Lebkowski and A.H. Davies. From Corning Incorporated we thank A. Frutos for critical reading and helpful comments on the manuscript, M. Lewis for his chemistry expertise, T. Garvey for his assistance in PAS fabrication, P. Gagnon and P. Szlosek for substrate plate supply and T. Heck for his confocal imaging help. From Geron Corporation we thank K. Delavan-Boorsma and S. Edell for their support of in vivo studies. AUTHOR CONTRIBUTIONS Z.M. conceived, designed, performed and analyzed PAS validation with H7 cells. J.L.W. conceived, designed, performed and analyzed PAS validation with H1 cells, wrote supplementary materials. P.D.-S. and J.Y. designed, performed and analyzed cardiomyocyte differentiation on PAS. J.Y. designed, performed and analyzed EB differentiation. L.Q. assisted J.Y. in performing and analyzing cardiomyocytes and EB differentiation. C.A.P. performed teratoma formation studies. D.M.W., A.G.F. and Y.Z. conceived, designed, developed and fabricated PAS. D.M.W. identified peptides for the study, characterized surface peptide uniformity, contributed to design of peptide conjugation scheme. A.G.F. developed and optimized peptide conjugation and characterized surface peptide density. C.S. assisted Y.Z. in acrylate coating development and fabrication. A.M., J.N. and P.W. developed coating conditions for PAS in T75 format, fabricated PAS-T75. Z.M. and R.B. wrote the manuscript. J.P.B., S.P. and R.B. provided direction and guidance for the various areas of the project. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturebiotechnology/. Published online at http://www.nature.com/naturebiotechnology/ . Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Thomson, J.A. et al. Embryonic stem cell lines derived from human blastocysts. Science 282, 1145–1147 (1998). 2. Reubinoff, B.E., Pera, M.F., Fong, C.Y., Trounson, A. & Bongso, A. Embryonic stem cell lines from human blastocysts: somatic differentiation in vitro. Nat. Biotechnol. 18, 399–404 (2000).

609

letters 13. Pierschbacher, M.D. & Ruoslahti, E. Cell attachment activity of fibronectin can be duplicated by small synthetic fragments of the molecule. Nature 309, 30–33 (1984). 14. Nomizu, M. et al. Cell binding sequences in mouse laminin α1 chain. J. Biol. Chem. 273, 32491–32499 (1998). 15. O’Connor, M.D. et al. Alkaline phosphatase-positive colony formation is a sensitive, specific, and quantitative indicator of undifferentiated human embryonic stem cells. Stem Cells 26, 1109–1116 (2008). 16. Li, Y.J., Chung, E.H., Rodriguez, R.T., Meri, T.F. & Healy, K.E. Hydrogels as artificial matrices for human embryonic stem cell self-renewal. J. Biomed. Mater. Res. A 79, 1–5 (2006). 17. Hern, D.L. & Hubbell, J.A. Incorporation of adhesion peptides into nonadhesive hydrogels useful for tissue resurfacing. J. Biomed. Mater. Res. 39, 266–276 (1998). 18. Allegrucci, C. & Young, L.E. Differences between human embryonic stem cell lines. Hum. Reprod. Update 13, 103–120 (2007). 19. Pekkanen-Mattila, M. et al. Substantial variation in the cardiac differentiation of human embryonic stem cell lines derived and propagated under the same conditionsa comparison of multiple cell lines. Ann. Med. 22, 1–15 (2009). 20. Laflamme, M.A. et al. Cardiomyocytes derived from human embryonic stem cells in pro-survival factors enhance function of infarcted rat hearts. Nat. Biotechnol. 25, 1015–1024 (2007). 21. Derda, R. et al. Defined substrates for human embryonic stem cell growth identified from surface arrays. ACS Chem. Biol. 2, 347–355 (2007). 22. Braam, S.R. et al. Recombinant vitronectin is a functionally defined substrate that supports human embryonic stem cell self-renewal via alphavbeta5 integrin. Stem Cells 26, 2257–2265 (2008).

© 2010 Nature America, Inc. All rights reserved.

3. Guhr, A., Kurtz, A., Friedgen, K. & Loser, P. Current state of human embryonic stem cell research: an overview of cell lines and their usage in experimental work. Stem Cells 24, 2187–2191 (2006). 4. Amit, M. et al. Human feeder layers for human embryonic stem cells. Biol. Reprod. 68, 2150–2156 (2003). 5. Richards, M. et al. Comparative evaluation of various human feeders for prolonged undifferentiated growth of human embryonic stem cells. Stem Cells 21, 546–556 (2003). 6. Xu, C. et al. Feeder-free growth of undifferentiated human embryonic stem cells. Nat. Biotechnol. 19, 971–974 (2001). 7. Stojkovic, P. et al. Human-serum matrix supports undifferentiated growth of human embryonic stem cells. Stem Cells 23, 895–902 (2005). 8. Lu, J., Hou, R., Booth, C.J., Yang, S.H. & Snyder, M. Defined culture conditions of human embryonic stem cells. Proc. Natl. Acad. Sci. USA 103, 5688–5693 (2006). 9. Li, Y., Powell, S., Brunette, E., Lebkowski, J. & Mandalam, R. Expansion of human embryonic stem cells in defined serum-free medium devoid of animal-derived products. Biotechnol. Bioeng. 91, 688–698 (2005). 10. Ludwig, T.E. et al. Derivation of human embryonic stem cells in defined conditions. Nat. Biotechnol. 24, 185–187 (2006). 11. Oldberg, A., Franzen, A. & Heinegard, D. The primary structure of a cell-binding bone sialoprotein. J. Biol. Chem. 263, 19430–19432 (1988). 12. Suzuki, S., Oldberg, A., Hayman, E.G., Pierschbacher, M.D. & Ruoslahti, E. Complete amino acid sequence of human vitronectin deduced from cDNA similarity of cell attachment sites in vitronectin and fibronectin. EMBO J. 4, 2519–2524 (1985).

610

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

ONLINE METHODS

© 2010 Nature America, Inc. All rights reserved.

Reagents. Cell biology reagents. X-VIVO10 (Lonza); hrbFGF, hrTGF-β1, Activin-A, BMP4 (R&D Systems); KnockOut SR, KnockOut DMEM, FBS, NEAA, l-glutamine, RPMI-B27, DPBS, Collagenase IV (Invitrogen); Matrigel (BD Biosciences); DMSO (Sigma-Aldrich); HSA (Baxter); HEPES (Invitrogen); AttoPhos (Promega); BCIP/NBT(KPL); paraformaldehyde (Electron Microscopy Sciences); flow cytometery staining buffer (BD Pharmingen); propidium iodide, β-mercaptoethanol (Sigma-Aldrich); Hoechst (Molecular Probes); ULA plates (Corning); gelatin-coated plates (StemCell Technologies Inc.). Antibodies: OCT4, TRA-1-60, SSEA-4 (Millipore Chemicon), α-actinin (Sigma), Nkx2.5 (Santa Cruz Biotechnology), β-tubulin III (Covance), smooth muscle actin and HNF3β (R&D Systems); all corresponding secondary antibodies (Invitrogen/Molecular Probes); custom rabbit anti-serum for BSP and VN peptides (Anaspec). Chemistry reagents. 2-Hydroxyethyl methacrylate, 2-carboxyethyl acrylate and tetra(ethylene glycol) dimethacrylate were obtained from Sigma-Aldrich. Durocur 1173 and Irgacure 819 were obtained from CIBA Specialty Chemicals. Ethanol was obtained from Pharmco Products. N-(3-Dimethylaminopropyl)N′-ethylcarbodiimide hydrochloride (EDC), N-hydroxysuccinimide (NHS), ethanolamine, and phosphate buffer were purchased from Sigma-Aldrich. N,N-dimethylformamide and hydrochloric acid were purchased from Fisher Scientific. Custom peptides were obtained from American Peptide Company. All materials were used as received. Acrylate polymer coating process. A mixture of 2-hydroxyethyl methacrylate, 2-carboxyethyl acrylate, and tetra(ethylene glycol) dimethacrylate was dissolved in ethanol. Darocur 1173 and Irgacure 819 were added as photoinitiators. The total monomer concentration was about 1%. The monomer solution was dispensed into plasma-treated multi-well plates and let spread to cover the culture area. Then the plates were dried in a vacuum chamber and cured in a nitrogen environment using Xenon pulsed UV light for 1 min with a dose of ~2 J/cm2. This resulted in culture ware with a swellable (meth) acrylate surface. Peptide-acrylate surface (PAS) preparation. The swellable (meth) acrylate surface was activated for 1 h with 1:1 EDC:NHS in DMF. After aspiration of the activation solution, the surface was treated for 1 h with 1 mM of the desired peptide in phosphate buffer. Finally, after aspiration of the peptide solution, the surface was treated for 1 h with ethanolamine adjusted to pH 8.0–8.5 with hydrochloric acid. These steps resulted in culture ware with a peptide acrylate surface. All reactions were performed at room temperature (20–25 °C). Plates were ethanol sanitized before cell culture. VN-PAS plates are available commercially from Corning. Peptide density assessment by fluorescent peptide labeling. The gradient peptide density 6-well plates were fabricated as described above for PAS. To track the change in conjugated peptide surface density, BSP peptide conjugation solution was supplemented with a rhodamine-labeled peptide (Rhod-P) using 1:0.0025 BSP:Rhod-P molar ratio. A dilution series of peptide conjugation solutions was implemented (1.0, 0.75, 0.5, 0.25, 0.125, 0.0625 mM) to vary conjugated peptide surface density. The average fluorescence intensity of the surfaces on a Cy3 channel was acquired using NovaRay microplate scanner (Alpha Innotech) at 999 ms acquisition time. Four independent measurements were performed for each surface. The average fluorescence intensity was analyzed with GenePix Pro software (MDS Analytical Technologies). hESC culture. H1 and H7 hESCs were maintained in chemically defined medium as previously described17. Briefly, cells were cultured on Matrigelcoated TCT plastic (1:30 dilution in KnockOut DMEM) in X-VIVO 10 basal medium supplemented with 80 ng/ml hrbFGF and 0.5 ng/ml hrTGF-β1 (X-VIVO 10 + GF medium). Cultures were passaged every 4–6 d, as cells become ~80% confluent, by incubation with 200 U/ml collagenase IV, followed by brief Ca2+/Mg2+ free phosphate buffered saline (DPBS) wash and gentle scraping. Seeding density was ~100,000 cells/cm2. Cells received fresh medium every day, except for the day after split. Cultures were routinely examined by flow cytometry for the expression of hESC markers, and cytogenetic analysis by G-banding for normal karyotype.

doi:10.1038/nbt.1629

hESC attachment and short-term growth assay. H1 and H7 cells were seeded at 35,000 cells/well in X-VIVO 10 + GF medium into 96-well plates with different PAS or Matrigel as control surface. After 48 h cells were fixed with 4% paraformaldehyde (PFA) for 10 min at room temperature (20–25 °C), washed with DPBS and incubated with 100 μl of soluble alkaline phosphatase substrate, AttoPhos reagent (1:3 dilution in DPBS) for 10 min. Fluorescent intensity at 485/535 nm was obtained using a Victor3 microplate reader (Perkin Elmer). Long-term culture of hESCs on synthetic surfaces. For each passage, H1 and H7 cells were sub-cultured in 6-well plates with PAS and Matrigel at the density of 100,000 cell/cm2 in X-VIVO 10 + GF medium. Cultures were fed every day, except for the day after split. Microscopic examination of cell and colony morphology was done daily. Cell viability and number were assessed at the end of each passage, by treating duplicate wells with collagenase IV, harvesting the cells with 0.25% trypsin/EDTA treatment followed by cell counting with automated cell number/viability analyzer, Vi-Cell (Beckman Coulter). Expression of hESC markers was assessed by flow cytometry at the end of each passage for more than ten consecutive passages. Immunofluorescent staining of cells on all surfaces was done as described below. To monitor genomic integrity, cell samples for all experimental conditions (PAS and Matrigel) were submitted for karyotyping analysis by G-banding (cytogenetics laboratories) at passages 0, 5 and 10. hESC doubling time was calculated as follows: doubling time (hours) = h × ln(2)/ ln(c 2 / c1) where h = hours in culture, c2 = cell number per vessel at harvest and c1 = cell number per vessel seeded. Cryopreservation of hESCs. To cryopreserve H7 hESCs the cells were harvested using collagenase and scraping as described above. Cells were then centrifuged at 300g in a Beckman Allegra 6 centrifuge for 5 min. The supernatant was aspirated and the cells resuspended in cryopreservation medium (X-VIVO 10 + 10% DMSO/1%HSA/0.1 M HEPES) at a final concentration of 5 × 106 or 1.5 × 107 cells/ml. 2 ml standard cryovials were then filled with 1.0 ml cell suspension. The vials were transferred to a controlled rate freezer and frozen at a rate of 1 °C per minute to −45 °C, and then at 10 °C per minute to −90 °C. The cells were then transferred to permanent storage in the vapor phase of liquid nitrogen. Peptide uniformity assessment by ELISA. All steps were performed at room temperature (20–25 °C) unless otherwise noted. The peptide acrylate surface was blocked for 1 h with 3% BSA in PBS, followed by incubation with anti-serum diluted with 1% BSA in PBS for 1 h. The anti-serum solution was aspirated and the surface was washed with 0.05% Tween 20 in PBS, followed by 1 h incubation with corresponding secondary antibody in 1% BSA in PBS. After washing with 0.05% Tween 20 in PBS, the surface was incubated for 1 h with BCIP/NBT, precipitating alkaline phosphatase substrate, at 37 °C followed by washing with water. Plates were dried before scanning. Flow cytometry. Live cells were used for the analysis of surface markers (TRA-1-60, SSEA-4). Cells stained with ethidium monoazide bromide (EMA), followed by PFA fixation and methanol permeabilization, were used for intra cellular markers (OCT4, Nkx2.5, α-actinin). Briefly, at the end of each passage, cells were harvested with 0.05% trypsin/EDTA. For each sample, 5 × 105 cells were incubated for 15 min in blocking buffer (10% heat-inactivated goat serum in DPBS) followed by 30 min incubation with 0.5 μg of primary antibody or corresponding isotype control in blocking buffer. After washing with 2 ml of staining buffer (2% heat inactivated FBS in DPBS), cells were incubated for 30 min with 0.25 μg of corresponding secondary antibody in staining buffer. Before acquisition, live cells were stained with propidium iodide (PI; 2 μg/ml in staining buffer) for 5 min, for viability assessment. 30,000 total events (gating was set for PI- or EMA-negative cell population) were acquired for each sample using FACS Calibur (BD Biosciences). All analyses were done using CellQuest Pro software (BD Biosciences). Immunocytochemistry. At the end of each passage, cells were fixed in 6-well plates with 4% PFA for 10 min at room temperature (20–25 °C). After DPBS wash, cells were permeabilized with 100% ethanol for 2 min, followed by

nature biotechnology

locking in 5% heat-inactivated (HI) FBS in DPBS for 1 h. Cells were exposed b to primary antibody or corresponding isotype control (in 2% HI-FBS in DPBS) for 1 h at room temperature (20–25 °C), followed by DPBS wash and treatment with corresponding secondary antibody (in 2% HI-FBS plus 1:2,000 dilution of Hoechst) for 30 min at 37 °C. After brief DPBS wash, cells were processed for fluorescent microscopy using Zeiss Axiovert 200M microscope with AxioVision Rel. 4.7.1 acquisition software or Zeiss LSM 510 confocal microscope with Zeiss Zen 2007 acquisition software.

Cardiomyocyte differentiation. H7 hESCs were differentiated into cardiomyocytes using a direct differentiation protocol, as described20. Briefly, undifferentiated H7 hESCs maintained on BSP-PAS for 14 passages were cultured on BSP-PAS in serum replacement medium (SRM; KnockOut DMEM + 20% KnockOut SR, 1 mM l-glutamine, 1% NEAA, 0.1 mM 2-ME, 80 ng/ml hrbFGF, 0.5 ng/ml hrTGF-β1) for 1 week (adaptation phase) before sub-cultivation on BSP-PAS at the density of 100,000 cell/cm2. Cells were fed daily with SRM for 6 d. Cardiac differentiation was induced by sequential exposure of cells to 100 ng/ml of activin A for 24 h, followed by 10 ng/ml BMP4 for 4 d in RPMI-B27 medium. Cells were fed every 2–3 d with RPMI-B27 medium without growth factors for additional 2–3 weeks. Widespread spontaneous beating activity was typically observed by 12 d post activin A induction. Differentiated cells were assessed for cardiomyocyte-specific markers (α-actinin, Nkx2.5) by flow cytometry and immunocytochemistry analyses as described above.

