No title

volume 29 number 3 MARCH 2011 1 5 t h a n n i v ersary i ss u e e d i tor i a l 171 Realigning interests news © ...

Author: Andrew Marshall

119 downloads 2242 Views 12MB 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 29 number 3 MARCH 2011

1 5 t h a n n i v ersary i ss u e

e d i tor i a l 171

Realigning interests

news

© 2011 Nature America, Inc. All rights reserved.

Nature Biotechnology celebrates 15 years of publishing the very best of biotech science and business. Cover art: Marina Corral.

pro f i l e 183 Gary Pisano 184 Stelios Papadopoulos 185 Paul Keckley 186 Merv Turner 187 Anthony Coyle 188 Elias Zerhouni 189 Edison Liu 190 Greg Winter 191 Lee Hood 192 Robert Weinberg 193 Arnold Demain 194 Irv Weissman 195 Barbara Mazur 196 Lee Lynd

op i n i o n a n d comme n t 215 215 218

C O M M E N TA R Y Biomedical technology and the clinic of the future point: Are we prepared for the future doctor visit? Stephen H Friend & Trey Ideker counterpoint: Do not opine before it’s time Isaac S Kohane & David M Margulies

f eat u re 221 Five more years of Nature Biotechnology research Monya Baker & Laura DeFrancesco 25 years of antibody display, p 245

229

pate n ts Unsettled expectations: how recent patent decisions affect biotech Brenda M Simon & Christopher T Scott

comp u tat i o n a l b i o l ogy pro f i l e 243 David Haussler

research 245

perspect i v e Beyond natural antibodies: the power of in vitro display technologies Andrew R M Bradbury, Sachdev Sidhu, Stefan Dübel & John McCafferty

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$4,048 (institution), US$4,658 (corporate institution). Canada add 5% GST #104911595RT001; Euro-zone: €202 (personal), €3,214 (institution), €4,011 (corporate institution); Rest of world (excluding China, Japan, Korea): £130 (personal), £2,077 (institution), £2,588 (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, 75 Varick Street, 9th Floor, New York, NY 10013-1917. 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 © 2011 Nature America, Inc. All rights reserved. Printed in USA.

i

volume 29 number 3 march 2011 news

© 2011 Nature America, Inc. All rights reserved.

Biotic video games, p 181

173 Fate of novel painkiller mAbs hangs in balance 174 Novel agents combined get own guidance 175 Dreamboat sinks prospects for fast approval of inhaled insulin 176 Courts back Prometheus IP 176 Accelerated approvals examined 177 Biosimilars encircle Rituxan, US debates innovator exclusivity 178 NCI revamps trials 178 Yardsticks for R&D 179 Industry exhales as USDA okays glyphosate-resistant alfalfa 179 DuPont swallows Danisco 179 Alzheimer’s genetic map 181 Video games played with live organisms 197 News Feature: Fresh from the biologic pipeline—2010 201 News Feature: The power of many

B i oe n trepre n e u r B u i l d i n g a b u s i n ess 205 Divining the path to a successful European exit Håkon Kirkeby Buch, Anna C Gustafsson, Viktor Drvota & Carl Johan Sundberg

op i n i o n a n d comme n t 208 210 212

C O R R E S P O ND E N C E Strengths and limitations of the federal guidance on synthetic DNA Partnering Brazilian biotech with the global pharmaceutical industry Survival and growth of Arabidopsis plants given limited water are not equal

f eat u re The promise of crowdsourcing, p 201

pate n ts 231

Recent patent applications in antibody fragments

N E W S A ND VI E W S 233 235 236 238 241

IPSCs put to the test see also p 279 Hyesoo Kim & Lorenz Studer Chemoproteomics quantifies complexity Edward B Holson & Stuart L Schreiber see also p 255 Biomarkers in aggregate Fred S Apple see also p 273 A modENCODE snapshot Markus Elsner & H Craig Mak Research highlights

Nanoparticles report biomarker accumulation, p 236

nature biotechnology

iii

volume 29 number 3 march 2011 research ARTICLEs 255 Chemoproteomics profiling of HDAC inhibitors reveals selective targeting of HDAC complexes Marcus Bantscheff, Carsten Hopf, Mikhail M Savitski, Antje Dittmann, Paola Grandi, Anne-Marie Michon, Judith Schlegl, Yann Abraham, Isabelle Becher, Giovanna Bergamini, Markus Boesche, Manja Delling, Birgit Dümpelfeld, Dirk Eberhard, Carola Huthmacher, Toby Mathieson, Daniel Poeckel, Valérie Reader, Katja Strunk, Gavain Sweetman, Ulrich Kruse, see also p 235 Gitte Neubauer, Nigel G Ramsden & Gerard Drewes Drug interactions with large protein complexes, p 255

© 2011 Nature America, Inc. All rights reserved.

267

273

279

l etters Generation of anterior foregut endoderm from human embryonic and induced pluripotent stem cells Michael D Green, Antonia Chen, Maria-Cristina Nostro, Sunita L d’Souza, Christoph Schaniel, Ihor R Lemischka, Valerie Gouon-Evans, Gordon Keller & Hans-Willem Snoeck Implantable magnetic relaxation sensors measure cumulative exposure to cardiac biomarkers Yibo Ling, Terrence Pong, Christophoros C Vassiliou, Paul L Huang & see also p 236 Michael J Cima reso u rce A functionally characterized test set of human induced pluripotent stem cells Gabriella L Boulting, Evangelos Kiskinis, Gist F Croft, Mackenzie W Amoroso, Derek H Oakley, Brian J ainger, Damian J Williams, David J Kahler, Mariko Yamaki, Lance Davidow, Christopher T Rodolfa, John T Dimos, Shravani Mikkilineni, Amy B MacDermott, Clifford J Woolf, Christopher E Henderson, Hynek Wichterle & see also p 233 Kevin Eggan

careers a n d recr u i tme n t 287

A mentoring program for women scientists meets a pressing need Masha Fridkis-Hareli

290

people

Testing iPSC differentiation, p 279

nature biotechnology

v

in this issue

© 2011 Nature America, Inc. All rights reserved.

Nature Biotechnology turns 15! Nature Biotechnology first launched 15 years ago. As we celebrate our crystal anniversary, the biotech sector is facing a whole host of funding, regulatory, legal, political and market challenges. As companies adapt to the new environment in which there is a shortfall in risk capital for innovative R&D [Editorial, p. 171], reformed healthcare systems seek to pay for performance, patent threats loom over not only small molecules but also biologics, increasing government and regulatory oversight becomes ever more stringent and the drug pipeline continues to be plagued by productivity issues. We went out and asked 15 experts to give us their views on the state of the sector and the prospects for translation of innovative research [Profiles, pp. 183–196]. We’ve also reached out to those interested in building biotech sectors from around the world and asked them to keep us updated in a new blog (http://blogs.nature.com/trade_secrets/). Readers interested in listening to a discussion of the challenges in life science commercialization can listen to our anniversary podcast (http://www. nature.com/nbt/podcast/index.html). One area where there’s been considerable upheaval is in patent case law. And uncertainty over eligibility, obviousness and disclosure has the potential to profoundly affect the sector [Patents, p. 229]. Because we remain some way off from implementation of personalized medicine, we have commissioned a discussion of the barriers to the application of systems approaches and largescale data in the practice of clinical medicine [Commentary, p. 215]. In addition, we talk to a leading authority about the challenges in annotating these large-scale data sets [Profile, p. 243]. We’ve also taken a brief look back over the most often cited papers published in our pages [Feature, p. 221]. Finally, it is now a little over 25 years since the invention of phage antibody display. To celebrate, we present an overview of the antibodies that have been obtained using display technologies. Many of these have characteristics that would have been extremely difficult, if not impossible, to derive through traditional immunization strategies [Perspective, p. 245]. AM

Dissecting complexity with chemoproteomics Efforts to characterize the interactions of drugs that target components of megadalton protein complexes traditionally involve only the purified subunit in question. Working with histone deacetylase (HDAC) inhibitors, Drewes and colleagues show the importance of conducting more physiologically relevant assays. The authors first use beads modified with HDAC inhibitors to trap interacting proteins from cell lysates containing varying concentrations of the HDAC inhibitor of interest. They then use quantitative mass spectrometry to determine drug affinities for the HDAC complexes and other targets recovered. Perhaps the most striking outcome is the demonstration that different HDAC complexes, all of which contain the HDAC1-HDAC2 catalytic core, show different selectivities for the 16 HDAC inhibitors tested. This demonstration that small molecules can discriminate between distinct protein complexes suggests that the concept of what constitutes a drug target may need to be refined for Written by Kathy Aschheim, Michael Francisco, Peter Hare, Andrew Marshall & Lisa Melton

nature biotechnology volume 29 number 3 march 2011

many small-molecule modulators. Other notable findings include the identification of several novel HDAC complexes and the demonstration that the orphan anti-inflammatory drug bufexamac is a class IIb HDAC inhibitor. [Articles, p. 255; News and Views, p. 235] PH

iPSCs no worse than ESCs How similar are induced pluripotent stem cells (iPSCs) to embryonic stem cells (ESCs)? Recent studies have reported differences between the two cell types in DNA methylation, gene expression and differentiation capacity. Eggan and colleagues have carried out the largest comparison to date at the functional level, establishing a test set of 16 iPSC lines and 6 ESC lines and analyzing differentiation to motor neurons. Contrary to previous work, they conclude that, on average, the performance of iPSCs in generating neurons is no worse than that of ESCs. Like ESC lines, however, individual iPSC lines do vary in their differentiation ability. The authors attempt to explain this variability and rule out several possible causes, including differences in donor age, transgene silencing, karyotype, donor health status and cell handling between labs. Two factors that may be relevant are donor identity and donor sex. This well-characterized set of iPSC lines is available from the authors and should provide a useful resource for further studies. [Resource, p. 279; News and Views, p. 233] KA vii

in this issue

Anterior foregut endoderm from pluripotent cells

© 2011 Nature America, Inc. All rights reserved.

Until now, research on differentiating human pluripotent stem cells to endodermal lineages has focused on cell types derived from the midgut or posterior foregut endoderm, such as pancreatic and hepatic cells. But the most rostral part of the endoderm—the anterior foregut endoderm (AFE)—also gives rise to tissues of great interest for regenerative medicine, including thymus, thyroid, parathyroid, lung and trachea. Snoeck and colleagues describe a method for generating AFE from embryonic stem cells and induced pluripotent stem cells. The approach relies on treating activin A–induced

Patent roundup As biosimilar developers launch phase 2 trials, a debate has erupted over innovator data and market exclusivity and the use of evergreening to extend patent life. [News, p. 177] LM Diagnostics companies cheered as the US Supreme Court in Prometheus Labs Inc v. Mayo Collaborative Srvs. upheld an earlier decision that two Prometheus Labs tests for defining drug dosage are patentable. [News, p. 176] LM Several recent court decisions have drastically changed the biotech patent landscape, according to Simon and Scott. A look back shows that broad patents are a thing of the past and biotech inventors face heightened requirements for patentability. [Patents, p. 229] MF Recent patent applications related to antibody fragments. [New patents, p. 231]

viii

MF

definitive endoderm with NOGGIN, an inhibitor of bone morphogenic protein signaling, and SB-431542, an inhibitor of transforming growth factor-β signaling. The authors also study the differentiation of AFE to downstream lineages, and show expression of parathyroid and lung markers. [Letters, p. 267] KA

Cumulative biomarker measurement Assaying biomarker levels at discrete time points may miss fluctuations relevant to research and therapeutic applications. Cima and colleagues describe the use of subcutaneously implanted sensors to integrate cumulative exposure to three markers of myocardial infarction over a 72-h period. The sensors contain superparamagnetic nanoparticles that aggregate upon exposure to the target protein. This generates a signal that can be read noninvasively by magnetic resonance relaxometry. Because total biomarker release closely reflects the extent of cardiac tissue death in a mouse model of myocardial infarction, the authors can correlate the sensor response with infarct size, a parameter that cannot be easily measured using measurements taken at a single time point. They further use a model of doxorubicin-induced cardiac damage to demonstrate the potential of cumulative biomarker measurements for use in drug screening. [Letters, p. 273; News and Views, p. 236] PH

Next month in • Label-free quantification of membrane binding • Exosomes deliver siRNA to the brain • Protein-based biorefinery • Exploring the function of essential genes

volume 29 number 3 march 2011 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 Production Editor: Amanda Crawford Senior Illustrator: Katie Vicari Illustrator: Marina Corral Cover design: Erin DeWalt Senior Editorial Assistant: Ania Levinson

© 2011 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 Exectutive Editor: Veronique Kiermer 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

Realigning interests Fifteen years after Nature Biotechnology was launched, the old paradigms in life science commercialization are no longer tenable. It’s time to realign the interests of companies, patients and payors so that innovation is prioritized.

© 2011 Nature America, Inc. All rights reserved.