© 2010 Nature America, Inc. All rights reserved.

In vivo pluripotency and immunohistochemistry. For in vivo pluri potency analysis, 1 × 10 7 hESCs were injected in a single intramuscular site in the flank of SCID/bg mice. When palpable masses were identified at the injection site, animals were euthanized and the hind limb was fixed in 10% neutral formalin, embedded in paraffin, sectioned and stained with hematoxylin and eosin using standard procedures. Tissues were examined using an Axioskop 2 Plus photomicroscope (Zeiss) and images were captured using a Penguin 600CL digital camera (Pixera); images were sized

and balanced for brightness and contrast using Photoshop (v. 7.0.1.; Adobe Systems) without additional image manipulation.

nature biotechnology

doi:10.1038/nbt.1629

letters

Long-term self-renewal of human pluripotent stem cells on human recombinant laminin-511

© 2010 Nature America, Inc. All rights reserved.

Sergey Rodin1, Anna Domogatskaya1, Susanne Ström2, Emil M Hansson3, Kenneth R Chien3,4, José Inzunza5, Outi Hovatta2 & Karl Tryggvason1 We describe a system for culturing human embryonic stem (hES) cells and induced pluripotent stem (iPS) cells on a recombinant form of human laminin-511, a component of the natural hES cell niche. The system is devoid of animal products and feeder cells and contains only one undefined component, human albumin. The hES cells self-renewed with normal karyotype for at least 4 months (20 passages), after which the cells could produce teratomas containing cell lineages of all three germ layers. When plated on laminin511 in small clumps, hES cells spread out in a monolayer, maintaining cellular homogeneity with approximately 97% OCT4-positive cells. Adhesion of hES cells was dependent on a6b1 integrin. The use of homogeneous monolayer hES or iPS cell cultures provides more controllable conditions for the design of differentiation methods. This xeno-free and feederfree system may be useful for the development of cell lineages for therapeutic purposes. Establishment of a chemically defined xeno-free (animal substance– free) and feeder cell–free environment that supports the self-renewal of hES cells has been a major goal in the field since hES cells were first derived and cultured on fetal mouse fibroblasts1. The presence of animal proteins, undefined human proteins or feeder cells in human stem cell cultures can cause problems related to immunogenicity, microbial and viral contamination and variability of experimental results2–4. A chemically defined cell culture system devoid of nonhuman substances and feeder cells would benefit human stem cell research and support the production of cells for future therapy. Extracellular matrix proteins, particularly basement membrane components, are an important part of in vivo niches for the differentiation, phenotype maintenance and function of many types of somatic and stem cells. These proteins have been shown to influence cellular differentiation, adhesion, proliferation, migration and self-renewal of cells. Laminins, the main component of basement membranes, are a family of heterotrimeric glycoproteins composed of α, β and γ chains that exist, respectively, as five, three and three genetically distinct types forming 15 different combinations in human tissues 5.

They are named according to chain composition; for example, LN-511 consists of α5, β1 and γ1 chains. Different laminins show various spatio-temporal expression patterns as well as tissue-specific locations and functions. Thus, LN-211 and LN-221 are primarily present in basement membranes of muscle cells and motor-neuron synapses6, LN-111 is restricted to the early embryo and certain epithelial cells7, LN-332 is specific for subepithelial basement membranes8, LN-411 is located in subendothelial basement membranes9 and LN-511 is practically ubiquitous10. Laminins are the first matrix proteins expressed in the embryo11; they are already apparent at the two-cell stage12. Matrix proteins have been used as coating substrata for in vitro cultures of hES cells in numerous studies, but they have usually been applied as undefined protein mixtures13,14, with undefined media2 or in short-term experiments15,16. Matrigel, a mouse tumor extract, is the most common non–feeder cell coating used for hES cell cultures. Matrigel contains mainly LN-111, type IV collagen, perlecan and nidogen, but also growth factors, and its composition varies from batch to batch. Apart from mouse LN-111, laminins are difficult or impossible to isolate in native form from tissues, and they can be produced only in minute amounts from normal cells. Only recently have some human laminins been produced as recombinant proteins17–19. We have shown that LN-511 and LN-332 support proliferation of mouse ES cells during long term passage, but that only LN-511 facilitates expansion of pluripotent cells that generate germline-competent chimeric mice upon blastocyst injection20. Mouse ES cell results cannot be extrapolated directly to human cells, as expression of cytokines, cell cycle regulation and mechanisms of self-renewal differ between human and mouse ES cells21,22. One group has reported14 a feederfree and xeno-free hES cell culture system containing human albumin and a culture dish–coating mixture consisting of collagen IV, vimentin, fibronectin and human placenta laminin, which includes partially degraded polypeptides from several laminin types and other basement membrane proteins23, but the component(s) providing selfrenewal remain unknown. We cultured hES cell lines HS420, HS207, HS401 (ref. 24) and two human iPS cell lines, BJ#12 and LDS1.4, on human recombinant

1Division

of Matrix Biology, Department of Medical Biochemistry and Biophysics, Karolinska Institute, Stockholm, Sweden. 2Division of Obstetrics and Gynecology, Department of Clinical Sciences, Intervention and Technology, Karolinska Institute and Karolinska University Hospital, Huddinge, Stockholm, Sweden. 3Cardiovascular Research Center, Massachusetts General Hospital, Charles River Plaza, Boston, Massachusetts, USA. 4Department of Stem Cell and Regenerative Biology, Cambridge, Massachusetts, USA. 5Department of Biosciences and Nutrition, Karolinska Institute, Novum, Huddinge, Karolinska Hospital, Huddinge, Stockholm, Sweden. Correspondence should be addressed to K.T. ([email protected]). Received 15 December 2009; accepted 8 March 2010; published online 30 May 2010; doi:10.1038/nbt.1620

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

611

letters

Cell contact area, relative

b

LN-511

3 P < 5%

P < 0.1%

2 74 1

0

93 45

LN-511

c

LN-332

MG

19

LN-111

25 PL

13 LN-411

Expression of laminin chains α1 α2 α3 α4 α5

β1 β2 β3

γ1 γ2 γ3

200 bp

LN-511 for long periods. The media used were O3, a variant of the chemically defined mTeSR1 medium14 containing BSA as the only animal-derived component; and H3, a variant of chemically defined, xeno-free TeSR1 medium 14 containing human serum albumin. Control cells were cultured on Matrigel in O3 or on feeder cells in a serum replacement–based medium. To compare adhesion properties of hES cells, we performed cell adhesion assays on LN-511, LN-332, LN-411, LN-111, Matrigel or poly-d-lysine substrata (Fig. 1a,b). The average contact area of an adherent hES cell grown on LN-511 was about 1.6 times larger than that of cells plated on Matrigel and about 1.2 times larger than that of cells plated on LN-332 (Fig. 1b). hES cells spreading on the other coatings had significantly smaller adherence areas than on LN-511. Several hES cell lines express laminin α5, β1 and γ1 chains16,25. To determine whether this is a unique property of those lines, we performed RT-PCR on cDNA of the HS207, HS420 and HS401 lines. Transcripts for laminin α5, β1 and γ1 chains along with α1, α2 and β2 were detected, demonstrating that LN-511 is expressed by all three lines (Fig. 1c). In agreement with previous reports, expression of α3 and β3 laminin chains was not observed, suggesting that LN-332 is not produced by pluripotent hES cells. Adhesion-blocking assays with function-blocking antibodies showed that α6 and β1 integrins were the most important ones for LN-511 binding (Fig. 2a). To further explore integrin expression of hES cells, we immobilized antibodies to integrin on plastic and identified antibodies that could bind and maintain attachment of hES cells. β1 integrin antibodies provided the strongest adhesion, whereas

β2, β3 and β4 antibodies bound with 19% efficacy or less (Fig. 2b). Immunofluorescence staining confirmed expression and colocation of α6 and β1 integrin subunits in undifferentiated (SOX2-positive) hES cells (Fig. 2c). The three hES cell lines HS207, HS420 and HS401 cultured on LN-511 in O3 or H3 media proliferated robustly for at least 28 passages (5–6 months in culture). The cells could be passaged in small clumps every 6–7 d in 1:2 to 1:6 ratios; the cells showed similar proliferation rates and phenotypes in both media. hES cells grown on LN-511 proliferated at a stable rate similar to that of cells grown on Matrigel (Fig. 3a). Karyotypes were normal after 20 passages (Supplementary Fig. 1). Immunofluorescence and RT-PCR analyses revealed that hES cells maintained high expression levels of pluripotency markers, such as OCT4, Nanog and SOX2 (Fig. 3b,c). Expression of pluripotency markers was quantified by real-time quantitative RT-PCR and quantitative western blot analysis and compared between cells grown on LN-511, Matrigel or feeder cells (Fig. 3d,e). hES cells grown on LN511 showed stable expression levels of OCT4, SOX2 and Nanog that were higher than those in cells plated on Matrigel but similar to levels when grown on feeder cells. The majority of hES cells cultured on LN-511 expressed the pluripotency markers OCT4, SSEA-4, TRA-1-60 and TRA-1-81 while expressing only a small amount of SSEA-1 (Fig. 3f and Supplementary Fig. 2). The cells cultured on Matrigel had slightly lower numbers of OCT4-positive and SSEA-4–positive cells (Supplementary Fig. 3). The hES cells were passed in small clumps, not as cell suspensions. However, on the following day they had already spread over the surface as a monolayer (appearing as thin disks), suggesting that affinity to the LN-511–coated surface was similar to or even higher

a 60%

612

P < 1%

20% 0% –20% –40%

Figure 2 Integrin receptors on hES cell surface and their role in hES cell adhesion. (a) Adhesion-blocking experiment: inhibition of hES cell adhesion to LN-511 by different integrin antibodies. Bars represent inhibition by antibodies to chains indicated below graph (all from Millipore). IgG was used as a control for uninhibited cell adhesion. Error bars show s.e.m. (n = 4). Statistical significance (P ) was calculated by Student’s t-test. (b) Adhesion of hES cells to surfaces coated by different integrin antibodies. Bars represent adhesion with antibodies to chains indicated below. Error bars show s.e.m. (n = 4). P calculated by Student’s t-test is shown; **P < 0.01. (c) Immunofluorescence: integrin α6 coexpression with β1 integrin subunit in pluripotent (SOX2-positive) hES cells cultured on LN-511. Scale bars, 37 mm.

P > 80%

40%

b Adhesion, %

© 2010 Nature America, Inc. All rights reserved.

Figure 1 Adhesion of hES cells to different coatings and expression of laminin chains in hES cells. (a) Crystal violet staining of hES cells adherent to LN-511 and LN-111. Scale bars, 220 mm (insets, 55 mm). (b) Contact area of ES cells with different adhesive substrata. Bars represent average relative contact area (compared with the cells plated on poly-d-lysine). Statistical significance (P) was calculated by Student’s t-test. Error bars show s.e.m.; number inside each bar indicates number of independent measurements. LN, laminins; MG, Matrigel; PL, poly-d-lysine. (c) RT-PCR analysis of total RNA isolated from HS420 cells. Primer sets for all known laminin chains were used. bp, base pairs.

LN-111

Inhibition, %

a

β1

α6

α1

α2

α3

α4

α5

αV αVβ3 β2

β3

β4

c

150%

** 100%

β1α6 P > 25%

50% SOX2 0%

β1

β2

β3

β4

100% 19% 4%

4%

lgG 0%

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters

N, doublings

20

b

LN-511_1 LN-511_2 Mg_1 Mg_2

c

Magnification: x20, bar size: 0.15 mm Nanog

SOX2

DAPI

G AP O DH C T N 4 an Br og ac h AF yur P y SO X PA 1 X6

a

Feeders MG, p.7 in O3

10 OCT4

SOX2

DAPI

LN-511, p.8 in H3 LN-511, p.27 in O3

30

20

e

100

60 40

in

H

3

O 3

1

in

101

102 FL1-Height

103

104

LN

-5

11

,p

,p

100

.1

.8

in 0 LN

-5

,p

.2

.1 ,p

11 -5

LN

11

3 6

in 11

-5

LN

LN

-5

11

,p

.1

0

.4 ,p

in

O

O

.7

in

,p G M

11

H 3

0

O 3

0

3

20

.4

0.5

-5

OCT4 (97%)

80 Counts

1.0

LN

.8 ,p

1.5

er s

H 3

3

in

O

3

in 0 .2

11 -5

,p

LN

-5

11

.1

6

in

O

3

3

O in 0

.1

,p 11

-5

,p

,p

11 -5

LN

LN

3

O

O

in .5

in 11

M

-5

G

p.

7

Fe

LN

LN

er s

0

2.0

,p

0.5

OCT4 SOX2

G

1.0

ed

mRNA expression level, relative

© 2010 Nature America, Inc. All rights reserved.

OCT4 Nanog

f 2.5

ed

T, days

M

10

Fe

0

d

Protein expression level, relative

0

Figure 3 Representative immunostaining analysis, RT-PCR, fluorescence-activated cell sorting (FACS) analysis, real-time quantitative RT-PCR and quantitative western blot analysis of HS207 cultured on LN-511, either in O3 medium or in H3 medium free from any animal-derived components. (a) Growth curves for hES cells cultured in O3 medium on LN-511 and Matrigel. The cells were passaged as described in the Online Methods for the long-term experiment. After each TrypLE Express treatment and subsequent washing, one-third of the cells were plated in clumps on fresh LN-511– or Matrigel-coated dishes. The rest were dissociated into single-cell suspension and counted. Two independent duplicate experiments were performed for each coating. After the fifth passage, a portion of the cells were fixed and analyzed by immunofluorescence staining, confirming that the majority of the cells still expressed Nanog, a marker of pluripotency. (b) Immunostaining of HS207 cells with antibodies to Nanog, SOX2 and OCT4 after 20 passages (6 months) on LN-511 in O3 medium. Right panels show nuclear 4,6-diamidino-2-phenylindole (DAPI) staining. Scale bars, 0.15 mm. (c) RT-PCR analysis of total RNA isolated from H207 cells grown on feeder cells (Feeders), on Matrigel after 7 passages in O3 medium (MG, p.7 in O3), on LN-511 after 8 passages in H3 medium (LN-511, p.8 in H3) and on LN-511 after 27 passages in O3 medium (LN-511, p.27 in O3). Primer sets were designed for pluripotency markers OCT4 and NANOG, along with differentiation markers Brachyury, α-fetoprotein (AFP), SOX1 and PAX6, and for a housekeeping gene encoding glyceraldehyde-3-phosphate dehydrogenase (GAPDH). (d) Real-time quantitative RT-PCR analysis was used to measure numbers of mRNA transcripts of the pluripotency markers OCT4 and NANOG at different time points in HS207 cells cultured on LN-511 and on Matrigel; values shown are normalized to OCT4 and NANOG expression levels in control HS207 cells cultured on feeder layer (Feeders). Number of passages, adhesion surface and medium are denoted as in c. Error bars show 95% confidence intervals. (e) Expression of pluripotency markers OCT4 and SOX2 in HS207 cells cultured on feeder cells, Matrigel and LN-511 at different time points and in different media (denoted as in c) was measured by western blotting and quantified by densitometry. Error bars represent range. (f) FACS analysis of HS207 cells after 25 passages on LN-511 in O3 medium for OCT4, a marker of pluripotency. The percentage of positive cells is listed in parentheses.