W

hen this journal celebrated its tenth anniversary five years ago, we exhorted the flagship companies to do more to reinvest profits in innovative science and technology. At that time, investors were increasingly unwilling to embrace the risk of early stage research. R&D productivity was in decline. Only a handful of drugs with novel mechanisms were making it to the clinic each year. Today, that situation is even bleaker. Fewer startups are being funded that center around really innovative science. And the types of companies and enterprises that do form are unlikely to be an adequate way of meeting the unprecedented demographic changes and disease burden that will face our societies. If these challenges are to be met, multiple interests must/have to be realigned so that investors and the commercial sector have more incentive to develop novel therapies. Uncovering new drugs is incredibly difficult. Even if we can address the scientific challenges of understanding human biology, the current investment, regulatory and reimbursement regimes that now govern the sector actively discourage industry from embracing innovation. From the investment standpoint, the mantra for access to the public markets is now a compound in proof-of-concept trials—a tall order for a company that has plowed substantial resources into developing a drug with a novel mechanism. Both big pharma and, increasingly, big biotech companies are focusing their efforts on satisfying investors via share buyback programs or modifying existing drugs or extending a product franchise to new indications. The risk equation for investing in innovative drug programs simply does not add up. Even if a company successfully brings an innovative compound through the clinic, regulators continue to raise the quality of data to be gathered in support of an approval, particularly for a novel drug with a novel mechanism. And facing an ever-stretching demand with finite resources, healthcare payors no longer guarantee premiums for innovative new drugs. At the same time, every stakeholder in the translation process— pharma, government and the research base—has its own diagnosis of the cause of the problem and its own bolt-on solutions. Pharmaceutical companies want to get closer to the academic and clinical research base both to spot opportunities earlier and to channel them appropriately into their R&D pipelines. But it’s hard not to view these initiatives as merely go-faster stripes on the pharma gas-guzzler. When these organizations engage generics companies in pay-for-delay collusions to keep cheaper replacement molecules off the market and axe research jobs by the thousand while only redeploying dozens in academic links, what message does that send about innovation? As part of his proposed budget for fiscal year 2012, US president Barack Obama has boosted funding for translational research via the US National Institutes of Health (NIH) by some $745 million— a large sum at a time of austerity. The NIH itself has also been taking

nature biotechnology volume 29 number 3 march 2011

translational activity seriously. Indeed, NIH director Francis Collins is proposing a specific institute devoted to translation—the National Center for Advancing Translational Sciences. The flaw in the plan is that the NIH is focused primarily on basic research questions. This is as it should be. But it begs the question as to whether the NIH—or academia for that matter—is well-placed to pick translational winners from basic research. As so many biotech research programs are simply ‘parked’ because continuation is not commercially supportable, the question is whether the public sector can be oriented to unblock such research areas and reopen commercial avenues. Fundamentally, then, neither the feeder research programs nor the commercial exploitation machinery is properly aligned with what both regard as the overall driver of healthcare innovation—improving health outcomes. How can the machinery be reoriented to prioritize truly innovative drug development? Simply providing more funding to venture capitalists in the present environment is unlikely to change the type of science that is funded. Investors will simply fund more of the same ventures rather than those based on truly innovative science. To prioritize investment in the type of science needed to address the healthcare system’s disease priorities, such as untreatable cancers, chronic diseases of aging and neurodegeneration or emerging infections, a market pull mechanism is needed to recalibrate the risk equation. One way this could be accomplished is for government to designate which indications are poorly served through current drugs, represent a burden on the healthcare system and require medical innovation. Incentives could then be offered to companies or institutions that develop experimental drugs addressing these indications. For example, the US government could co-fund late-stage pivotal clinical research in indications identified as broader health priorities, especially for drugs that have novel mechanisms of action. In doing so, it could both signal its specific priorities and recognize and address the risk in innovation. The extent of government participation could be adjusted according to the order of its healthcare priorities. Alternatively, an increased period of data exclusivity—the time between approval of an innovator drug and the entry onto the market of the same or highly similar product from a competitor—could be offered. The potential returns of five extra years of exclusivity for an innovative drug might make the risk worthwhile, especially if the drug turns out to be a blockbuster. Government thus has a choice. Keep the alignment in status quo and risk a partial or total eclipse of the innovative biomedical enterprise. Or intervene and realign market incentives so that investors and companies can embrace the high risk science needed to develop the medicines of tomorrow. 171

news

in this section Inhaled insulin’s last gasp p175

Biosimilar mAbs begin onslaught

USDA green light for glyphosateresistant alfafa

p177

p179

© 2011 Nature America, Inc. All rights reserved.

Fate of novel painkiller mAbs hangs in balance A major new class of pain medications suffered a severe setback in December when the US Food and Drug Administration (FDA) placed a hold on most clinical trials for experimental therapies targeting nerve growth factor (NGF). These biologics were poised to be the first important new class of drugs for general pain since the prototype non-steroidal anti-inflammatory drug (NSAID), aspirin, came into general use at the end of the nineteenth century. “We’ve had over a hundred years without having really a major new pain drug,” says Thomas Schnitzer, a professor of medicine at Northwestern University in Chicago. With this latest stumbling block, the fate of antiNGF painkillers hangs in the balance. New York–based Pfizer leads the quest for novel painkillers with tanezumab (PF-4383119), a humanized monoclonal antibody (mAb) that blocks NGF. Preclinical models of pain suggest that this mAb is equally or more efficacious than opiates or NSAIDs. Tanezumab was originally discovered by Genentech spin-off Rinat of South San Francisco. The antibody passed to Pfizer in 2006 when the pharma giant acquired Rinat. But tanezumab’s steady progress faltered last summer when some individuals in phase 3 osteoarthritis trials developed cases of joint damage. Sixteen of them needed surgery to replace joints when they developed progressively worsening osteoarthritis with evidence of bone necrosis inferred from radiographic images. The FDA responded by placing on hold phase 3 trials testing Pfizer’s NGF inhibitor in osteoarthritis and some other pain indications in June and July. Then in December, an additional case of joint failure cropped up, suggesting problems with the entire class of drugs. The FDA did not release information on which drug caused the additional case, yet it still decided to extend the hold to other companies developing competing NGF-targeting antibodies (Table 1). But according to insiders, anti-NGF antibodies are far from dead. Pfizer’s tanezumab phase 3 studies involved >10,000 patients, with four trials now complete. And though at least six other trials have been terminated, the company is not yet giving up. “From Pfizer’s perspective, no causal relationship has been established between tanezumab and the reported adverse events,” says company spokesperson

MacKay Jimeson. Pfizer’s phase 2 trials in cancer pain have been allowed to proceed. Other companies investing in NGF painkillers are also cautiously optimistic. “We plan to continue working with the regulators, such as the FDA, to try to understand whether and where there is sufficient risk-benefit ratio for this class of drugs to be used,” says George Yancopoulos, CSO of Regeneron of Tarrytown, New York. “And whether there really is any class effect that might be causing these problems.” Regeneron has brought its anti-NGF antibody REGN475/ SAR164877, generated using the company’s VelocImmune technology (mice in which the murine variable regions have been replaced with their human heavy and light chain counterparts), into phase 2 trials, in partnership with Sanofi-aventis of Paris. Fueling hope is the fact that no one, except possibly the FDA, knows how frequent joint failure was in the tanezumab study’s treatment groups as opposed to the control groups. (The FDA declined comment.) “Since many of the patients in the trial had advanced osteoarthritis with pain, we would expect at least some of them to continue to advance to the point of needing joint replacement, no matter what treatment they were given,” writes Richard Loeser, a

professor of molecular medicine at Wake Forest University in Winston-Salem, North Carolina, in an e-mail to Nature Biotechnology. “So we need all the numbers from all of the subjects in the control and treatment arms from all of the trials…to see if there is a significant difference between placebo control and anti-NGF.” The placebo groups were small relative to the treatment arms, making comparisons difficult. “When you have events that are so rare, the numbers are so low, it’s hard to prove that it’s really drug related,” says Yancopoulos. And anti-NGF antibodies control pain extremely well. These drugs work through a completely novel analgesic mechanism of action by reducing signaling through NGF receptors TrkA and p75, dampening excitability in pain neurons. Tanezumab “is a very effective drug,” says Schnitzer, co-author of the New England Journal of Medicine (363, 1521–1531, 2010) paper that reported phase 2 results for tanezumab in individuals with knee osteoarthritis. In that trial, tanezumab injections eight weeks apart reduced knee pain by a mean of 45–62% from baseline with various doses, compared with 22% reduction for placebo. Regeneron reported similar phase 2 efficacy for its molecule at the American College of Rheumatology annual Sensitizers/receptors

Signal transduction Peripheral terminal

Cell body Dorsal root ganglion

B2

Bradykinin

EP

PGE2 NGF

TRKA

TNFα

TNFR1 NK1

Substance P

Prokinetican PKR

Trafficking

ETAR

Endothelin

Transcription Translation Trafficking

PKCε PKA P13K ERK

PAR2 TRPV1

TRPV1

Phosphorylation P

TRPV1

TRPV1

NGF/ GDNF

Anterograde transport

Ca2+ TRPA1 B2

Reduced pain threshold

Peripheral sensitization

Targeting pain with anti-nerve growth factor antibodies. Inflammation produces several inflammatory factors, most notably nerve growth factor, which sensitize nerve cells by acting on their cognate receptors and activating signal transduction. These activated pathways phosphorylate transient receptor potential (TRP) channels, which alter their trafficking and reduce the membrane’s threshold, resulting in an increased excitability of pain neurons. Image courtesy of Nat. Rev. Drug. Disc. (8, 55–56, 2009).

nature biotechnology volume 29 number 3 MARCH 2011

173

© 2011 Nature America, Inc. All rights reserved.

Studio Heinemann/Westend61/Newscom

NEWS

in brief

Table 1 Selected anti-NGF antibodies in clinical development for pain

Novel agents combined get own guidance

Company

Antibody

Pfizer

Tanezumab (RN-624), a humanized anti-NGF mAb

Phase 3

All phase 3 trials either completed or terminated Phase 2 in cancer pain ongoing

Amgen (Thousand Oaks, California)/J&J (New Brunswick, New Jersey)

Fulranumab, a fully human anti-NGF mAb (JNJ42160443) (AMG-403)

Phase 2

Trials suspended

Regeneron/Sanofi-aventis

REGN475/SAR164877, a fully human anti-NGF mAb

Phase 2

Program active, osteoarthritis pain trials on hold

Medimmune (Gaithersburg, Medi-578, a monoclonal single chain variable fragment (scFv) Maryland)/AstraZeneca against NGF (London)

Phase 2

Trials suspended

Abbott Laboratories (Abbott Park, Illinois)

Phase 1

Osteoarthritis trial enrollment complete, trial ongoing

Companies welcome the draft guidance published by the US Food and Drug Administration (FDA) clarifying the regulatory issues involved in developing novel, experimental drugs used as combinations. Recognizing that there are occasions when coDrug combinations developing two or more gain regulatory path. investigational agents may provide significant therapeutic advantages, the FDA issued a public call for comments. Fifteen firms and organizations including Celgene of Summit, New Jersey, MedImmune of Gaithersburg, Maryland, the Melanoma Research Alliance and the Biotechnology Industry Organization submitted responses. “These comments were considered in the development of the draft guidance,” says Crystal Rice, a spokesperson for the FDA. Until this draft guidance was published, companies were forging their own regulatory paths for the co-development of investigational drugs. Because many large companies already have a number of such combinations in early clinical trial testing (Nat. Biotechnol. 28, 765–766, 2010), the opportunity to receive consistent advice from the regulators has been particularly well received. However, Rice explains that the guidance does not describe a one-sizefits-all development program for combination therapies. She acknowledges that the amount and types of clinical data needed and appropriate study designs will vary depending on the nature of the combination being developed, the disease and other factors. “The FDA anticipates that the finer details of individual development programs will usually be worked on a case-bycase basis,” says Rice. Importantly, the new guidance (http://www.fda.gov/downloads/Drugs/ GuidanceComplianceRegulatoryInformation/ Guidances/UCM236669.pdf) does not apply to already marketed drugs in fixed-dose combinations or to the development of an investigational drug and an approved drug. “Historically, the independent contribution of each participating drug in a combination needed to have been characterized beyond reasonable doubt, which often led sponsors to combine novel drugs with registered standards of care,” says Andrew Hughes, global clinical vice president of early oncology development at AstraZeneca. Instead, “the guidance document facilitates the earlier clinical appraisal of promising combinations of two unregistered drugs,” adds Hughes, for instance, in vivo and in vitro evidence to support the biological rationale for a particular combination. For AstraZeneca, the FDA’s input is timely because they have just announced a strategic alliance with Cancer Research UK to take combinations of experimental cancer drugs into early phase clinical trials. Bethan Hughes

174

Latest stage Status

ABT-110 (PG110), a humanized anti-NGF mAb

meeting in November. In both trials, neurosensory problems like abnormal sensitivity to touch were the most common side effects. “Almost all of them are limited in duration and not particularly significant in terms of how people feel or function,” says Schnitzer. “I don’t see that as a major issue.” Joint failure is another story. “Clearly it occurs, and clearly that’s an important issue that needs to be further evaluated,” says Schnitzer. In a New England Journal of Medicine editorial, neuroscientist John Wood of the Wolfson Institute for Biomedical Research, University College London, put forth a controversial explanation for the joint failures. He suggested that pain control may be so successful in these patients that they overuse their joints. “A complete quenching of pain in osteoarthritis may not be a good thing,” wrote Wood, whose research focuses on pain pathways. Does tanezumab mask pain so well that patients hurt themselves? “There’s every reason to believe that if you have a very effective analgesic agent that certainly can happen,” says Schnitzer. But he also points out that no one knows whether the failed joints were already damaged before treatment or whether healthy joints were caused to fail because baseline X-rays of all joints were not taken—crucial information in any risk-benefit calculation for individual patients. Others find the overuse hypothesis questionable. “We know that exercise, including walking, does not exacerbate osteoarthritis or cause it to progress more rapidly,” says Wake Forest University’s Loeser. “And if it did, [we] would not expect it to be in the form of necrosis.” Loeser thinks a direct anti-NGF effect on bone is equally plausible. Animal experiments in the late 1990s showed that Trk receptors (which bind NGF) are present on bone-forming cells. Also, giving NGF to rats had beneficial effects on fracture healing. “If you block NGF action with monoclonal antibodies it may interfere with bone remodeling,” writes Brian Grills, a bone researcher at La Trobe University in