than the adhesion between cells. Usually the cells first formed a monolayer, but later could grow on top of each other. To assess the level of spontaneous differentiation in hES cultures on LN-511, Matrigel or feeders, we compared expression levels of the differentiation markers PAX6, SOX17 and SOX7. Real-time quantitative RT-PCR revealed similar levels of expression of all three markers in LN-511 cultures after 20 passages (4 months) and Matrigel cultures after 4 passages (1 month) (Supplementary Fig. 4). To explore whether LN-511 can be used to derive new hES cell lines, we isolated the inner cell masses (ICMs) of day 6 or day 7 blastocysts and plated them on LN-511. In H3 medium, in a completely xeno-free environment, 10 of 12 plated ICMs successfully attached, of which five gave outgrowths (Supplementary Fig. 5 and Supplementary Table 1). In mTeSR1 medium supplemented with LN-511 (1 μg ml−1 in solution in addition to the laminin-511 used for coating the dish), all nine ICMs used in the experiment attached and gave outgrowths (Supplementary Table 1). HS207, HS420 and HS401 cells cultured, respectively, for 15, 20 and 20 passages on LN-511 in O3 medium, and HS207 cells cultured for 23 passages in H3 medium, formed teratomas after they were grafted into the testes of severe combined immunodeficiency (SCID) mice. Histological examination confirmed the ability of the cells to nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

ifferentiate into cells of all three germ lineages of the human embryo d (Fig. 4a–d). Cells of all three hES lines grown for 20 passages on LN-511 in O3 medium expressed markers of mesoderm (smooth-muscle actin), ectoderm (nestin and MAP-2) and endoderm (α-fetoprotein) (Fig. 4e), providing additional evidence of pluripotency. To determine whether LN-511 generally supports self-renewal of hES and iPS cells, we cultured the widely used H1 and H9 hES lines as well as two iPS cell lines on LN-511 in O3 or mTeSR1 media. The H1 and H9 cells had phenotypes and proliferation rates similar to those of HS207, HS420 and HS401 cells under the same conditions. Immunofluorescence analysis revealed that the H1 and H9 cells maintained expression of pluripotency markers, such as OCT4, Nanog, SOX2 and SSEA-4 (Fig. 5a and Supplementary Fig. 6a) after five passages (1 month). Similar results were obtained with the BJ#12 (ref. 26) and LDS 1.4 iPS cell lines, which expressed pluripotency markers OCT4, Nanog, SOX2 and TRA-1-60 (Fig. 5b and Supplementary Fig. 6b,c). The level of Nanog expression was similar to that of the same cells grown on Matrigel (Supplementary Fig. 6c). The present results demonstrate that LN-511 provides an artificial niche supporting the survival and self-renewal of pluri potent human stem cells in culture in a xeno-free environment for 613

letters a

b

c

d

Figure 4 Pluripotency of HS207 cells after extensive passaging on LN-511. Teratomas containing components of the three germ layers were formed after HS207 cells that had been cultured for 15 passages on LN-511 were injected subcutaneously into SCID mice. (a) Cartilage, stained with hematoxylin and eosin (HE). Magnification, ×100. (b) Developing neural tissue and intestinal endoderm, with HE and periodic acid-Schiff (HE-PAS) staining. Goblet cells are shown in red. Magnification, ×400. (c) Developing kidney glomerulus, HE staining. Magnification, ×400. (d) Retinal pigment epithelium, HE staining. Magnification, ×400. (e) Immunostaining of embryoid bodies formed from HS207 cells after 20 passages on LN-511 revealed expression of markers for the three embryonic cell layers: smooth-muscle (SM) actin, nestin, MAP-2 and AFP. Scale bars, 55 mm.

© 2010 Nature America, Inc. All rights reserved.

e SM Actin

Nestin

MAP-2

AFP

DAPI

DAPI

DAPI

DAPI

pivotal role in the binding of hES cells to the matrix (Fig. 2a,b). We surmise that hES cells abundantly expressing α6β1 integrins can attach quickly and migrate efficiently an LN-511–coated surface, which, in turn, facilitates their self-renewal. Hence, the role of LN-511 could be to provide hES cells with focal adhesion contacts to the surface and to enable mobility. The fact that LN-511 expression is not restricted to early embryos, but rather LN-511 is a ubiquitous basement-membrane component, supports this hypothesis. There is a great need for chemically defined, xeno-free, feederfree culture systems for hES cells28. The human LN-511 coating may therefore have considerable advantages for the standardization of stable hES cell cultures. A chemically defined substratum is preferable to feeder cells because feeders vary in their production of bioactive molecules, such as cytokines, growth factors and other unknown proteins, and they carry a risk of microbial and viral contamination. A defined environment for culturing hES cells may be helpful in investigating the molecular mechanisms of differentiation and in designing more reproducible methods for differentiation. HES cell conditions based on recombinant proteins may also be more acceptable to regulatory authorities in many countries.

at least 20 passages, or 4 months. LN-511 appears to be part of the stem cell niche in the human embryo, as it is expressed in the ICM of blastocysts27, from which hES cells originate. Furthermore, LN-511 is present in hES cell colonies cultured in vitro on feeder cells25, and hES cells themselves express LN-511 (Fig. 1c). Thus, LN-511 provides a biologically relevant coating matrix for the self-renewal of hES cell in vitro. It has been OCT4 SOX2 DAPI Nanog SOX2 DAPI a reported that hES cells remain pluripotent on several recombinant human laminins, such as LN-111, LN-332 and LN-511, for H9 96 h (ref. 16). However, we have shown that mouse pluripotent ES cells can survive and proliferate on both LN-511 and LN-332 for at least 169 d, although only cells cultured on LN-511 were able to generate germlineH1 competent chimeric mice20. Notably, we found that hES cells formed monolayers after being passaged in clumps to new LN-511–coated plates. This suggests that DAPI Nanog OCT4 Merge b LN-511 provides the cells with a migration potential in the absence of differentiation. Monolayers of hES cells should be beneficial BJ#12 for the development of differentiation procedures, as equal availability of soluble factors can create more homogeneous populations of differentiated cells. All hES cell LDS 1.4 populations contain a proportion of differentiated cells, probably owing to spontaneous differentiation, but in this study the proportion of undifferentiated hES cells grown on Figure 5 Immunostaining analysis of different hES and iPS cells grown on LN-511. (a) H1 and H9 LN-511 was high and stable throughout the cells after five passages (1 month) on LN-511 in O3 medium expressed pluripotency markers OCT4 whole experiment (Fig. 3f). (green), Nanog (green) and SOX2 (red). DAPI staining is in blue. Scale bars, 75 mm. (b) BJ#12 The results support previous findings 16 and LDS 1.4 iPS cells after five passages on LN-511 in mTeSR1 medium expressed Nanog (red) that β1 integrins, primarily α6β1, have a and OCT4 (green). Scale bars, 75 mm. 614

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

letters Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Note: Supplementary information is available on the Nature Biotechnology website.

© 2010 Nature America, Inc. All rights reserved.

Acknowledgments We thank A.-S. Nilsson and A.-M. Strömberg for excellent technical assistance, C. Cowan for collaboration in procuring the iPS cell lines, F. Holm and R. Bergström for their help with the cell cultures and D. Baker for carrying out karyotyping of hES cells. This work was supported in part by grants from the Knut and Alice Wallenberg Foundation (K.T.), the Novo Nordisk Foundation (K.T.), the Söderberg’s Foundation (K.T.), the Swedish Research Council (K.T., O.H.), the Swedish Cancer Foundation (K.T.), the Harvard Stem Cell Institute (K.R.C.) and the Leducq Foundation (K.R.C.). E.M.H. is a Wenner-Gren Foundation fellow. EU: ESTOOLS (O.H.) has not been used for the derivation of new hES cell lines. Author contributions S.R. and A.D. contributed to the production and purification of human recombinant laminins, conducted all in vitro experiments with the hES cells and contributed to the planning and design of experiments and to the writing of the manuscript. O.H. established and provided the hES cell lines and contributed to manuscript writing and karyotyping. S.S. contributed to the establishment of the new hES cell lines. E.M.H. and K.R.C. contributed to the iPS cell work. J.I. carried out the teratoma experiments in nude mice. K.T. planned and designed the project and contributed to the writing of the manuscript. COMPETING FINANCIAL INTERESTS The authors declare competing financial interests: details accompany the full-text HTML version of the paper at http://www.nature.com/naturebiotechnology/. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Reubinoff, B.E., Pera, M.F., Fong, C.Y., Trounson, A. & Bongso, A. Embryonic stem cell lines from human blastocysts: somatic differentiation in vitro. Nat. Biotechnol. 18, 399–404 (2000). 2. Xu, C. et al. Feeder-free growth of undifferentiated human embryonic stem cells. Nat. Biotechnol. 19, 971–974 (2001). 3. Hovatta, O. et al. A culture system using human foreskin fibroblasts as feeder cells allows production of human embryonic stem cells. Hum. Reprod. 18, 1404–1409 (2003). 4. Martin, M.J., Rayner, J.C., Gagneux, P., Barnwell, J.W. & Varki, A. Evolution of human-chimpanzee differences in malaria susceptibility: relationship to human genetic loss of N-glycolylneuraminic acid. Proc. Natl. Acad. Sci. USA 102, 12819–12824 (2005). 5. Aumailley, M. et al. A simplified laminin nomenclature. Matrix Biol. 24, 326–332 (2005). 6. Miner, J.H. & Yurchenco, P.D. Laminin functions in tissue morphogenesis. Annu. Rev. Cell Dev. Biol. 20, 255–284 (2004).

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

7. Ekblom, P., Lonai, P. & Talts, J.F. Expression and biological role of laminin-1. Matrix Biol. 22, 35–47 (2003). 8. Kallunki, P. et al. A truncated laminin chain homologous to the B2 chain: structure, spatial expression, and chromosomal assignment. J. Cell Biol. 119, 679–693 (1992). 9. Iivanainen, A. et al. Primary structure, developmental expression, and immunolocalization of the murine laminin alpha4 chain. J. Biol. Chem. 272, 27862–27868 (1997). 10. Miner, J.H., Lewis, R.M. & Sanes, J.R. Molecular cloning of a novel laminin chain, alpha 5, and widespread expression in adult mouse tissues. J. Biol. Chem. 270, 28523–28526 (1995). 11. Cooper, A.R. & MacQueen, H.A. Subunits of laminin are differentially synthesized in mouse eggs and early embryos. Dev. Biol. 96, 467–471 (1983). 12. Dziadek, M. & Timpl, R. Expression of nidogen and laminin in basement membranes during mouse embryogenesis and in teratocarcinoma cells. Dev. Biol. 111, 372–382 (1985). 13. Klimanskaya, I. et al. Human embryonic stem cells derived without feeder cells. Lancet 365, 1636–1641 (2005). 14. Ludwig, T.E. et al. Derivation of human embryonic stem cells in defined conditions. Nat. Biotechnol. 24, 185–187 (2006). 15. Braam, S.R. et al. Recombinant vitronectin is a functionally defined substrate that supports human embryonic stem cell self-renewal via alphavbeta5 integrin. Stem Cells 26, 2257–2265 (2008). 16. Miyazaki, T. et al. Recombinant human laminin isoforms can support the undifferentiated growth of human embryonic stem cells. Biochem. Biophys. Res. Commun. 375, 27–32 (2008). 17. Yurchenco, P.D. et al. The alpha chain of laminin-1 is independently secreted and drives secretion of its beta- and gamma-chain partners. Proc. Natl. Acad. Sci. USA 94, 10189–10194 (1997). 18. Doi, M. et al. Recombinant human laminin-10 (α5β1γ1). Production, purification, and migration-promoting activity on vascular endothelial cells. J. Biol. Chem. 277, 12741–12748 (2002). 19. Kortesmaa, J., Yurchenco, P. & Tryggvason, K. Recombinant laminin-8 (α4β1γ1). Production, purification, and interactions with integrins. J. Biol. Chem. 275, 14853–14859 (2000). 20. Domogatskaya, A., Rodin, S., Boutaud, A. & Tryggvason, K. Laminin-511 but not -332, -111, or -411 enables mouse embryonic stem cell self-renewal in vitro. Stem Cells 26, 2800–2809 (2008). 21. Ginis, I. et al. Differences between human and mouse embryonic stem cells. Dev. Biol. 269, 360–380 (2004). 22. Humphrey, R.K. et al. Maintenance of pluripotency in human embryonic stem cells is STAT3 independent. Stem Cells 22, 522–530 (2004). 23. Wondimu, Z. et al. Characterization of commercial laminin preparations from human placenta in comparison to recombinant laminins 2 (α2β1γ1), 8 (α4β1γ1), 10 (α5β1γ1). Matrix Biol. 25, 89–93 (2006). 24. Strom, S. et al. Mechanical isolation of the inner cell mass is effective in derivation of new human embryonic stem cell lines. Hum. Reprod. 22, 3051–3058 (2007). 25. Evseenko, D. et al. Identification of the critical extracellular matrix proteins that promote human embryonic stem cell assembly. Stem Cells Dev. 18, 919–928 (2009). 26. Maherali, N. et al. A high-efficiency system for the generation and study of human induced pluripotent stem cells. Cell Stem Cell 3, 340–345 (2008). 27. Klaffky, E. et al. Trophoblast-specific expression and function of the integrin alpha 7 subunit in the peri-implantation mouse embryo. Dev. Biol. 239, 161–175 (2001). 28. Unger, C., Skottman, H., Blomberg, P., Dilber, M.S. & Hovatta, O. Good manufacturing practice and clinical-grade human embryonic stem cell lines. Hum. Mol. Genet. 17, R48–R53 (2008).

615

© 2010 Nature America, Inc. All rights reserved.

ONLINE METHODS

hES and iPS cell cultures. hES cells of HS207, HS420 and HS401, originally derived in our laboratory at the Karolinska Institute, as described previously3,29, were cultured on LN-511–coated laboratory dishes in chemically defined O3 medium (a variant of mTeSR1 medium14; see below) and chemically defined, xeno-free H3 medium (a variant of TeSR medium14; see below) at 37 °C in 5% CO2. Clinical grade ≥96 pure human albumin was purchased from Octapharma AB. Initially, we transferred cells from the lines on LN-511 in small pieces from a feeder-cell layer by careful scratching using a sterile knife. Cells were fed once a day with fresh medium prewarmed in an incubator for 1 h, except for the first day after a passage, when only a few drops of fresh medium were added. Cells were routinely passed once every 6–7 d by exposure to TrypLE Express (GIBCO Invitrogen) for 1.5 min at room temperature. They were then washed twice on the dish with the medium, gently scraped, pipetted to break them into small pieces (not a single-cell suspension) and plated in a ratio of 1:2 or 1:3 (up to 1:6 if a large number of cells was needed). Control cells of the same line were cultured on Matrigel (BD Biosciences) in O3 medium as described previously14 and on mitotically inactivated human foreskin fibro blasts as described elswewhere29. Laboratory dishes were coated as previously described20. Before use, dishes were prewarmed in an incubator for 1 h and then carefully washed twice with the prewarmed medium. The derivation and extensive characterization of BJ#12 hiPS cells has been described previously26. Notably, BJ#12 hiPS cells were found to express pluripotency markers from the endogenous loci and lack expression of the viral transgenes. The pluripotency of BJ#12 cells was confirmed by in vivo (injection into SCID mice) and in vitro (embryoid bodies formation) experiments. The generation of LDS 1.4 cells will be described in detail elsewhere (E.M.H. and K.R.C., unpublished data). Briefly, fibroblasts were infected with Moloney-type retroviruses encoding OCT4, SOX2 and KLF4, and valproic acid was added during the reprogramming phase to enhance the efficiency. The LDS 1.4 cells showed a high degree of viral silencing (determined by quantitative RT-PCR analysis as described30) and expression levels of endogenous pluripotency genes that were comparable to levels seen in hES cells. When injected into SCID mice, the LDS 1.4 cells gave rise to teratomas that had contributions from all three germ layers. Laminins and other coating materials. Human recombinant LN-511 was produced in human embryonic kidney cells (HEK293; ATCC CRL-1573) sequentially transfected with full-length laminin γ1, β1 and α5 constructs, essentially as described previously18. Human recombinant LN-511 has recently become available from BioLamina. For protein production, the HEK293 cells were cultured in DMEM supplemented with GlutaMax I and 4.5 g l−1 glucose (GIBCO) for up to 6 d. The LN-511 molecules were affinity-purified using anti-Flag matrix (Sigma) as previously described18 and then characterized using 3–8% (Supplementary Fig. 7) and 4–15% gradient SDS-PAGE under reducing and nonreducing conditions. The proteins were visualized using SYPRO Ruby (Bio-Rad) protein staining and immunostaining of the chains on polyvinylidene difluoride membranes. To further characterize the protein, we performed western blot analysis with antibodies against the laminin α5, β1 and γ1 chains. The preparations contained all three chains of the right size, as described18. Human recombinant LN-411 was produced in the same way as LN-511, as described19. All other extracellular matrix (ECM) proteins were obtained as described previously20. Cell contact area measurement. MaxiSorp 96-well plates (Sarstedt) were coated with ECM proteins as previously described20 and blocked with 1% (wt/vol) BSA solution. Undifferentiated ES cells were split into single-cell suspension, filtered through a 40-μm sterile cell sieve, plated at a density of 700 cells mm−2 on ECM-coated plates and left to adhere for 1 h at 37 °C. Nonadherent cells were washed away, and adherent cells were fixed for 20 min with 5% glutar aldehyde (vol/vol), washed and stained with 0.1% crystal violet (wt/vol). Photos of six to ten random fields were taken, and the cell contact area of 13–93 cells was measured using Volocity imaging software (Improvision). To measure the cell area of unspread hES cells, the cells were plated on poly-d-lysine for 20 min, fixed and stained as described above. Adhesion-blocking assay using integrin antibody. Adhesion-blocking assays were performed as described previously20. Briefly, plates were coated with