Victoria, Australia, in an e-mail. Grills, who performed the rat fracture experiments, explains that because sensory nerves have a trophic influence on bone, another possibility is that NGF blockade could interfere with bone metabolism. A third scenario is that NGF blockade could have negative effects on blood vessels, leading to avascular necrosis, a mechanism that Regeneron is actively investigating. “Right now I would say that the evidence is to the contrary,” says Yancopoulos. Overly focusing on mechanism is premature, he adds, especially because it is still uncertain that the joint failures were drug related, much less that they’re a class effect. “Until we understand clinically what the problem is, it becomes double speculation,” he says. And without more information, it’s impossible to know whether the agency’s hold is temporary, or whether it harbingers the demise of anti-NGF antibodies for pain, at least for musculoskeletal pain like osteoarthritis—the prize indication. “My view is that the class is dead,” says Seamus Fernandez, a pharmaceutical analyst for Leerink Swann in Boston. “The only potential resurrection of the class would be in… cancer-related pain.” That’s premature, counters Yancopoulos. “We don’t disagree that one has to be cautious here,” he says. “But I think it’s very premature to call something dead when one doesn’t even know what the problem really is, or whether it’s drug-related.” Schnitzer hopes that the FDA will convene a panel of experts to publicly examine the data and advise regulators and industry. “It would clear the air a lot to just look at the data and have a frank and open discussion about it, a scientific exchange,” he says. That anti-NGF antibodies have proven to be effective against pain is an important scientific advance, Schnitzer says. But turning them into successful drugs might take more work. “It may be that playing with the pain pathway [is] going to be a balancing act,” he says. “We clearly don’t want an anesthetic agent.” Ken Garber, Ann Arbor, Michigan

volume 29 number 3 MARCH 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

Studio Heinemann/Westend61/Newscom

NEWS

in brief

Table 1 Selected anti-NGF antibodies in clinical development for pain

Novel agents combined get own guidance

Company

Antibody

Pfizer

Tanezumab (RN-624), a humanized anti-NGF mAb

Phase 3

All phase 3 trials either completed or terminated Phase 2 in cancer pain ongoing

Amgen (Thousand Oaks, California)/J&J (New Brunswick, New Jersey)

Fulranumab, a fully human anti-NGF mAb (JNJ42160443) (AMG-403)

Phase 2

Trials suspended

Regeneron/Sanofi-aventis

REGN475/SAR164877, a fully human anti-NGF mAb

Phase 2

Program active, osteoarthritis pain trials on hold

Medimmune (Gaithersburg, Medi-578, a monoclonal single chain variable fragment (scFv) Maryland)/AstraZeneca against NGF (London)

Phase 2

Trials suspended

Abbott Laboratories (Abbott Park, Illinois)

Phase 1

Osteoarthritis trial enrollment complete, trial ongoing

Companies welcome the draft guidance published by the US Food and Drug Administration (FDA) clarifying the regulatory issues involved in developing novel, experimental drugs used as combinations. Recognizing that there are occasions when coDrug combinations developing two or more gain regulatory path. investigational agents may provide significant therapeutic advantages, the FDA issued a public call for comments. Fifteen firms and organizations including Celgene of Summit, New Jersey, MedImmune of Gaithersburg, Maryland, the Melanoma Research Alliance and the Biotechnology Industry Organization submitted responses. “These comments were considered in the development of the draft guidance,” says Crystal Rice, a spokesperson for the FDA. Until this draft guidance was published, companies were forging their own regulatory paths for the co-development of investigational drugs. Because many large companies already have a number of such combinations in early clinical trial testing (Nat. Biotechnol. 28, 765–766, 2010), the opportunity to receive consistent advice from the regulators has been particularly well received. However, Rice explains that the guidance does not describe a one-sizefits-all development program for combination therapies. She acknowledges that the amount and types of clinical data needed and appropriate study designs will vary depending on the nature of the combination being developed, the disease and other factors. “The FDA anticipates that the finer details of individual development programs will usually be worked on a case-bycase basis,” says Rice. Importantly, the new guidance (http://www.fda.gov/downloads/Drugs/ GuidanceComplianceRegulatoryInformation/ Guidances/UCM236669.pdf) does not apply to already marketed drugs in fixed-dose combinations or to the development of an investigational drug and an approved drug. “Historically, the independent contribution of each participating drug in a combination needed to have been characterized beyond reasonable doubt, which often led sponsors to combine novel drugs with registered standards of care,” says Andrew Hughes, global clinical vice president of early oncology development at AstraZeneca. Instead, “the guidance document facilitates the earlier clinical appraisal of promising combinations of two unregistered drugs,” adds Hughes, for instance, in vivo and in vitro evidence to support the biological rationale for a particular combination. For AstraZeneca, the FDA’s input is timely because they have just announced a strategic alliance with Cancer Research UK to take combinations of experimental cancer drugs into early phase clinical trials. Bethan Hughes

174

Latest stage Status

ABT-110 (PG110), a humanized anti-NGF mAb

meeting in November. In both trials, neurosensory problems like abnormal sensitivity to touch were the most common side effects. “Almost all of them are limited in duration and not particularly significant in terms of how people feel or function,” says Schnitzer. “I don’t see that as a major issue.” Joint failure is another story. “Clearly it occurs, and clearly that’s an important issue that needs to be further evaluated,” says Schnitzer. In a New England Journal of Medicine editorial, neuroscientist John Wood of the Wolfson Institute for Biomedical Research, University College London, put forth a controversial explanation for the joint failures. He suggested that pain control may be so successful in these patients that they overuse their joints. “A complete quenching of pain in osteoarthritis may not be a good thing,” wrote Wood, whose research focuses on pain pathways. Does tanezumab mask pain so well that patients hurt themselves? “There’s every reason to believe that if you have a very effective analgesic agent that certainly can happen,” says Schnitzer. But he also points out that no one knows whether the failed joints were already damaged before treatment or whether healthy joints were caused to fail because baseline X-rays of all joints were not taken—crucial information in any risk-benefit calculation for individual patients. Others find the overuse hypothesis questionable. “We know that exercise, including walking, does not exacerbate osteoarthritis or cause it to progress more rapidly,” says Wake Forest University’s Loeser. “And if it did, [we] would not expect it to be in the form of necrosis.” Loeser thinks a direct anti-NGF effect on bone is equally plausible. Animal experiments in the late 1990s showed that Trk receptors (which bind NGF) are present on bone-forming cells. Also, giving NGF to rats had beneficial effects on fracture healing. “If you block NGF action with monoclonal antibodies it may interfere with bone remodeling,” writes Brian Grills, a bone researcher at La Trobe University in