nature biotechnology

LN-511 and blocked by 1% (wt/vol) heat-denatured BSA solution. ES singlecell suspension was incubated with function-blocking antibodies to integrin (concentration as recommended by supplier) for 30 min, plated on LN-511– coated plates and allowed to adhere for 1 h at 37 °C. Unattached cells were removed, the remaining adherent cells were fixed for 20 min with 5% (vol/vol) glutaraldehyde, washed and stained with 0.1% (wt/vol) crystal violet. After 1 h, we extracted crystal violet from cells using 10% (vol/vol) acetic acid and quantified it by measuring optical density at 570 nm. Assay of cell adhesion to surfaces coated by integrin antibodies. The assay was designed to identify integrin receptors that are expressed in sufficient amounts to retain cells attached to a surface coated with integrin-specific antibody. MaxiSorp 96-well plates (Nunc) were coated with purified integrin antibodies at a concentration of 10 μg ml−1 at 4 °C overnight and later washed and blocked with 1% (wt/vol) BSA solution. ES cells were plated on antibody-coated plates and allowed to adhere for 1 h at 37 °C. Unattached cells were removed, and the remaining cells were fixed, stained and quantified as described above. RT-PCR. Total RNA was isolated using the Absolutely RNA Microprep kit (Stratagene) according to the manufacturer’s instructions. cDNA was synthesized using 0.2 μg of total RNA in a 20-μl reaction mixture containing oligo(dT)12–18 primers and Superscript II reverse transcriptase (GIBCO Invitrogen), according to the manufacturer’s instructions. To compensate for variable cDNA yields, the amount of cDNA for each PCR reaction was calibrated using the expression level GAPDH as a standard. Amounts of cDNA yielding an equivalent amount of GAPDH PCR product (at 20 cycles; data not shown) were used for subsequent PCR reactions. To analyze expression of different markers of pluripotency or differentiation of hES cells, we amplified cDNAs using primers described in Supplementary Table 2. To analyze expression of different laminin chains, we used primers from ref. 25. All PCR reactions were run for 30 cycles (including those GAPDH PCRs that are shown in the figures) and were performed in 20 μl under standard conditions using 1 U of recombinant Taq DNA polymerase (GIBCO Invitrogen). The PCR products were analyzed on a 1.5% agarose gel containing ethidium bromide. For each RNA sample, RT-PCR without reverse transcriptase was performed to confirm that no genomic DNA was isolated. Immunofluorescence. For immunofluorescence studies, ES cells were cultured and fixed in 8-well slide chambers (BD Biosciences) or 96-well plate wells with 4% (wt/vol) paraformaldehyde, permeabilized with 0.1% (vol/vol) Triton-X and blocked with 10% (vol/vol) fetal bovine serum (FBS; GIBCO Invitrogen) in PBS containing 0.1% (vol/vol) Tween-20 (Sigma-Aldrich) for 1 h. Cells were incubated with primary antibody for 1.5 h at room temperature, and with secondary antibody and DAPI (Molecular Probes) for 40 min. Between incubations, specimens were washed with 0.1% (vol/vol) Tween-20 in PBS buffer three to five times. Specimens were preserved in fluorescence mounting medium (Dako), and observed under a fluorescence microscope (Leica). Real-time PCR quantification of mRNAs. Total RNA was isolated and cDNA was synthesized as described above for RT-PCR. Real-time quantitative RT-PCR Taqman assays were performed using the Applied Biosystems 7300 Real-Time PCR System. All reactions were done in quadruplicate with predeveloped gene expression assay mix (Applied Biosystems) containing primers and a probe for the mRNA of interest. Additional reactions for each experiment included predeveloped gene expression assay mix for GAPDH, used to normalize the RNA input. All data were analyzed with 7300 System SDS Software version 1.4. For Nanog mRNA quantification in iPS cells, SYBR green assays were used. In this case, total RNA was purified from each sample using an RNeasy Mini kit (Invitrogen). cDNA was synthesized using iScript (Bio-Rad), and quantitative PCR was performed on an Eppendorf Mastercycler using the HotStart-IT SYBR Green qPCR Master Mix (USB). Primer sequences are available upon request. Western blot and densitometry analysis. hES cells were collected, counted and pelleted by centrifugation, mixed with nonreduced SDS-PAGE sample buffer to equal concentrations of 2,000 cells μl−1 and sonicated five times for 15 s. Gradient 4–12% gels were used for SDS electrophoresis, and the proteins

doi:10.1038/nbt.1620

© 2010 Nature America, Inc. All rights reserved.

were transferred to PVDF membranes. Membranes were blocked using 5% (wt/vol dry) milk solution in PBS buffer with 0.1% (vol/vol) Tween for 2 h. Primary antibodies against OCT4 and SOX2 (both from Millipore) in 5% milk solution in PBS buffer with 0.1% Tween were incubated with the membranes overnight at 4 °C. After being washed four times, HRP-conjugated secondary antibodies in 5% milk solution in PBS with 0.1% Tween buffer (diluted 1:1,000) were incubated with the membranes for 40 min at room temperature and washed five times with PBS. Chemoluminescent horseradish peroxidase substrate from Amersham Biosciences was used for visualization. Films were scanned at 2,400 d.p.i. and analyzed with the ChemiImager5500 program (1DMulti Line densitometry mode). hES cells cultured on Matrigel and on feeder cells were used as positive controls. FACS analysis. Cells were removed from the culture dish with Trypsin-EDTA, dissociated into a single-cell suspension, resuspended in ice-cold FACS buffer (2% (vol/vol) FBS, 0.1% (wt/vol) sodium azide in Hank’s buffer). Cells were incubated with primary antibodies against SSEA-4, SSEA-1 (both from R&D Systems), Tra1-60 or Tra1-81 (both from Millipore) for 1 h on ice, then washed three times with ice-cold FACS buffer. They were then probed in FACS buffer with 1:400 dilutions of Alexa Fluor mouse secondary antibodies (GIBCO Invitrogen) for 30 min in the dark and washed four times. Control cells were incubated with mouse immunoglobulins and then with the secondary antibody as described above. Cells were analyzed on FACSCalibur Flow Cytometer (Becton Dickinson). Data were analyzed with CellQuest software (Becton Dickinson). OCT4 expression was analyzed as described14. Karyotyping. Karyotyping of the cell lines was carried out using standard Q-banding techniques at passage 20 on LN-511. Samples of cells were treated with colcemid KaryoMAX (0.1 μg ml−1; GIBCO Invitrogen) for up to 4 h, then dissociated with TrypLE Express (GIBCO Invitrogen). The cells were pelleted via centrifugation, resuspended in prewarmed 0.0375 M KCl hypotonic solution and incubated for 10 min. After centrifugation, the cells were resuspended in fixative (3:1 methanol/acetic acid). Metaphase spreads were prepared on glass microscope slides, G-banded by brief exposure to trypsin and stained with 4:1 Gurr’s/Leishmann’s stain (Sigma-Aldrich). A minimum of 10 metaphase spreads were analyzed, and an additional 20 were counted.

(smooth-muscle actin, nestin, MAP-2 and α-fetoprotein; all four antibodies were from Millipore) and analyzed as described above for immunofluorescence. Statistics. Statistical significance was determined with Student’s two-tailed t-test for unequal variances. O3 medium. O3 medium is a variant of mTeSR1. We prepared stock A by adding 165 mg of thiamine and 50 mg of reduced glutathione to 500 ml of distilled water (as described32, but without l-ascorbic acid). The distilled water was purchased from GIBCO Invitrogen. The solution was then filtered (0.22-μm filter), divided into aliquots and frozen at −20 °C. Stock B was prepared as described32, but without selenium, insulin and holo-transferrin. Then 6 mg of phenol red was added per 100 ml, and the stock was carefully stirred and filtered. Stock B could be stored at 4 °C up to 2 months. Stocks of transforming growth factor β1 (TGF-β1), pipecolic acid, GABA (γ-aminobutyric acid) and LiCl were prepared and stored as described32. To prepare 100 ml of O3 medium, DMEM-F12 medium was supplemented with 20 ml of stock B, 200 μl of TGF-β1 stock, 13 μl of pipecolic acid stock, 200 μl of GABA stock, 200 μl of LiCl stock, 1 ml of MEM nonessential amino acid solution (GIBCO Invitrogen), 1 ml of 200 mM l-glutamine solution (GIBCO Invitrogen) and 2 ml of insulin-transferrin-selenium supplement (GIBCO Invitrogen). To compensate the salt balance and to adjust the pH of the medium, 145 mg of NaCl and 56 mg of NaHCO3 were added. The solution was thoroughly mixed and the pH of the medium at room temperature was adjusted to 7.4 using 10 N NaOH. The solution was filtered using a 0.22-μm filter, and 200 μl of chemically defined lipid concentrate (GIBCO Invitrogen) was added. O3 medium could be stored at 4 °C up to 1 month. Before use, the medium was supplemented with 96 ng ml−1 recombinant human FGF basic (R&D Systems) and 40 μg ml−1 ascorbic acid (Sigma-Aldrich).

Teratoma formation. Teratoma-formation experiments were done as described29,31 by implantation of approximately 106 cells beneath the testicular capsule of a young (7-weeks-old) SCID mouse. Three animals per cell line were used. Teratoma growth was determined by palpation every week, and the mice were killed 8 weeks after the implantation. The teratomas were fixed, and sections were stained with HE or HE-PAS. Tissue components of all three embryonic germline layers were present in the stained sections. All animal experiments were performed at the infection-free animal facility of the Karolinska University Hospital, where they were approved by the ethical committee.

H3 medium. H3 medium is a variant of TeSR1. Stock A was prepared as described above for O3 medium. Human albumin solution (Albuminativ) was purchased from Octapharma AB. The solution was dialyzed three times against cell culture PBS for 3 h each time, using a 12- to 14-kDa dialysis membrane (Spectrum Laboratories), and then once against DMEM-F12 medium. We measured optical density to assess the final concentration of the protein in the solution. Stock B was mixed using the appropriate volume (depending on the concentration) of the dialyzed human albumin solution to achieve the same concentration of albumin as in stock B for O3 medium (described above). Trace elements and phenol red were added, and DMEM-F12 was used instead of water. Stock of TGF-β1 was prepared as described32, but the dialyzed human albumin was used instead of BSA. All other stocks were prepared as described above. H3 medium was mixed as described for O3, but NaCl was not added. Before use, the medium was supplemented with 96 ng ml−1 carrier-free recombinant human FGF basic (R&D Systems) and 40 μg ml−1 ascorbic acid (Sigma-Aldrich).

Embryoid body formation. ES cells were dissociated from LN-511–coated cell culture dishes as described above for passaging. They were broken into pieces and cultured in suspension 96-well plates (Sarstedt) in Knockout DMEM medium (GIBCO Invitrogen) supplemented with 2 mM l-glutamine, 20% (vol/vol) FBS (GIBCO Invitrogen), 0.1 mM β-mercaptoethanol (GIBCO Invitrogen) and 1% (wt/vol) nonessential amino acids (GIBCO Invitrogen). After 1–2 weeks in suspension, the embryoid bodies were transferred into gelatin-coated tissue cell culture 96-well plates (Sarstedt), cultured for 1–2 weeks, then fixed, stained with antibodies against markers of all three embryonic germline layers

29. Inzunza, J. et al. Derivation of human embryonic stem cell lines in serum replacement medium using postnatal human fibroblasts as feeder cells. Stem Cells 23, 544–549 (2005). 30. Dimos, J.T. et al. Induced pluripotent stem cells generated from patients with ALS can be differentiated into motor neurons. Science 321, 1218–1221 (2008). 31. Inzunza, J. et al. Comparative genomic hybridization and karyotyping of human embryonic stem cells reveals the occurrence of an isodicentric X chromosome after long-term cultivation. Mol. Hum. Reprod. 10, 461–466 (2004). 32. Ludwig, T.E. et al. Feeder-independent culture of human embryonic stem cells. Nat. Methods 3, 637–646 (2006).

doi:10.1038/nbt.1620

nature biotechnology

resource

© 2010 Nature America, Inc. All rights reserved.

Analysis of a genome-wide set of gene deletions in the fission yeast Schizosaccharomyces pombe Dong-Uk Kim1,14, Jacqueline Hayles2,14, Dongsup Kim3,14, Valerie Wood2,4,14, Han-Oh Park5,14, Misun Won1,14, Hyang-Sook Yoo1,14, Trevor Duhig2, Miyoung Nam1, Georgia Palmer2, Sangjo Han3, Linda Jeffery2, Seung-Tae Baek1, Hyemi Lee1, Young Sam Shim1, Minho Lee3, Lila Kim1, Kyung-Sun Heo1, Eun Joo Noh1, Ah-Reum Lee1, Young-Joo Jang1, Kyung-Sook Chung1, Shin-Jung Choi1, Jo-Young Park1, Youngwoo Park1, Hwan Mook Kim6, Song-Kyu Park6, Hae-Joon Park5, Eun-Jung Kang5, Hyong Bai Kim7, Hyun-Sam Kang8, Hee-Moon Park9, Kyunghoon Kim10, Kiwon Song11, Kyung Bin Song12, Paul Nurse2,13 & Kwang-Lae Hoe1,6 We report the construction and analysis of 4,836 heterozygous diploid deletion mutants covering 98.4% of the fission yeast genome providing a tool for studying eukaryotic biology. Comprehensive gene dispensability comparisons with budding yeast—the only other eukaryote for which a comprehensive knockout library exists—revealed that 83% of single-copy orthologs in the two yeasts had conserved dispensability. Gene dispensability differed for certain pathways between the two yeasts, including mitochondrial translation and cell cycle checkpoint control. We show that fission yeast has more essential genes than budding yeast and that essential genes are more likely than nonessential genes to be present in a single copy, to be broadly conserved and to contain introns. Growth fitness analyses determined sets of haploinsufficient and haploproficient genes for fission yeast, and comparisons with budding yeast identified specific ribosomal proteins and RNA polymerase subunits, which may act more generally to regulate eukaryotic cell growth. Systematic genome-wide gene deletion collections of eukaryotic organisms provide powerful tools for investigating molecular mechanisms in basic biology and for identifying pathways that can be targeted in bioengineering or medical applications, as shown by pioneering studies with the budding yeast Saccharomyces cerevisiae1–5. The construction of systematic gene deletion collections is difficult, although RNA interference (RNAi) provides a popular alternative approach to ablate gene activity in many organisms. However, RNAi approaches suffer from drawbacks such as partial knockdown of gene expression and off-target effects. For example, RNAi screens in fly and human cells revealed only a 10–38% overlap in genes identified as being required for the cell cycle between these two organisms6. We have constructed a genome-wide gene deletion set for the fission yeast Schizosaccharomyces pombe. Fission and budding yeast are not closely related and differ in a number of aspects including organization of the cell cycle, heterochromatin, complexity of centromeres and DNA replication origins and the prevalence of introns7, which makes their comparison valuable for defining genes and processes required more generally in eukaryotes. Here, we have identified similarities and differences in gene dispensability between

the two yeasts and have used growth fitness profiling to identify genes haploinsufficient or haploproficient for growth. RESULTS Deletion construction and gene dispensability We have constructed 4,836 heterozygous deletions covering 98.4% of the 4,914 protein coding open reading frames (ORFs) based on the annotated genome sequence7 (http://www.genedb.org/genedb/ pombe, 01/04/08) (Online Methods and Supplementary Table 1; for all the PCR primer sets and the mapping data, see Supplementary Data 1 and 2; also available at http://pombe.kaist.ac.kr/nbtsupp/). In addition, we have deleted 9 Tf2 transposons, 39 dubious genes8 and 48 pseudogenes (Supplementary Table 2). Each gene was deleted and replaced using homologous recombination by a ‘deletion cassette’ containing the KanMX marker gene9 (Supplementary Data 3) flanked by a pair of unique molecular bar codes (Fig. 1a, Supplementary Fig. 1 and Supplementary Table 1). Several pilot scale deletion studies have been carried out10–12 and it was suggested that 40~80 bp of homology is not always sufficient for the recombination required for the systematic deletion of genes in fission yeast12. Both block PCR

1Integrative

Omics Research Centre, Korea Research Institute of Bioscience and Biotechnology (KRIBB), Yuseong, Daejeon, Korea. 2Cancer Research UK, The London Research Institute, London, UK. 3Department of Bio and Brain Engineering, Korea Advanced Institute of Science & Technology (KAIST), Yuseong, Daejeon, Korea. 4Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge, UK. 5Bioneer Corp., Daedeok, Daejeon, Korea. 6Bioevaluation Centre, Korea Research Institute of Bioscience and Biotechnology (KRIBB), Ochang, Chungcheongbuk-do, Korea. 7Department of Bioinformatics & Biotechnology, Korea University, Jochiwon, Chungnam, Korea. 8School of Biological Sciences, Seoul National University, Seoul, Korea. 9Department of Microbiology, Chungnam National University, Yuseong, Daejeon, Korea. 10Division of Life Sciences, Kangwon National University, Chuncheon, Kangwon-do, Korea. 11Department of Biochemistry, Yonsei University, Seoul, Korea. 12Department of Food and Nutrition, Chungnam National University, Yuseong, Daejeon, Korea. 13The Rockefeller University, New York, New York, USA. 14These authors contributed equally to this work. Correspondence and requests for material should be addressed to K.-L.H. ([email protected]). Received 6 January; accepted 30 March; published online 16 May 2010; doi:10.1038/nbt.1628

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

617

resource

© 2010 Nature America, Inc. All rights reserved.