Victoria, Australia, in an e-mail. Grills, who performed the rat fracture experiments, explains that because sensory nerves have a trophic influence on bone, another possibility is that NGF blockade could interfere with bone metabolism. A third scenario is that NGF blockade could have negative effects on blood vessels, leading to avascular necrosis, a mechanism that Regeneron is actively investigating. “Right now I would say that the evidence is to the contrary,” says Yancopoulos. Overly focusing on mechanism is premature, he adds, especially because it is still uncertain that the joint failures were drug related, much less that they’re a class effect. “Until we understand clinically what the problem is, it becomes double speculation,” he says. And without more information, it’s impossible to know whether the agency’s hold is temporary, or whether it harbingers the demise of anti-NGF antibodies for pain, at least for musculoskeletal pain like osteoarthritis—the prize indication. “My view is that the class is dead,” says Seamus Fernandez, a pharmaceutical analyst for Leerink Swann in Boston. “The only potential resurrection of the class would be in… cancer-related pain.” That’s premature, counters Yancopoulos. “We don’t disagree that one has to be cautious here,” he says. “But I think it’s very premature to call something dead when one doesn’t even know what the problem really is, or whether it’s drug-related.” Schnitzer hopes that the FDA will convene a panel of experts to publicly examine the data and advise regulators and industry. “It would clear the air a lot to just look at the data and have a frank and open discussion about it, a scientific exchange,” he says. That anti-NGF antibodies have proven to be effective against pain is an important scientific advance, Schnitzer says. But turning them into successful drugs might take more work. “It may be that playing with the pain pathway [is] going to be a balancing act,” he says. “We clearly don’t want an anesthetic agent.” Ken Garber, Ann Arbor, Michigan

volume 29 number 3 MARCH 2011 nature biotechnology

news

In a move that could spell the end of a long- bridge to the new inhaler. This is not disasstanding dream of inhaled insulin, in January trous, it’s just a delay.” the US Food and Drug Administration (FDA) Still, during the conference call, Mann, issued a complete response letter to Valencia, whose largesse has kept the company afloat in California–based MannKind, asking for more the past, left himself room for maneuvering. data on its inhaled insulin product Afrezza. “I can’t make any commitment at this point. Specifically, the agency requested more infor- Obviously, I believe in this product, and I have mation on the bioequivalence of a second- invested in it, because I believe in it, and I have generation inhaler device, known as Dreamboat, got to find a solution for this,” he said. to the MedTone inhaler used in clinical trials. Not everyone was taken aback by the agency’s After initially putting a brave face on the devel- decision. “It’s not completely surprising because opment, the company announced plans to lay FDA has been conservative of late and insisting off 179 employees, 41% of its workforce, during that all the i’s be dotted and the t’s crossed,” says a quarterly earnings call on February 10. Jay S. Skyler, associate director of the Diabetes The demand for more data was unexpected, Research Institute, University of Miami School as the agency had already issued MannKind a of Medicine in Hollywood, Florida. “MannKind complete response letter last March requesting was changing to a new device—a really sexy bioequivalence data for the devices that dispense new device, I must say—but the FDA said that powdered insulin. The company responded one bridging study was not enough. That’s not with a comparison study and assumed that necessarily an unfair position,” adds Skyler, who would suffice. “We thought we had come to an has acted as consultant for all companies in the agreement as to what would be required, so this field at one time or another. came as a bit of a surprise. But it’s within FDA’s MannKind switched to Dreamboat because purview to [ask for more data],” says Matthew it offers significant advantages, and its developPfeffer, MannKind’s CFO. ment progressed more rapidly than expected. MannKind will soon meet with FDA to dis- “Our estimate was that Dreamboat could be cuss how to proceed. The company had already [ready to launch] as little as 6 months behind decided to conduct further trials that can now [the market approval with MedTone],” says be adapted to include a MedTone comparison Pfeffer. Both devices dispense a human insuarm. “They were primarily intended to be mar- lin formulation of the company’s Technosphere keting studies, but we have the equipment to fill technology. The dry insulin particles are MedTone inhalers,” says Pfeffer. As the trials inhaled and, upon contact with the neutral pH weren’t designed to include a comparison arm, of the alveoli, change to liquid. The Dreamboat that might complicate matters, but at least on is significantly smaller and uses a 10-unit dose the surface it appears most of the FDA’s con- of inhaler powder compared with MedTone’s cerns could be answered. 15-unit dose. In the meantime, to keep afloat, MannKind A lower powder load can not only cut the may have to receive another cash infusion from device’s cost but also cause fewer side effects. Alfred Mann’s deep pockets. MannKind’s chair- When patients start using an inhaler of any man and CEO has already personally invested $925 million of the $1.5 billion that has been ploughed into the company. A partnership or acquisition could also be on the horizon. The company has no intention of giving up, Pfeffer says. “[FDA’s requests are] relatively straightforward. They’re not raising safety or efficacy issues. They’re Thumb-sized device. MannKind’s inhaled insulin product, here shown with just questioning the Dreamboat dispenser, suffered a surprise setback. MannKind

© 2011 Nature America, Inc. All rights reserved.

Dreamboat sinks prospects for fast approval of inhaled insulin

nature biotechnology volume 29 number 3 MARCH 2011

175

NEWS

in brief

© 2011 Nature America, Inc. All rights reserved.

Courts back Prometheus IP In a ruling closely watched by developers of companion diagnostics, the US Court of Appeals for the Federal Circuit recently concluded that two methods for determining the optimal dosage of drugs to treat autoimmune diseases are patentable. The December 17, 2010, ruling reaffirmed the court’s earlier decision in Prometheus Labs. Inc. v. Mayo Collaborative Services. Prometheus Laboratories of San Diego sued the Mayo Clinic for patent infringement when the medical group applied an in-house diagnostic test instead of sending samples to Prometheus. The Mayo Clinic claimed the process of giving a drug, observing its effects and adjusting the dosage is an abstract idea that was around before Prometheus patented the test. But the Federal Circuit upheld the patent. Then soon after the Supreme Court’s Bilski v. Kappos decision (Nat. Biotechnol. 28, 767, 2010), in which the court determined that the ‘machine-or-transformation’ test was only one of the considerations for an invention’s patentability, it vacated the Federal Circuit’s ruling and ordered the court to issue a new one. Prometheus argued the Bilski decision did not merit a reversal, as the tests “involve a particular transformation of a patient’s body and bodily sample and use particular machines to determine metabolite concentrations in a bodily sample.” The court came back with the same decision—good news for companies wanting to develop and patent companion diagnostic tests. Michael Francisco