Figure 1 Deletion construction and gene dispensability. (a) Gene deletion cassette containing the KanMX4 gene flanked by unique bar codes (UPTAG/DNTAG) and regions of homology to the gene of interest (RHG). The cassette replaced the ORF of interest by homologous recombination at the RHG regions. (b) Construction of deletion mutants. All 4,836 protein coding genes were deleted using serial extension PCR (31.3%), block PCR (63.2%) or total gene synthesis (5.4%). The remaining 78 genes could not be confirmed as deleted owing to ambiguous sequencing results, recombination failure or inviability of the heterozygous diploids. (c) Dispensability of 4,836 protein coding genes. For 3,626 (2,729 + 897) genes the dispensability was previously unknown. ND, not done.

and total gene synthesis methods13 were developed to overcome this problem by increasing the length of homology from ~80 bp to ~350 bp (Fig. 1b and Supplementary Figs. 2–4). We confirmed that the deletion mutants were correctly replaced with the KanMX marker using PCR and dideoxy sequencing (Supplementary Fig. 5). For some genes constraints on primer selection for block PCR resulted in <100% of the ORF being deleted (for the amount of ORF deleted see Supplementary Table 1, column C KO%). Of the 4,836 genes deleted, at least 4,328 genes (87.6%) have >80% of their ORFs removed. In addition we carried out Southern blot analysis to determine the frequency with which the deletion cassette integrated elsewhere in the genome and estimated it to be <1% (Supplementary Fig. 6). We determined the essentiality of 4,836 genes by sporulating each heterozygous deletion diploid strain and then observing the germinating haploid spores microscopically. Essentiality was confirmed by tetrad analysis for all genes initially characterized as essential. We found that 26.1% of fission yeast genes (1,260/4,836) were essential and 73.9% (3,576/4,836) were nonessential for viability of haploid cells in the growth conditions we used. This analysis determined the dispensability for 3,626 genes that had not been deleted previously (Fig. 1c). Comparisons with published data for 1,210 genes revealed that the dispensability data for 98.4% of our deletions are similar or our data are more likely to be correct, leaving 1.6% as the maximum estimate of the error rate in our study (Supplementary Table 3). These results contrast with budding yeast where 17.8% (1,033/5,776) of genes are essential for viability (http://www.yeastgenome.org/). Fission yeast therefore has 227 more essential genes (1,260–1,033) than budding yeast despite having fewer genes in total (4,836 versus 5,776). Fission yeast has fewer duplicated genes than budding yeast14,15. It is therefore possible that there are more essential genes in fission yeast, because duplication in budding yeast is masking potential essentiality. We examined this possibility by identifying all of the essential genes for each organism with duplications in the other organism (ortho logous relationships Sp|Sc, one|many, many|one and many|many). We then identified the cases where these essential genes had orthologs in the other organism that were both nonessential and duplicated (Supplementary Table 1 and Online Methods). This revealed only 67 essential genes in fission yeast and 32 essential genes in budding yeast where essentiality of the orthologs in the other organism could be masked by redundancy. Thus redundancy could account for maximally only 35 (67–32) of the 227 extra essential genes in fission yeast. We conclude that redundancy is not the major reason for the additional essential genes in fission yeast. Analysis of gene dispensability Essential and nonessential genes are distributed evenly throughout the fission yeast genome except within 100 kb of the telomeres on chromosomes 1 and 2 (Fig. 2a). As in other organisms4 genes in the subtelomeric regions showed low essentiality (1.2%) compared to a 618

a

UPTAG

RHG

. DN

Chr

A

Gene b synthesis

RHG

Not made (78)

KanMX4

DNTAG

ORF

RHG

Chr.

RHG

DNA

c

(263)

Serial PCR (1,515)

Block PCR (3,058)

Nonessential (this study) 2,729

Nonessential (known) 847

Essential (this study) 897

Essential ND (known) 363 78

genome average of 26.1%. These regions are enriched for paralogs (68.7%, 79/115) and we have shown that duplicated genes are less likely to be essential than single-copy genes (Supplementary Table 4). These regions are also enriched for nonessential species-specific genes related to meiosis and the response to nitrogen starvation, which are less likely to be essential under these assay conditions15. In fission yeast ~46% of genes have introns7 and we found that the essentiality of genes with one or more introns is significantly higher than genes lacking introns (33% versus 21%, P < 10−14) (Fig. 2b). One possible explanation is that essential genes are less likely to be rapidly regulated given that it has already been shown that rapidly regulated stress response–related genes are less likely to contain introns16. Alternatively if introns arose early during eukaryotic evolution, this may be reflected as a bias toward intron-containing essential genes because essential genes are more likely to be ancient than nonessential genes12. The relationship between our gene essentiality data and previously published ORFeome localization17 was also analyzed for ten different cellular locations (Fig. 2c). As in budding yeast3,18 we found that the greatest percentage of essential gene products was localized to the nucleolus, nuclear envelope and the spindle pole body. As previously shown for budding yeast4, essential genes in fission yeast were more likely to be unique, with 93.1% of essential genes (1,173/1,260) being present in single copy compared to 73.9% of nonessential genes (2,643/3,576). In contrast nonessential genes were more likely to be duplicated or species-specific. Comparison of Gene Ontology (GO) term enrichment between the two yeasts revealed that the essential gene sets for both yeasts were significantly (P < 10−2) enriched for core cellular processes, such as macromolecular (DNA, RNA, protein and lipid) metabolism and cellular biosynthesis (transcription initiation and/or translation and ribosome assembly) (Fig. 2d and Supplementary Table 5). In contrast, nonessential genes were significantly (P < 10−2) enriched for regulatory functions (control of gene expression and cell communication) (Fig. 2d and Supplementary Table 6). Nonessential genes were also enriched for conditional or life-cycle specific processes, such as stress response, transmembrane transport and meiosis or sexual reproduction, together with processes that are less likely to be essential in the rich medium and mitotic growth used in our assay conditions. Genes of unknown function were also highly enriched (93%) in the nonessential genes. We predict that many of these genes are involved in biological regulation or condition-specific processes and are not directly involved in primary processes. VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

resource a

Chr 1

Cen1

0

1

Chr 2

2

3

(5.6 M)

4

Cen2

d Biological process

5

(4.5 M)

Cellular component org. Cellular macromol. biosyn.

1

Chr 3

Cen3

0

1

2

2

c

Cytoplasm

25 20 15 10 5 0 1 2 3 >4 Number of introns

RNA processing Ribosome biogenesis Protein localization Translation

Essential (Fission/budding) Nonessential

/

Mitotic cell cycle General transcription

Mitochondrion ER Cell periphery

Enriched nonessential

Cellular localization

Essential genes (%)

30

Nucleotide/nucleic acid met.

DNA replication

Nucleus

35

0

4

(2.5 M)

b

Nucleolus Golgi Spindle pole body Nuclear envelope Microtubule 0 5 10 15 20 25 30 35 40 45 Essential genes (%)

Unknown Meiotic cell cycle Response to stress Transmembrane transport Cell communication Reg. of gene expression 0

500

1,000

1,500

2,000

Number of genes

Figure 2 Analysis of gene dispensability. (a) Chromosome distribution of gene dispensability. Essential genes (tall bars) and nonessential genes (short bars) are distributed randomly throughout the genome except within 100 kb of the telomeres (gray boxes), where nonessential genes are enriched. Upper bars represent genes transcribed left to right and lower bars represent genes transcribed right to left. Filled circles in orange represent centromeres. (b) Percentage of essential genes versus number of introns. Percentage of essential genes was plotted against the number of introns within genes. In fission yeast, the percentage of essential genes containing introns is significantly (P < 10−14) higher than the percentage of those lacking introns. The dotted line represents the average percentage of essential genes in the total gene set (26.1%). (c) Percentage of essential genes versus ORFeome localization. The percentage of essential genes was plotted against ten different cellular locations in fission yeast. The dotted line represents the average percentage of essential genes for the total gene set (26.1%). The number of essential gene products localized to the nucleolus, spindle pole body and nuclear envelope is higher than average. The number of essential genes compared to the total for each location is: (i) cytoplasm 564/2,113; (ii) nucleus 601/2,068; (iii) mitochondrion 128/450; (iv) ER 98/436; (v) cell periphery 55/326; (vi) nucleolus 89/217; (vii) Golgi 27/224; (viii) spindle pole body 69/181; (ix) nuclear envelope 29/76; and (x) microtubule 20/71. (d) Comparison of GO analyses of fission yeast and budding yeast genes. Bar chart shows a selection of broad, biologically informative GO terms significantly (P ≤ 0.01) enriched for essential and nonessential genes in fission yeast and budding yeast. For the complete list of processes and for methods used to extract these data, see Supplementary Tables 5 and 6.

Species distribution of essential genes The dispensability profiles for the 4,836-deletion gene set were classified by their gene copy numbers according to their relationship with budding yeast genes (Supplementary Table 4 and x axis in Fig. 3) and into five categories by their species distribution (Supplementary Essential Fission yeast specific

Nonessential

Variable phyla Species distribution

© 2010 Nature America, Inc. All rights reserved.

3

Enriched essential

0

Fungi

Eukaryotes + bacteria

Eukaryotes

n=1 m≥1

n>1 m≥1

n≥1 m=0

Gene copy number

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

Table 4 and y axis in Fig. 3). In a comparison of the entire deletion gene set (4,836) there are 2,841 single-copy genes (n = 1, m ≥1) (n and m are gene copy number in fission yeast and budding yeast, respectively), 855 duplicated genes in fission yeast that are conserved in budd ing yeast (n > 1, m ≥1) and 1,140 genes found in fission yeast but not conserved in budding yeast (n ≥ 1, m = 0). The 1,260 essential genes were distributed across species as follows: (i) 883 genes conserved only in eukaryotes including humans, (ii) 207 conserved in both bacteria and eukaryotes, including humans, (iii) 91 genes found only in fungi, (iv) 39 genes found with a variable distribution throughout the phyla and (v) 40 fission yeast–specific genes. Essential genes were more likely than nonessential genes to be single copy and to be conserved broadly across species. Of the 1,260 essential fission yeast genes, 1,173 were single copy and only 87 have duplicates (Supplementary Tables 1 and 4). From the total of 974 (883 + 91) essential genes found only Figure 3 Comparative analysis of gene dispensability profiles of fission yeast. Gene dispensability profiles of 4,836 deletion mutants by gene copy number of fission yeast orthologs compared to budding yeast (x axis) and species distribution (y axis). Compared to budding yeast, fission yeast genes consist of 2,841 single-copy genes (n = 1, m ≥1), 855 duplicated genes (n > 1, m ≥1) and 1,140 genes found in fission yeast but not in budding yeast (n ≥ 1, m = 0), where ‘n’ is the number of genes in fission yeast and ‘m’ is the number of genes in budding yeast. The term ‘eukaryotes’ includes human and the term ‘variable phyla’ includes plants. The area of each circle represents the numbers of genes, where essential and nonessential genes are represented by yellow and blue, respectively.

619

© 2010 Nature America, Inc. All rights reserved.

resource in eukaryotes, 59 are probably related to genes found in Archaea (Supplementary Table 7). The remaining 915 genes (72.6% of all essential genes) are likely to have arisen within the eukaryotic lineage. This implies that many essential novel gene functions arose with the evolution of the eukaryotic cell. The fidelity of cell division in ancestral unicellular eukaryotes may have been very low, which could be tolerated in evolutionary terms as long as there was overall population growth. However, a multicellular eukaryote requires greater fidelity at each cell division than a unicellular eukaryote, because even moderate levels of random cell death would lead to poor survival of a multicellular organism. It has been estimated that it took around 500 million years for multicellular organisms to arise from an ancestral unicellular eukaryote19, and we propose that during this period there was considerable genomic innovation to generate a unicellular eukaryote with sufficient fidelity at cell division to allow the evolution of multicellularity. Essential genes broadly conserved both in bacteria and eukaryotes were significantly (207 genes, P < 10−2) enriched for respiratory function and primary metabolism of low molecular weight molecules, such as nucleotide or glucose metabolism (Supplementary Table 8). Of 445 fission yeast–specific genes, only 40 were essential for viability (Supplementary Table 9). Some of these genes are implicated in aspects of mitotic and meiotic chromosome segregation10 and such species-specific genes may have played a role in speciation by reinforcing reproductive isolation20. As the majority of essential genes are broadly conserved, it is possible that distant orthologs exist in other eukaryotes, including budding yeast, if some of these apparently species-specific genes are rapidly diverging. To investigate this possibility we re-interrogated the nonconserved essential genes from both yeasts using the same criteria used to build the manual ortholog data set, but relaxing thresholds for candidates to generate seed alignments and building alignments starting from the budding yeast genes rather than the fission yeast genes. This revealed a further four potential orthologs (Supplementary Table 10). This indicates that more

a

b

in-depth comparisons of the essential nonconserved gene sets may reveal further distant evolutionary relationships and functions. Dispensability comparison of orthologous pairs from the two yeasts Access to deletion collections for both fission yeast and budding yeast allows a robust comparative analysis of dispensability between two evolutionarily distant eukaryotic organisms. To eliminate any complications due to functionally redundant paralogous genes, 2,438 single-copy orthologous pairs (one to ones) for which deletion data are available in both organisms were used for this analysis (Supplementary Table 11). Overall 83% of these genes (2,027/2,438) had the same dispensability in both yeasts (Fig. 4a), suggesting that conserved orthologs in other organisms may also have conserved dispensability. GO enrichment of the conserved one-to-one essential genes in fission and budding yeasts was similar to that of all essential genes (compare Supplementary Table 12 with Supplementary Table 5), whereas the nonessential one-to-one pairs (compare Supplementary Table 13 with Supplementary Table 6) were enriched for additional GO terms, such as DNA damage, Golgi and/or endoplasmic reticulum (ER)-related processes and catabolic processes. As conserved genes can be expected to be under positive selection, these single-copy nonessential genes are likely to contribute to overall cell fitness. For example, the inability to repair nonlethal DNA damage will reduce cell fitness. It is also likely that some processes still take place in the absence of certain components, albeit less efficiently, because of flexibility and plasticity in the processes concerned21. The Golgi/ER-related processes may be complemented by different but related membrane trafficking pathways or components substituting one for the other. The remaining 17% of orthologous pairs (411/2,438) differ in essentiality between the two yeasts; of these, 268 are essential only in fission yeast and 143 are essential only in budding yeast (Fig. 4a). Therefore, there are 125 extra essential genes (268–143) in fission yeast

Fission yeast

Essential Nonessential

NE:NE 1,312 (54%)

E:E 715 (29%)

E:NE 268 (11%)

NE:E 143 (6%)

80 75

30

Budding yeast

Biological process Mitochondrial translation Other mitochondrial function Iron-sulphur cluster assembly Other processes Spindle/kinetochore associated DNA recombination/repair DNA replication checkpoint Mitotic/SIN signaling Glycosylation/ER associated V-type ATPase Proteosome/ubiquitin associated SUMOylation associated Neddylation associated Proteolysis (peptidases) Met/Thr/Glu metabolism Tryptophan metabolism Ergosterol metabolism Other amino acid metabolism Heme metabolism Purine/pyrimidine metabolism Tubulin-specific chaperone Actin cytoskeleton related

20

10

Number of genes with different dispensability

0

Essential Nonessential

0

10

20

30

75 80

Number of genes with different dispensability

Figure 4 Dispensability comparison of orthologous pairs from the two yeasts. (a) Essentiality of nonredundant 2,438 orthologous pairs were compared between the two yeasts. Eighty-three percent of orthologs show conserved dispensability and the remaining 17% show different dispensability. E, essential; NE, nonessential. (b) Functional distribution of orthologs with different dispensability. The 17% of the orthologous pairs with different dispensability were allocated to one of 31 biological terms, 22 of which are shown here. For the complete list of processes and genes, see Supplementary Table 14. Note that genes annotated to mitochondrial functions, certain amino acid metabolic pathways and protein degradation pathways such as neddylation and sumoylation are mostly essential in one yeast and nonessential in the other yeast, whereas other categories show essential genes (although the specific genes are different) in both yeasts under the conditions used in this study. Because there are some differences in the constituents of the standard rich media used for each organism, it is possible that in a few cases different dispensability between the two organisms are due to these differences.