Accelerated approvals examined The US Food and Drug Administration (FDA) is seeking to improve the much-criticized accelerated approval program by reviewing six drugs approved under this pathway. The agency’s Oncologic Drugs Advisory Committee (ODAC) held a meeting on February 8 to scrutinize Eli Lilly’s Erbitux (cetuximab), GlaxoSmithKline’s Bexxar (tositumomab) and Arranon (nelarabine); Genzyme’s Clolar (clofarabine), Amgen’s Vectibix (panitumumab) and Novartis’ Gleevec (imatinib). The committee’s intention was to analyze the process that brought these drugs to market without full confirmation that they are safe and effective. ODAC concluded that to grant accelerated approval, the agency should require a randomized trial, which could measure a surrogate endpoint. The panel also proposed that at the time of gaining accelerated approval, two randomized controlled trials should be under way. “The real issue is that lots of drugs are approved that are not terribly efficacious,” argues Laurence Baker, chairman of the Southwest Oncology Group, Ann Arbor, Michigan, who was not on the panel. Recently, for instance, the agency withdrew the breast cancer indication for Avastin (bevacizumab), given accelerated approval in 2008, after studies found the drug did not provide a survival advantage (Nat. Biotechnol. 29, 3–15, 2011). Emma Dorey

176

kind, the most common side effect is a cough, Pfeffer says. “We see less of that with the new inhaler in the small, early studies that we’ve done so far.” Inhalers also produce a measurable reduction in lung function, though it is reversible and clinically insignificant. That side effect is also reduced with Dreamboat, he says. “It’s kind of intuitive that inhaling less powder can’t help but be a good thing.” If MannKind had gone to market with the earlier MedTone, “Bringing in a new inhaler would be very confusing,” says Pfeffer, explaining why the company decided to swap as soon as the pivotal clinical trials had been completed. Inhaled insulin has a checkered history. First developed in the mid-1990s, New York–based Pfizer’s Exubera was the first inhaled insulin to receive FDA approval in 2006. Original estimates predicted $2 billion in sales, but the inhaler was unpopular with patients. Perhaps the biggest strike against it was the large, awkward delivery device—so ungainly it was nicknamed ‘the bong’. In addition, lingering uncertainty over a putative association between inhaled insulin and lung cancer also compromised patient uptake. The lukewarm reception for the product and poor sales prompted Pfizer to pull the product from the market only two years later, citing lung cancer concerns (Nat. Biotechnol. 26, 479–480, 2008). Soon after, Novo Nordisk of Bagsvaerd, Denmark, also cancelled its phase 3 program for an inhaled insulin (Nat. Biotechnol. 26, 255, 2008). “When you go and look at the original data, I don’t know anybody who buys the argument that there was an increased lung cancer risk. The word on the street was that it wasn’t selling and they needed an excuse to pull out of it,” says Ananth Annapragada, a professor of entrepreneurial biomedical informatics and bioengineering at The University of Texas Health Science Center in Houston. Inhaled insulin has clear advantages: it is simpler than injections and hypoglycemia incidents are sharply reduced, “which is probably the biggest fear of both doctors and patients regarding the use of injected insulin,” says Pfeffer. Skyler also notes Afrezza’s extremely fast action. Peak insulin is achieved at 14 minutes, compared with 49 minutes for Exubera. Rapid-acting insulin analogs like Novo Nordisk’s Novolog, which peaks at 52 minutes, still don’t become available rapidly enough to deal with the spike in glucose during a meal, Skyler says. “[Afrezza] is the first really super-rapidacting insulin. I think this will allow much better control. The second advantage is that the thumb-sized device makes it really easy to use. Those are two very attractive features that should resonate with patients and doctors,”

Skyler says. Afrezza could also ease concerns about potential effects on the lung. “This is so rapidly absorbed that there’s little exposure in the lung. I don’t think [safety] is an issue, but one never knows for sure. I think eventually it [will be] approved,” says Skyler. Although Afrezza is on hold, eyes are turning toward Generex of Toronto. The company’s oral insulin Oral-lyn—an aerosolized, mixedmicelle liquid formulation, comprising recombinant insulin with excipients (alkali metal alkyl sulfate), absorption enhancers, phenol stabilizers and propellant—insulin spray is in phase 3 clinical trials. The product is delivered directly into the mouth and absorbed through the mucosa of the cheeks and the back of the throat. No product enters the lungs, according to the company. Like Afrezza, Oralin is rapidly absorbed, with a dose taken immediately before eating, followed by a second dose after the meal. That profile mimics insulin patterns in nondiabetics, who experience an insulin peak 30 to 60 seconds after beginning to eat, says James Anderson, a professor at Indiana University in Bloomington, and a member of Generex’s advisory board. “When you inject insulin, you don’t get that large spike early on. [With Oral-lyn], you can get better control of the glucose rise following a meal than you can with injected insulin.” Overall, the reviews for MannKind and Generex are mixed. Clinicians still worry that insulin could be linked to lung cancer. “The long-term results of giving inhaled insulin are still not very clear. I think primary-care physicians will be reluctant to use it,” says Joel Zonszein, director of the clinical diabetes center at Montefiore Medical Center, in the Bronx, New York, and a professor of clinical medicine at the Albert Einstein College of Medicine, also in the Bronx. Also, improvements in injected insulin have eroded the need for alternatives. “The needles are not very painful—it’s more painful to check blood sugar level. Patients don’t complain much. If [Afrezza] is approved it will be a niche product but maybe a small niche for individuals who have needle phobia,” says Zonszein. But others still hold out hope that inhaled insulin can transform insulin therapy. “The reality is that superiority in efficacy is pretty much a given because of compliance. With injected insulin, compliance is terrible,” says Annapragada. People are often reluctant to inject themselves in front of others, whereas Afrezza’s Dreamboat is more like an asthma inhaler. “I don’t know any [asthma sufferer] who won’t take a puff in the middle of a meeting,” he adds. Jim Kling, Bellingham, Washington

volume 29 number 3 MARCH 2011 nature biotechnology

NEWS

in brief

© 2011 Nature America, Inc. All rights reserved.

Courts back Prometheus IP In a ruling closely watched by developers of companion diagnostics, the US Court of Appeals for the Federal Circuit recently concluded that two methods for determining the optimal dosage of drugs to treat autoimmune diseases are patentable. The December 17, 2010, ruling reaffirmed the court’s earlier decision in Prometheus Labs. Inc. v. Mayo Collaborative Services. Prometheus Laboratories of San Diego sued the Mayo Clinic for patent infringement when the medical group applied an in-house diagnostic test instead of sending samples to Prometheus. The Mayo Clinic claimed the process of giving a drug, observing its effects and adjusting the dosage is an abstract idea that was around before Prometheus patented the test. But the Federal Circuit upheld the patent. Then soon after the Supreme Court’s Bilski v. Kappos decision (Nat. Biotechnol. 28, 767, 2010), in which the court determined that the ‘machine-or-transformation’ test was only one of the considerations for an invention’s patentability, it vacated the Federal Circuit’s ruling and ordered the court to issue a new one. Prometheus argued the Bilski decision did not merit a reversal, as the tests “involve a particular transformation of a patient’s body and bodily sample and use particular machines to determine metabolite concentrations in a bodily sample.” The court came back with the same decision—good news for companies wanting to develop and patent companion diagnostic tests. Michael Francisco