620

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

resource

Growth profiling of diploids All fission deletion mutants constructed in this study have been barcoded (Supplementary Table 1), enabling the strains to be examined as an entire set in pooled experiments. Parallel analysis for changes in the growth rate of heterozygous deletion diploid strains has been used in budding yeast to identify potentially rate-limiting steps for cellular growth2,28,29. Using a similar methodology30 (Online Methods and Supplementary Figs. 7–9), we examined the growth rates in yeast extract medium for 4,334 fission yeast heterozygous deletion diploids (Supplementary Table 15; for the microarray raw data see Supplementary Data 4 and 5) and we further examined the growth rate of the 10 slowest haploinsufficient mutants as a proof-of-principle experiment (Supplementary Fig. 10). The growth rates of these ten mutants were found to be comparable to the relative fitness results from the microarray parallel analysis. Comparisons were also made for the haploinsufficient (slower growth) and haploproficient (faster growth) genes in fission yeast and budding yeast (Fig. 5). There were considerably more haplo insufficient genes in fission yeast compared to budding yeast (455 versus 356) when using a growth rate cut-off of <0.97 (Fig. 5 and Supplementary Table 16), whereas there were a similar number of haploproficient genes in both yeasts. The budding yeast life cycle is predominantly diploid and so reduced expression of potentially haploinsufficient genes in diploid cells is likely to have been subject to strong negative selection; this would not be the case for the predominantly haploid fission yeast. To make a more direct comparison between the fission and budding yeasts, we compared the fastest 3% of haploproficient genes (136 versus 183) and the slowest 3% of haploinsufficient genes (138 versus 184) from each organism (Table 1 and Supplementary Table 17). In fission yeast the haploproficient gene set showed GO enrichment for nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

1,500

Number of genes

100 50

1,000

0.88

0.91

0.94

0.97

Fission yeast

0

500

0 0.88

0.91

0.94

0.97

1.00

1.03

1,500 100 Number of genes

compared to budding yeast in this category, making the difference in dispensability of one-to-one orthologs a major reason why overall there are 227 more essential genes in fission yeast than in budding yeast. To analyze these differences further we identified a set of broad biological processes that encompass the entire set and sorted each gene pair into the most biologically relevant group (Fig. 4b and Supplementary Table 14). The most striking difference was for mitochondrial function (95 orthologous pairs). Of these 89 genes were essential in fission yeast and only six genes in budding yeast. Many of these genes encode components of the mitochondrial translation machinery (69 genes), which is required for mitochondrial DNA (mtDNA) stability. Loss of mtDNA is lethal in fission yeast but not in budding yeast where ‘petite’ mutants lacking mtDNA are viable and mitochondrial translation is not essential22. Conversely, the DNA replication checkpoint genes rad3, rad26 and cds1 are nonessential during normal growth, whereas their respective budding yeast orthologs are essential because of the requirement for degradation of a ribonucleotide reductase inhibitor23. This inhibitor can be degraded by a second checkpoint-independent pathway in fission yeast24,25 and other eukaryotes but not in budding yeast. Other examples of differential essentiality include the biological processes relating to RNA processing and/or export pathways, Golgi/ER transport, spindle/kinetochore/centromere, transcription and/or other chromatin-associated and glycosylation and/or other ER-associated processes. These differences may reflect dissimilarities in the numbers of introns7, centromere structure7, the organization of the Golgi network26,27 and membrane trafficking. Although 83% of the orthologous pairs have conserved dispensability, different essentiality of specific biological processes and defined complexes in 17% of gene pairs may represent life-style differences between these distantly related yeasts.

50

1,000

0.88

0.91

0.94

0.97

Budding yeast

0

500

0 0.88

0.91 0.94 0.97 1.00 Relative growth rate in rich media

1.03

Figure 5 A comparison of the relative growth rates for the total set of heterozygous deletion diploids in fission yeast (4,334 genes) and budding yeast (5,921 genes). In fission yeast there are more haploinsufficient genes with a relative growth rate of <0.97 compared to budding yeast (455 versus 356), as shown in the expanded region 0.88–0.97 (Supplementary Table 16).

macromolecule biosynthesis (P < 2.1 × 10−19) particularly ribosomal proteins (Table 1a and Supplementary Table 18). The TOR pathway genes (tor2, tsc1 and mip1) and genes encoding Rab-GTPase activating proteins were also found to be enriched in the haploproficient gene set. The loss of heterozygosity in TSC131, RAB-GTPases32 and also certain ribosomal proteins33 has been implicated in certain human cancers. None of these haploproficient genes showed any enrichment in the budding yeast haploproficient gene set. If fission yeast evolved in a nutrition-poor niche, then these pathways may have evolved to fine-tune optimal growth in these conditions, which may result in a sub-maximal growth rate in rich media. Haploinsufficient genes from budding yeast showed a significant GO enrichment for ribosomal-related function29, whereas those from fission yeast did not (Supplementary Table 18). We reasoned that any genes common to both haploinsufficient gene sets are likely to be important for regulating growth in both yeasts. A comparison of these gene sets in the two yeasts (138 versus 184) revealed 14 common orthologous groups and 15 genes (Table 1b). These included three genes encoding small subunit ribosomal proteins (S3, S6 and S7), five genes encoding large subunit ribosomal proteins (L6, redundant L13, L35 and L39) and another five genes involved in transcriptional functions including a predicted transcription factor TFIID complex subunit A/SAGA complex subunit (taf12), DNA-directed RNA polymerase II–specific subunits, (rpb3 and rpb7) and DNA-directed RNA polymerase subunits (rpb6 and rpc10), which are common to DNA-directed RNA polymerases I, II and III. Because the haplo insufficiency of these genes has been conserved between two distantly related organisms, it is likely that the amount of protein encoded by them is particularly important for the growth rate of the cell. It is therefore possible that their dosage is also important for the regulation of growth in other eukaryotes. 621

resource Table 1 Haploinsufficient and haploproficient genes in the two yeasts (a) GO term

Haploproficient (HP) gene

Translation & ribosome 60S biogenesis (GO:0006412)

rpl301, rpl501, rpl702, rpl801, rpl901, rpl902, rpl1001, rpl1101, rpl1701, rpl1801, rpl2001, rpl2002, rpl1901, rpl2101, rpl2102, rpl2301, rpl2502, rpl2802, rpl3001, rpl3201, rpl3202, rpl3401, rpl3601, rpl3602, rpl3702, rpl4301, rpl3801, rpp201 (28 genes) rps001, rps002, rps401, rps402, rps403, rps502, rps801, rps802, rps901, rps1001, rps1002, rps1101, rps1102, rps1201, rps13, rps1501, rps1502, rps1602, rps1701, rps1702, rps1801, rps1902, rps23, rps2302, rps2402, rps2802 (26 genes) gyp1, gyp7, gyp51, SPAC1952.17c

40S

© 2010 Nature America, Inc. All rights reserved.

Regulation of Rab GTPase activity (GO:0032313) TOR signaling pathway (GO:0031929)

tor2, tsc1, mip1, tco89, gad8

HP gene annotation

Total gene annotation

P-value (uncorrected)

54

316

2.10 × 10−19

4

13

5.30 × 10−4

5

14

8.55 × 10−3

Of the genes deleted in the 136 haploproficient mutants with the fastest growth rates, 54 genes (39.7%) encode ribosomal subunit proteins and nine genes encode Rab GAP and TOR pathway-related proteins. For GO enrichment of the haploproficient genes in fission yeast, see Supplementary Tables 17 and 18.

(b) Gene product

Gene category

Budding yeast ID

Fission yeast ID

Sc:Sp

Ribosomal protein S3 Ribosomal protein S6 Ribosomal protein S7 Ribosomal protein L6 Ribosomal protein L13 Ribosomal protein L35 Ribosomal protein L39 TFIID subunit A (Taf12) RNA pol II Rpb3 RNA pol Rpb6 RNA pol II Rpb7 RNA pol Rpc10 U3 snoRNP Utp4 ATPase Rvb2

Ribosomal subunit Ribosomal subunit Ribosomal subunit Ribosomal subunit Ribosomal subunit Ribosomal subunit Ribosomal subunit Transcription Transcription Transcription Transcription Transcription RNA processing Chromatin remodeling

YNL178W YPL090C YNL096C|YOR096W YLR448W|YML073C YDL082W|YMR142C YDL136W|YDL191W YJL189W YDR145W YIL021W YPR187W YDR404C YHR143W-A YDR324C YPL235W

SPBC16G5.14c SPAPB1E7.12 SPAC18G6.14c SPCC622.18 SPAC664.05|SPBC839.13c SPCC613.05c SPCC663.04 SPAC15A10.02 SPCC1442.10c SPCC1020.04c SPACUNK4.06c SPBC19C2.03 SPBC19F5.02c SPBC83.08

1:1 1:1 2:1 2:1 2:2 2:1 1:1 1:1 1:1 1:1 1:1 1:1 1:1 1:1

Genes common to the haploinsufficient gene sets of both fission yeast and budding yeast (Supplementary Tables 17 and 18). Of these 15 genes, 13 (86.7%) are involved in transcription or translation.

DISCUSSION Fission yeast is an important model eukaryotic organism and the availability of a genome-wide deletion collection will facilitate further studies such as genetic interaction assays, phenotypic analysis4, comparative genomics, gene dispensability analysis of higher eukaryotes and drug-induced haploinsufficiency screening34. For example, a partial collection of the viable haploid deletions has been distributed to ~25 laboratories and studies from two laboratories have shown that there is considerable conservation of synthetic lethal genetic interactions with budding yeast as well as rewiring of some functionally conserved modules35,36. Our comparisons of orthologous gene pairs between budding and fission yeast showed that 83% had the same dispensability despite being distantly related. This high level of conservation in dispensability will be helpful for the interpretation of more complex RNAi data from other organisms 6,37–39. We have also shown that there is a relationship between gene essentiality and the presence of introns, which may indicate that essential genes are less likely to be rapidly regulated 16. There are orthologs for 3,492 fission yeast genes in other eukaryotes, including humans. Of these genes, 454 are not conserved in budding yeast suggesting that fission yeast may be a valuable alternative organism to budding yeast for certain experiments, for example, optimization of drug screening protocols. However, there are ~3,038 genes conserved in 622

both yeasts and other eukaryotes including humans, which encourages us in the view that conclusions drawn from analyses in the two yeasts concerning molecular and cell biology will be relevant to, and improve our understanding of, metazoan cells. We have also identified a small set of genes required for translation and transcription, including genes encoding specific ribosomal proteins and RNA polymerase subunits that are haploinsufficient for growth in both the yeasts. These specific gene products may play a critical role in regulating the growth of eukaryotic cells. The identification of genes encoding elements of the TOR pathway, Rab-GTPase activating proteins and ribosomal proteins, as haploproficient, is also of interest given the involvement of these gene products in cancer31–33. The availability of a near-complete, genome-wide deletion collection for fission yeast provides a useful tool for the functional studies of eukaryotic molecular and cell biology and for biotechnological applications. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Note: Supplementary information is available on the Nature Biotechnology website. Acknowledgments We thank members of our laboratories for their participation in the construction and analysis of the deletion mutants, particularly H.-R. Hwang, H.-S. Ahn,

VOLUME 28 NUMBER 6 JUNE 2010 nature biotechnology

resource Y.-D. Kim, S. Park, H.-J. Lee, J.-H. Ahn, Y.-S. Kil, S.-Y. Park, J.-H. Lim, J.-H. Song, Y.-K. Ryoo, J.-Y. Kim, M.-J. Oh, S. Kong, J. Ahn, N. Sun, N. Peat, R. Mandeville and J.-J. Li. We also thank J.-H. Roe and W.-K. Huh for reading this manuscript and for their insightful comments and O. Nielsen for his patience with the many requests for pON177. This work was supported by the intramural research program of KRIBB (Mission 2007), the Chemical Genomics Research Program and the 21st Century Frontier Research Program from the Ministry of Education, Science and Technology (MOEST) of Korea. This work was also supported by Bioneer Corp., The Wellcome Trust, Cancer Research UK, The Breast Cancer Research Foundation (BCRF) and The Rockefeller University. AUTHOR CONTRIBUTIONS D.-U.K., J.H., H.-O.P., M.W., H.-S.Y., P.N. and K.-L.H. conceived the project; D.-U.K., J.H., D.K., V.W., M.W., T.D., M.N., G.P., S.H., L.J., S.-T.B., H.L., Y.S.S., M.L., L.K., K.-S.H., E.J.N., A.-R.L., Y.-J.J., K.-S.C., S.-J.C., J.-Y.P., Y.P., H.M.K., S.-K.P., H.B.K., H.-S.K., H.-M.P., K.K., K.S. and K.B.S. performed experiments and data analysis; D.K., H.-J.P., E.-J.K. and H.-M.P. performed primer design; D.K. and V.W. performed bioinformatics; D.-U.K., J.H., D.K., V.W., P.N. and K.-L.H. wrote the paper.

© 2010 Nature America, Inc. All rights reserved.

COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/. 1. Jorgensen, P. et al. High-resolution genetic mapping with ordered arrays of Saccharomyces cerevisiae deletion mutants. Genetics 162, 1091–1099 (2002). 2. Hillenmeyer, M.E. et al. The chemical genomic portrait of yeast: uncovering a phenotype for all genes. Science 320, 362–365 (2008). 3. Giaever, G. et al. Functional profiling of the Saccharomyces cerevisiae genome. Nature 418, 387–391 (2002). 4. Winzeler, E.A. et al. Functional characterization of the S. cerevisiae genome by gene deletion and parallel analysis. Science 285, 901–906 (1999). 5. Entian, K.D. & Kotter, P. Methods in Microbiology 36, edn. II. 629–666 (Elsevier, 2007). 6. Kittler, R. et al. Genome-scale RNAi profiling of cell division in human tissue culture cells. Nat. Cell Biol. 9, 1401–1412 (2007). 7. Wood, V. et al. The genome sequence of Schizosaccharomyces pombe. Nature 415, 871–880 (2002). 8. Fisk, D.G. et al. Saccharomyces cerevisiae S288C genome annotation: a working hypothesis. Yeast 23, 857–865 (2006). 9. Wach, A., Brachat, A., Pohlmann, R. & Philippsen, P. New heterologous modules for classical or PCR-based gene disruptions in Saccharomyces cerevisiae. Yeast 10, 1793–1808 (1994). 10. Gregan, J. et al. Novel genes required for meiotic chromosome segregation are identified by a high-throughput knockout screen in fission yeast. Curr. Biol. 15, 1663–1669 (2005). 11. Martin-Castellanos, C. et al. A large-scale screen in S. pombe identifies seven novel genes required for critical meiotic events. Curr. Biol. 15, 2056–2062 (2005). 12. Decottignies, A., Sanchez-Perez, I. & Nurse, P. Schizosaccharomyces pombe essential genes: a pilot study. Genome Res. 13, 399–406 (2003). 13. Smith, H.O., Hutchison, C.A. III, Pfannkoch, C. & Venter, J.C. Generating a synthetic genome by whole genome assembly: phiX174 bacteriophage from synthetic oligonucleotides. Proc. Natl. Acad. Sci. USA 100, 15440–15445 (2003).

nature biotechnology VOLUME 28 NUMBER 6 JUNE 2010

14. Sipiczki, M. Where does fission yeast sit on the tree of life? Genome Biol. 1, reviews 1011.1–1011.4 (2000). 15. Wood, V. Schizosaccharomyces pombe comparative genomics; from sequence to systems, in Comparative Genomics: Using Fungi as Models (eds. Sunnerhagen, P. & Piskur, J.), 233–285 (Springer Berlin, Heidelberg, 2006). 16. Jeffares, D.C., Penkett, C.J. & Bahler, J. Rapidly regulated genes are intron poor. Trends Genet. 24, 375–378 (2008). 17. Matsuyama, A. et al. ORFeome cloning and global analysis of protein localization in the fission yeast Schizosaccharomyces pombe. Nat. Biotechnol. 24, 841–847 (2006). 18. Huh, W.K. et al. Global analysis of protein localization in budding yeast. Nature 425, 686–691 (2003). 19. Benton, M.J. & Ayala, F.J. Dating the tree of life. Science 300, 1698–1700 (2003). 20. Hoskin, C.J., Higgie, M., McDonald, K.R. & Moritz, C. Reinforcement drives rapid allopatric speciation. Nature 437, 1353–1356 (2005). 21. Harrison, R., Papp, B., Pal, C., Oliver, S.G. & Delneri, D. Plasticity of genetic interactions in metabolic networks of yeast. Proc. Natl. Acad. Sci. USA 104, 2307–2312 (2007). 22. Chiron, S., Suleau, A. & Bonnefoy, N. Mitochondrial translation: elongation factor tu is essential in fission yeast and depends on an exchange factor conserved in humans but not in budding yeast. Genetics 169, 1891–1901 (2005). 23. Choi, D.H., Oh, Y.M., Kwon, S.H. & Bae, S.H. The mutation of a novel Saccharomyces cerevisiae SRL4 gene rescues the lethality of rad53 and lcd1 mutations by modulating dNTP levels. J. Microbiol. 46, 75–80 (2008). 24. Ralph, E., Boye, E. & Kearsey, S.E. DNA damage induces Cdt1 proteolysis in fission yeast through a pathway dependent on Cdt2 and Ddb1. EMBO Rep. 7, 1134–1139 (2006). 25. Liu, C. et al. Cop9/signalosome subunits and Pcu4 regulate ribonucleotide reductase by both checkpoint-dependent and -independent mechanisms. Genes Dev. 17, 1130–1140 (2003). 26. Preuss, D., Mulholland, J., Franzusoff, A., Segev, N. & Botstein, D. Characterization of the Saccharomyces Golgi complex through the cell cycle by immunoelectron microscopy. Mol. Biol. Cell 3, 789–803 (1992). 27. Ayscough, K., Hajibagheri, N.M., Watson, R. & Warren, G. Stacking of Golgi cisternae in Schizosaccharomyces pombe requires intact microtubules. J. Cell Sci. 106, 1227–1237 (1993). 28. Roemer, T. et al. Large-scale essential gene identification in Candida albicans and applications to antifungal drug discovery. Mol. Microbiol. 50, 167–181 (2003). 29. Deutschbauer, A.M. et al. Mechanisms of haploinsufficiency revealed by genomewide profiling in yeast. Genetics 169, 1915–1925 (2005). 30. Pierce, S.E. et al. A unique and universal molecular barcode array. Nat. Methods 3, 601–603 (2006). 31. Jozwiak, J., Jozwiak, S. & Wlodarski, P. Possible mechanisms of disease development in tuberous sclerosis. Lancet Oncol. 9, 73–79 (2008). 32. Cheng, K.W., Lahad, J.P., Gray, J.W. & Mills, G.B. Emerging role of RAB GTPases in cancer and human disease. Cancer Res. 65, 2516–2519 (2005). 33. McGowan, K.A. et al. Ribosomal mutations cause p53-mediated dark skin and pleiotropic effects. Nat. Genet. 40, 963–970 (2008). 34. Lum, P.Y. et al. Discovering modes of action for therapeutic compounds using a genome-wide screen of yeast heterozygotes. Cell 116, 121–137 (2004). 35. Roguev, A. et al. Conservation and rewiring of functional modules revealed by an epistasis map in fission yeast. Science 322, 405–410 (2008). 36. Dixon, S.J. et al. Significant conservation of synthetic lethal genetic interaction networks between distantly related eukaryotes. Proc. Natl. Acad. Sci. USA 105, 16653–16658 (2008). 37. Kamath, R.S. et al. Systematic functional analysis of the Caenorhabditis elegans genome using RNAi. Nature 421, 231–237 (2003). 38. Dietzl, G. et al. A genome-wide transgenic RNAi library for conditional gene inactivation in Drosophila. Nature 448, 151–156 (2007). 39. Ravi, D. et al. A network of conserved damage survival pathways revealed by a genomic RNAi screen. PLoS Genet. 5, e1000527 (2009).

623

© 2010 Nature America, Inc. All rights reserved.

ONLINE METHODS

Construction of genome-wide deletion mutants. Heterozygous deletion mutants of 4,836 protein coding genes in fission yeast were constructed using a method based on homologous recombination of a deletion cassette containing a pair of unique molecular bar codes (up-tag and down-tag in Supplementary Table 1) and the KanMX marker gene9. The sequences of bar codes was generated using a BioPerl-based computer program to meet the following criteria; melting temperature (Tm) = 60 °C, no cross-hybridization, no secondary structures and no similarities to genomic sequences. RNAfold and mfold freeware (http://rna.tbi.univie.ac.at/cgi-bin/RNAfold.cgi) was used for checking secondary structure, and the BLAST program was used for checking similarity with genomic sequence. Deletion cassettes were generated by a modified PCR-based strategy. For one-third of deletion cassettes, the conventional serial-extension PCR method3,4 was used. For the remaining two-thirds, the block PCR method or an innovative gene synthesis method13 was employed, resulting in the increase in the length of homologous recombination regions from ~80 bp to 250~450 bp. Oligonucleotides used in construction of the deletion cassettes were supplied by Bioneer Corporation. The deletion cassettes were transformed into SP286 (ade6-M210/ade6-M216, leu1-32/leu1-32, ura4-D18/ura4-D18 h+/h+) using a lithium acetate method40, and then incubated for 5 d to select positive colonies on YES agar containing 100 μg/ml G418 (Duchefa Biochemie). Confirmation of genome-wide deletion mutants. To verify the integration of deletion cassettes at the correct locus, colony PCR was carried out. Dideoxy sequencing of the PCR product from each successful deletion mutant was carried out to confirm the sequences of up- and down-tags as well as the junctions to accurately define the deleted region. To estimate how often the deletion cassette integrated at additional sites in the genome, Southern blot analysis of chromosomal DNA from 61 different deletion strains was carried out using KanMX4 as a probe. All the strains and check-PCR primers described here are available from Bioneer (http://pombe.bioneer.co.kr). Determination of essentiality. General growth conditions and media were used as described41. Essentiality was determined by a microscopic observation of colony-forming ability of spores on YES (yeast extract medium supplemented with adenine, leucine, uracil and histidine at 250 mg/l) at 25 °C and 32 °C. The spores were derived from corresponding heterozygous diploid deletion strains transformed with the pON177 plasmid 42 using a modified version of the PLATE method 43. About 5% of the heterozygous deletion diploids could not be transformed using this high-throughput method and these were repeated using a standard transformation protocol40. Briefly, four batches each of 48 heterozygous diploid strains were patched on to YE (yeast extract medium supplemented with leucine and uracil at 250 mg/l) + G418 agar plates in two 96-well microtiter plates (each strain is represented four times) and left to grow for 2~3 d at 32 °C. Cells were inoculated into 200 μl YE + G418 and left to grow into stationary phase. The cells were harvested and transformed with pON177 (ref. 42), plated on minimal agar + leucine (250 mg/l) and incubated for a week at 32 °C. Transformants were inoculated into minimal media lacking nitrogen and left for 2~3 d at 25 °C to induce sporulation. The asci were treated with helicase (Bio Sepra) diluted 1 in 250 to eliminate vegetative cells, washed with water and the haploid spores were plated on YES agar at 25 °C and 32 °C. Essentiality was determined by a microscopic observation of the germinating spores on plates after 1 and 2 d before replica plating to YES +100 μg/ml G418 to confirm that the deletion phenotype was associated with G418 resistance. Essential genes were further analyzed by tetrad analysis. Briefly, cells harboring pON177 were left to germinate for 4~5 d on minimal plates. Using a Singer MSM microscope, spores were dissected on YES plates for 4~5 d at 30 °C. Viable colonies were patched onto YES plates + 100 μg/ml G418 to confirm that viability was linked to G418 sensitivity (Supplementary Methods). While analyzing gene dispensability, we found that a subset of the deletion collection harbored a recessive temperature-sensitive mutation unrelated to the gene deletion. This ts mutation was removed from the entire nonessential haploid deletion library after sporulation of the diploid heterozygous deletion strains of nonessential genes. There were originally 416 of the 1,260 essential heterozygous deletion diploid strains that harboured the ts mutation. Of these 416 strains, 364 have been remade and the remaining 52 are

nature biotechnology

currently being remade (Supplementary Table 1, column U for the list of heterozygous diploid strains that still contain the ts mutation). Redundancy and essentiality. To assess the effect of redundancy on masking essentiality and its contribution to the extra essential genes in fission yeast, we identified all genes in the one|many, many|one and many|many categories where data were available for both organisms (Supplementary Table 1). We eliminated all orthologous groups with an equal number of essential genes in each organism (e.g., ev|ev) and those where redundancy could not contribute to the difference in essentiality (e.g., vv|vv). The remaining essential genes where redundancy could mask essentiality in one or the other organism were counted for both yeasts. There were 67 essential genes in fission yeast and 35 essential genes in budding yeast where redundancy in the other yeast could potentially be masking essentiality. Data source and URLs. DNA and protein information of fission yeast were from the S. pombe GeneDB database ftp://ftp.sanger.ac.uk/pub/yeast/pombe/ Mappings/OLD/allNames.txt_27Aug2008, and the budding yeast data set from http://www.yeastgenome.org/. Budding yeast deletion data3 were from http://downloads.yeastgenome.org/literature_curation/archive/phenotypes. tab.20080202.gz. Interspecies comparisons used manually curated species distribution from GeneDB on 24/06/2008 and Version 13 of the manually curated fission yeast/budding yeast ortholog table. Distant ortholog detection. The detection of distant orthologs used all essential S. pombe and all S. cerevisiae proteins that were not already members of an existing orthologous group, based on the manually curated S. cerevisae/ S. pombe ortholog table version 13 (ref. 15). Ortholog candidate detection used PSI blast, and the criteria as described15 were used to support orthologous cluster predictions. Individual multiple alignments are provided in Supplementary Table 10. One ortholog prediction SPAC1006.42/YPR085C pair has since been confirmed experimentally (PMID 19040720). GO analysis. GO enrichment analysis used the Princeton implementation of GO term finder44 (http://go.princeton.edu/cgi-bin/GOTermFinder) with gene association files from November 2008; GO TermFinder calculates P-value using the hypergeometric distribution, and Bonferroni method is used for multiple hypothesis correction. Analysis used a P-value cut off of 0.01 and all evidence codes except RCA (reviewed computational analysis) are included. The whole genome comparison in Figure 2d and Supplementary Tables 5 and 6 used the total protein coding data sets for fission yeast (4,836) and budding yeast (5,776). Some biologically uninformative terms were omitted from the results (that is, when parent and child terms show identical enrichment only the child term is included). GO process enrichment of essential genes, which are conserved in single copy (Supplementary Table 12), versus nonessential genes conserved in single copy (Supplementary Table 13) used fission yeast annotations and background set. Parallel analysis using microarray. The custom-made GeneChip (48 K) was designed and manufactured according to the Affymetrix GeneChip guide (KRIBBSP2, Part No. 520506). Construction of mutant library pools, sampling, PCR amplification of probes, hybridization and washes were carried out following modified budding yeast protocols29,30. Genomic DNA was prepared from frozen cell stocks using a kit (Zymo Research ZR-Fungal/Bacterial DNA kit). For each sample, 10~20 OD600 corresponding to 2~4 × 108 cells/ml was used for the genomic DNA preparation. To amplify and label the tags the following sets of primers were used for PCR using 0.2 μg genomic DNA as a template; uptag, forward (5′U-2) 5′-GCTCCCGCCTTACTTCGCAT-3′, reverse (biotin-Kan5′U-2) 5′-biotin-CGGGGACGAGGCAAGCTAA-3′; downtag, forward (DN3-F-biotin) 5′-biotin-GCCGCCATCCAGTGTCG-3′, reverse (DN3-R) 5′-TTGCGTTGCGTAGGGGGG-3′. For growth profiling, data were collected from six independent experiments using two different pool sets. For details, see Supplementary Figures 7–9. Analysis of microarray results. Out of 4,441 mutants in the deletion pool, 3,523 mutants were represented by both up-tag and down-tag, and 811 mutants were represented by at least one of two tags. Therefore, at least one of the tags

doi:10.1038/nbt.1628

40. Bahler, J. et al. Heterologous modules for efficient and versatile PCR-based gene targeting in Schizosaccharomyces pombe. Yeast 14, 943–951 (1998). 41. Moreno, S., Klar, A. & Nurse, P. Molecular genetic analysis of fission yeast Schizosaccharomyces pombe. Methods Enzymol. 194, 795–823 (1991). 42. Styrkarsdottir, U., Egel, R. & Nielsen, O. The smt-0 mutation which abolishes mating-type switching in fission yeast is a deletion. Curr. Genet. 23, 184–186 (1993). 43. Elble, R. A simple and efficient procedure for transformation of yeasts. Biotechniques 13, 18–20 (1992). 44. Boyle, E.I. et al. GO:TermFinder–open source software for accessing Gene Ontology information and finding significantly enriched Gene Ontology terms associated with a list of genes. Bioinformatics 20, 3710–3715 (2004).

© 2010 Nature America, Inc. All rights reserved.

from 4,334 strains was detectable by chip analysis. The remaining 107 tags were removed from the analysis, as they had intensities less than fourfold that of background. For the analysis of microarray results, the analysis of covariance (ANCOVA) model was used as a statistical tool. Each array signal was normalized by a mean-intensity (that is, 2,500 arbitrary units) and interpreted by ANCOVA as a linear regression corresponding to a multiple-regression model on time (measured in generations and treated as a quantitative predictor) and replicate series (treated as a categorical predictor) simultaneously. This analysis provides estimates of statistical significance (P-values) using the F-statistic (Supplementary Methods).

doi:10.1038/nbt.1628

nature biotechnology

e r r ata

Erratum: The cancer vaccine roller coaster Bruce Goldman & Laura DeFrancesco Nat. Biotechnol. 27, 129–139 (2009); published online 7 February 2009; corrected after print 7 June 2010 In the version of this article initially published, the Mologen product description in Table 5, page 139, was incomplete and its status incorrectly stated to be compassionate use in India. The product description should have read: Genetically modified allogeneic (human) tumor cells for the expression of IL-7, GM-CSF, CD80 and CD154, in fixed combination with a DNA-based double stem loop immunomodulator (dSLIM). The status should have read: Orphan drug status granted by EMEA in 2006. The error has been corrected in the HTML and PDF versions of the article.

Erratum: Irish bioethics council axed Cormac Sheridan Nat. Biotechnol. 28, 112 (2010); published online 5 February 2010; corrected after print 7 June 2010

© 2010 Nature America, Inc. All rights reserved.

In the version of this article initially published, a researcher at University College Cork was incorrectly named. His name is Tom (not Barry) Moore. The error has been corrected in the HTML and PDF versions of the article.

Erratum: Never again Chris Scott Nat. Biotechnol. 28, 131 (2010); published online 5 February 2010; corrected after print 7 June 2010 In the version of this article initially published, Art Levinson is incorrectly described as a founder of Genentech, Sandra Horning as senior vice president of global clinical development and Richard Scheller as chief of operations. Their titles should have read: CEO Arthur Levinson moved up to the board of directors…. Sandra Horning…took over as senior vice president, global head, clinical development, hematology/oncology. Executive vice president, research, Richard Scheller…. The errors have been corrected in the HTML and PDF versions of the article.

Erratum: Resuscitated deCODE refocuses on diagnostics Mark Ratner Nat. Biotechnol. 28, 192 (2010); published online 8 March 2010; corrected after print 7 June 2010 In the version of this article initially published, it was reported that deCODE had “shuttered its Emerald Biosciences and Emerald Biostructures drug discovery operations”; in fact, the companies were sold to investors. In addition, the correct name of Emerald Biosciences is Emerald BioSystems. The error has been corrected in the HTML and PDF versions of the article.

Erratum: Biotech in a blink Ken Garber Nat. Biotechnol. 28, 311–314 (2010); published online 8 April 2010; corrected after print 15 April 2010 In the version of the article originally published, Michael Tolentino was misquoted to the effect that bevasiranib had been shown to persist indefinitely in post-mitotic cells. Tolentino actually stated that the RNA-induced signaling complex persists. The error has been corrected in the HMTL and PDF versions of the article.

624

volume 28 number 6 JUNE 2010 nature biotechnology

careers and recruitment

The ABC’s of industry: a postdoc program provides a sneak peek into industry careers Adnan O Abu-Yousif, Erik C Hett, Ann M Skoczenski & Tayyaba Hasan

© 2010 Nature America, Inc. All rights reserved.

An innovative partnership allows local companies to educate postdocs about careers in industry.

A

cademia or industry? It’s a question that scientists grapple with early and often throughout their training as graduate students and postdoctoral researchers. Most scientists receive training at academic institutions from mentors who offer great insight about how to excel in academia; however, any concerted effort towards providing postdocs with a broad understanding of careers beyond academia is rare. In fact, postdocs in hospital-based environments are exposed to vastly different cultural experiences when compared to counterparts in basic science departments. The career options for these hospital-based basic scientists can be very different from those of the clinical trainees that are the priority of hospitals, and historically, the infrastructure for research career development and institutional support to explore science career options have been lacking. In 2005, to address this and other issues of basic scientists, Massachusetts General Hospital (MGH) created the Office for Research Career Development (ORCD; http://www.massgeneral. org/orcd) to (i) clarify promotion requirements and career development pathways for research faculty in an academic medical environment; (ii) strengthen career guidance and mentoring offered to faculty and postdoctoral fellows; and (iii) enhance communication within the research community. Soon after the creation of the ORCD, MGH founded the Mass General Postdoc Association (MGPA; https://www. massgeneral.org/mgpa) to serve as a postdocled extension of the ORCD and provide research advancement, career development, communication and networking, quality-of-life training and advocacy on behalf of MGH’s 1,000+ postdocs. Adnan O. Abu-Yousif, Erik C. Hett, Ann M. Skoczenski and Tayyaba Hasan are at Massachusetts General Hospital, Boston, Massachusetts, USA e-mail: [email protected]

Initially the MGPA hosted programs for postdocs covering topics from grant writing to mentor development programs. Seeking to leverage the full resources of the greater BostonCambridge area, one of us, MGPA co-chair A.O.A., conceived the Industry Exploration Program to establish a connection between MGH postdocs and the many R&D companies in the area. Local organizations such as the Massachusetts Biotechnology Education Foundation (MassBioEd; http://www.massbioed. org/) and the Massachusetts Biotechnology Council (MassBio) created many important programs for high school and college stu-

dents, but had not yet delved into providing PhD level trainees with exposure to the culture and opportunities that exist in industry. When approached, Lance Hartford, executive director of MassBioEd, saw a great opportunity to work together to achieve the common goal of educating young PhD level scientists. Creating the program The objective was simple: create a program that provides opportunities for postdocs to learn about industry careers by visiting local companies and interacting with industry scientists. As part of the experience, postdocs

Box 1 Considerations when starting an Industry Exploration Program •S eek internal support and guidance from your institution’s research career development leadership. • Involve the Research Ventures and Licensing (RVL) or technology-transfer division of your institution early in the planning process. To protect potential intellectual property of researchers at your institutions and the participating companies, participating postdocs should be required to attend a seminar hosted by your RVL office that educates postdocs on interactions between academia and industry. • Prepare a one-page summary of your program that can be shared with companies that clearly outlines the goals of the program and potential activities. • Use a simple application process, such as the two-part application described below: Part 1: postdocs prepare a 250-word summary statement that outlines the background of the participant including training, skills and interests. These summary statements should not include specific experimental data in order not to jeopardize intellectual property. Part 2: postdocs obtain a signed consent form from their advisor to participate. • If possible, connect with a business association like MassBio to help make valuable connections to industry. • Limit the number of visits that an individual postdoc can make so that the greatest number of postdocs can participate. • Require participants to complete a survey upon the completion of the visit and suggest ways to improve the program. • Select participating postdocs and companies to participate in a panel discussion about the program to help improve the program and to garner further interest in the program from attendees of the panel seminar. • Develop a secure website that can host applications so that companies can rapidly identify candidates with mutual interests.

nature biotechnology volume 28 number 6 JUNE 2010

625

careers a n d recruitme n t Box 2 Potential activities for the day of postdoc visit

© 2010 Nature America, Inc. All rights reserved.

• Introductory session describing the company structure, size, and mission statement. • Tour of the relevant areas of the facility. • Panel discussion with company scientists describing their career path and what it’s like to work in industry. • One-on-one meetings with several company scientists (at different levels) to discuss science, their transition from academia into industry and career paths. • Laboratory meeting in the department that best suits the participating postdoc, allowing the postdoc to interact with the group in an intellectual setting. • Human Resources session that informs postdocs about the application and interview process at the company.

have the opportunity to network with industry experts, engage in scientific discussions, and learn how to prepare for the transition from academia to industry. The program demonstrates the value of collaborating with business associations like MassBioEd, which provided support from inception to execution. As a result of the partnership, countless PhD scientists will have the opportunity to gain invaluable insight into industry careers. We hope that the information provided here will serve as a platform to (i) guide other postdoc associations on how to launch similar programs in their region, and (ii) help companies realize the value of interacting with postdocs as potential colleagues. In the fall of 2009, we began with the support and guidance of the ORCD at MGH to determine an appropriate course of action. Once consensus was achieved on the direction and approach, we worked with Lance Hartford to explore the collaboration and discuss best practices for enlisting local companies. Among the opportunities to connect with industry organizations was the 2009 Annual MassBio meeting in Boston, which provided an opportunity to introduce the program to a handful of influential representatives from surrounding companies. Among the first companies to sign on for the Industry Exploration Program were New England Biolabs, EMD-Serono and AstraZeneca. Interested parties received a one-page summary outlining the goals of the program after the annual meeting and one-on-one meetings were held to determine criteria for the application process (Box 1). In tandem with these exchanges, we consulted the Research Ventures and Licensing (RVL) division at MGH to ensure that the intellectual property of postdocs, advisors and companies were protected. Ultimately, we concluded that postdoc participants would be required to attend a RVL Division seminar outlining how to protect intellectual property.

626

We solicited postdoc applications in the fall, which were completed and submitted to AstraZeneca, New England Biolabs and EMD-Serono. Companies evaluated the applications and selected program participants based on interest and expertise. Once matched, we worked with the postdocs and companies to coordinate a single visit for each participating organization. Site visits were planned at the sole discretion of the company and varied by location. Participants benefitted from a myriad of opportunities with industry scientists and associates during one-on-one meetings, site tours and panel discussions (Box 2). Surveys distributed immediately after the site visits revealed that postdocs enjoyed these personal interactions with industry scientists and learning about the structure of the organization. In fact, 70% of participants indicated an increased level of interest in the company after their visit. One participant, Ulysses Sallum, secured a collaboration with Barton Slatko, scientist at New England Biolabs, as a direct result of his participation in the Industry Exploration Program. “It was a coincidence of similar scientific interests that sparked a natural conversation,”

Box 3 A mutually beneficial collaboration Benefits to participating postdocs: • A deeper appreciation for what it is like to work in industry • Insight into the culture at different companies • Potential scientific collaborations • Career guidance Benefits to participating companies: • Improvement of the company’s perception among postdoc scientists • Exposure to potential colleagues • Establishment of a working relationship between local academic institutes • Potential scientific collaborations

said Sallum. “The Industry Exploration Program gives participants an opportunity to see inside the black box of industry and allows participants to experience a broad range of philosophies and opportunities available within industry, which is too frequently considered a single alternative to academia.” The program has been well received by the research community and had one of the three award-winning posters at the 8th Annual National Postdoc Association Meeting in Philadelphia (http://www.nationalpostdoc.org/). Feedback from participating companies has also been positive. When asked, industry scientists indicated that they enjoyed the experience and said they wished the program had been available to them when they were making their transition from academy to industry. With the help of MassBioEd the program will expand to include additional companies this fall to match the broad range of interests that are represented in the MGH postdoc community. These include additional pharmaceutical R&D companies as well as biomedical engineering, patent law firms and medical science liasions. Ultimately, our goal is is to bridge the gap between academia and industry, expanding the long-term benefit to both parties (Box 3). For example, at MGH, the relationships have extended beyond the Industry Exploration program to the Annual Postdoc Poster Day, which has been transformed into a Career Day. The session will include a career option panel featuring many representatives from our industry partners. Time will tell how many participants of the industry exploration program will pursue careers outside academia, but their experience left all with a better understanding of how to answer the academia or industry question for themselves. Looking forward we can see programs like the one described here leading to many exciting scientific opportunities for academia and industry to work together on pressing issues, and creating more comprehensive programs that train postdocs for successful careers upon the completion of their training. ACKNOWLEDGMENTS The authors would like to thank L. Hartford of MassBioEd for his continued enthusiasm and support of this endeavor; and J. Dyhrfjeld-Johnsen, H. Luderer, D. Reif, J. Ellard, F. Firouz, B. Bodnaryk, I. Nasrullah, W. Jack, B. Slatko, L. Tonello, J.-O. Funk, K.-M. Lo, B. Tedeschi, I. Erenburg, J. Davis, MGH postdoc participants, D. Marzullo, A. Lew and many others who were instrumental to the success of this program. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

volume 28 number 6 JUNE 2010 nature biotechnology

© 2010 Nature America, Inc. All rights reserved.

people

Optimer Pharmaceuticals (San Diego) has announced the appointment of Pedro Lichtinger (left) as president, CEO and member of the board of directors. He joins Optimer with more than 30 years of experience in the pharmaceutical and animal health industries, most recently as president of Pfizer’s global primary care business unit, where he oversaw operations in North America, Europe, Korea and Australia. Optimer cofounder Michael N. Chang, who has been president and CEO since the company’s inception, will serve as chairman of the board and as a consultant. “Pedro’s successful track record in the global pharmaceutical industry will benefit us as we focus on the commercialization of our lead drug candidate, fidaxomicin,” says Chang. “His experience leading R&D and commercial operations will be invaluable.” Optimer has also announced that Alain B. Schreiber, managing partner at ProQuest Investments, has stepped down from his position as a director. Schreiber has served on the company’s board since 2001.

David L. Barker has been elected to the board of directors of BioNanomatrix (Philadelphia). Barker was vice president and CSO at Illumina from 2000 through 2006. He currently serves on the boards of directors of Cell Biosciences, Integrated Diagnostics, IntegenX and NextBio, and is a scientific advisor to Illumina, GenVault, Diagnostics for the Real World, Helixis, BiOptix and DNA Direct. Silence Therapeutics (London) has announced the promotion of Jorg Kaufmann to the position of vice president of research. Kaufmann, who previously served as senior director, technologies, has been at Silence since 2000. Immunovaccine (Halifax, Nova Scotia, Canada) has announced the appointment of Paul Kirkconnell to its board of directors. Kirkconnell is founder of PAK Limited, a biopharmaceutical consulting firm, and was most recently managing director of DRI Capital, a $1 billion healthcare investment company. In addition, Denis Ryan has stepped down from Immunovaccine’s board after 5 years of service. Emmanuel le Poul has announced his departure as head of the CNS business unit at Addex Pharmaceuticals (Geneva), where he has served since 2003. Addex CEO Vincent Mutel will resume responsibility for the CNS business unit, a position he held before the promotion of le Poul. NOXXON Pharma (Berlin) has named Aram Mangasarian as chief business offi628

cer and member of the executive committee. Mangasarian brings more than 10 years of biotech and pharma industry business development and strategic planning experience. Previously, he served as vice president of business development at French biotech companies Novexel and ExonHit Therapeutics. Tarek S. Mansour has been appointed executive vice president, R&D of Xenon Pharmaceuticals (Vancouver, BC, Canada). He was most recently vice president, chemical sciences at Wyeth Pharmaceuticals (now Pfizer). Cobalis (Irvine, CA, USA) has named Martin Marion CEO. He has over 30 years of strategic planning and healthcare marketing experience. He was formerly chief marketing officer for Cobalis from 2003 through 2006 and has served as a consultant to the company for the past year. Epizyme (Cambridge, MA, USA) has named Mikel Moyer vice president of molecular discovery. He joins Epizyme from the Broad Institute, where he was director of medicinal chemistry in the Stanley Center for Psychiatric Research. Before going to the Broad, Moyer held leadership positions in medicinal chemistry groups at Pfizer for nearly 20 years. Stephen Murray has been appointed chief medical officer of Affectis Pharmaceuticals (Martinsried, Germany). He previously

served in the same capacity at Memory Pharmaceuticals before it was sold to Roche in 2009. Nancy Kreis Newman has joined Vestaron (Kalamazoo, MI, USA) as vice president of finance and CFO. She was previously CFO of A.M. Todd Group. David R. Parkinson has been appointed to the board of directors of Threshold Pharmaceuticals (Redwood City, CA, USA). He is president and CEO of Nodality and was formerly senior vice president, oncology R&D at Biogen Idec. Privately held Nuon Therapeutics (San Mateo, CA, USA) has appointed Lee M. Rauch president and CEO. Rauch was most recently acting head of corporate development for Onyx Pharmaceuticals. She also served as chief business officer at Point Biomedical and at Onyx. Previously, she was senior vice president of corporate development at COR Therapeutics. Novasep (Pompey, France) has announced the appointments of Stephen F. Stefano as president and CEO of its North American business and Patrick A. Glaser as president and CEO of the Novasep Synthesis division. Both men also join Novasep’s executive management committee. Stefano was previously senior vice president at GlaxoSmithKline Pharmaceuticals in charge of business development for North America and new product planning. Glaser was most recently vice president of Dr. Reddy’s Laboratories’ active pharmaceutical ingredient business and head of strategic contract manufacturing acquisition and integration. Karen Zaderej has been named CEO of AxoGen (Alachua, FL, USA). She previously served as COO. Outgoing CEO Jamie Grooms will retain the role of chairman of the board for AxoGen. WuXi PharmaTech (Shanghai) has announced that it has appointed Hao Zhou as CFO. Zhou joined the company in May 2009 as vice president of finance. Before joining WuXi, he worked for 11 years at General Electric, most recently serving as finance director, Greater China, for GE Healthcare Medical Diagnostics and Life Sciences.

volume 28 number 6 JUNE 2010 nature biotechnology