Accelerated approvals examined The US Food and Drug Administration (FDA) is seeking to improve the much-criticized accelerated approval program by reviewing six drugs approved under this pathway. The agency’s Oncologic Drugs Advisory Committee (ODAC) held a meeting on February 8 to scrutinize Eli Lilly’s Erbitux (cetuximab), GlaxoSmithKline’s Bexxar (tositumomab) and Arranon (nelarabine); Genzyme’s Clolar (clofarabine), Amgen’s Vectibix (panitumumab) and Novartis’ Gleevec (imatinib). The committee’s intention was to analyze the process that brought these drugs to market without full confirmation that they are safe and effective. ODAC concluded that to grant accelerated approval, the agency should require a randomized trial, which could measure a surrogate endpoint. The panel also proposed that at the time of gaining accelerated approval, two randomized controlled trials should be under way. “The real issue is that lots of drugs are approved that are not terribly efficacious,” argues Laurence Baker, chairman of the Southwest Oncology Group, Ann Arbor, Michigan, who was not on the panel. Recently, for instance, the agency withdrew the breast cancer indication for Avastin (bevacizumab), given accelerated approval in 2008, after studies found the drug did not provide a survival advantage (Nat. Biotechnol. 29, 3–15, 2011). Emma Dorey

176

kind, the most common side effect is a cough, Pfeffer says. “We see less of that with the new inhaler in the small, early studies that we’ve done so far.” Inhalers also produce a measurable reduction in lung function, though it is reversible and clinically insignificant. That side effect is also reduced with Dreamboat, he says. “It’s kind of intuitive that inhaling less powder can’t help but be a good thing.” If MannKind had gone to market with the earlier MedTone, “Bringing in a new inhaler would be very confusing,” says Pfeffer, explaining why the company decided to swap as soon as the pivotal clinical trials had been completed. Inhaled insulin has a checkered history. First developed in the mid-1990s, New York–based Pfizer’s Exubera was the first inhaled insulin to receive FDA approval in 2006. Original estimates predicted $2 billion in sales, but the inhaler was unpopular with patients. Perhaps the biggest strike against it was the large, awkward delivery device—so ungainly it was nicknamed ‘the bong’. In addition, lingering uncertainty over a putative association between inhaled insulin and lung cancer also compromised patient uptake. The lukewarm reception for the product and poor sales prompted Pfizer to pull the product from the market only two years later, citing lung cancer concerns (Nat. Biotechnol. 26, 479–480, 2008). Soon after, Novo Nordisk of Bagsvaerd, Denmark, also cancelled its phase 3 program for an inhaled insulin (Nat. Biotechnol. 26, 255, 2008). “When you go and look at the original data, I don’t know anybody who buys the argument that there was an increased lung cancer risk. The word on the street was that it wasn’t selling and they needed an excuse to pull out of it,” says Ananth Annapragada, a professor of entrepreneurial biomedical informatics and bioengineering at The University of Texas Health Science Center in Houston. Inhaled insulin has clear advantages: it is simpler than injections and hypoglycemia incidents are sharply reduced, “which is probably the biggest fear of both doctors and patients regarding the use of injected insulin,” says Pfeffer. Skyler also notes Afrezza’s extremely fast action. Peak insulin is achieved at 14 minutes, compared with 49 minutes for Exubera. Rapid-acting insulin analogs like Novo Nordisk’s Novolog, which peaks at 52 minutes, still don’t become available rapidly enough to deal with the spike in glucose during a meal, Skyler says. “[Afrezza] is the first really super-rapidacting insulin. I think this will allow much better control. The second advantage is that the thumb-sized device makes it really easy to use. Those are two very attractive features that should resonate with patients and doctors,”

Skyler says. Afrezza could also ease concerns about potential effects on the lung. “This is so rapidly absorbed that there’s little exposure in the lung. I don’t think [safety] is an issue, but one never knows for sure. I think eventually it [will be] approved,” says Skyler. Although Afrezza is on hold, eyes are turning toward Generex of Toronto. The company’s oral insulin Oral-lyn—an aerosolized, mixedmicelle liquid formulation, comprising recombinant insulin with excipients (alkali metal alkyl sulfate), absorption enhancers, phenol stabilizers and propellant—insulin spray is in phase 3 clinical trials. The product is delivered directly into the mouth and absorbed through the mucosa of the cheeks and the back of the throat. No product enters the lungs, according to the company. Like Afrezza, Oralin is rapidly absorbed, with a dose taken immediately before eating, followed by a second dose after the meal. That profile mimics insulin patterns in nondiabetics, who experience an insulin peak 30 to 60 seconds after beginning to eat, says James Anderson, a professor at Indiana University in Bloomington, and a member of Generex’s advisory board. “When you inject insulin, you don’t get that large spike early on. [With Oral-lyn], you can get better control of the glucose rise following a meal than you can with injected insulin.” Overall, the reviews for MannKind and Generex are mixed. Clinicians still worry that insulin could be linked to lung cancer. “The long-term results of giving inhaled insulin are still not very clear. I think primary-care physicians will be reluctant to use it,” says Joel Zonszein, director of the clinical diabetes center at Montefiore Medical Center, in the Bronx, New York, and a professor of clinical medicine at the Albert Einstein College of Medicine, also in the Bronx. Also, improvements in injected insulin have eroded the need for alternatives. “The needles are not very painful—it’s more painful to check blood sugar level. Patients don’t complain much. If [Afrezza] is approved it will be a niche product but maybe a small niche for individuals who have needle phobia,” says Zonszein. But others still hold out hope that inhaled insulin can transform insulin therapy. “The reality is that superiority in efficacy is pretty much a given because of compliance. With injected insulin, compliance is terrible,” says Annapragada. People are often reluctant to inject themselves in front of others, whereas Afrezza’s Dreamboat is more like an asthma inhaler. “I don’t know any [asthma sufferer] who won’t take a puff in the middle of a meeting,” he adds. Jim Kling, Bellingham, Washington

volume 29 number 3 MARCH 2011 nature biotechnology

NEWS

in brief

© 2011 Nature America, Inc. All rights reserved.

Courts back Prometheus IP In a ruling closely watched by developers of companion diagnostics, the US Court of Appeals for the Federal Circuit recently concluded that two methods for determining the optimal dosage of drugs to treat autoimmune diseases are patentable. The December 17, 2010, ruling reaffirmed the court’s earlier decision in Prometheus Labs. Inc. v. Mayo Collaborative Services. Prometheus Laboratories of San Diego sued the Mayo Clinic for patent infringement when the medical group applied an in-house diagnostic test instead of sending samples to Prometheus. The Mayo Clinic claimed the process of giving a drug, observing its effects and adjusting the dosage is an abstract idea that was around before Prometheus patented the test. But the Federal Circuit upheld the patent. Then soon after the Supreme Court’s Bilski v. Kappos decision (Nat. Biotechnol. 28, 767, 2010), in which the court determined that the ‘machine-or-transformation’ test was only one of the considerations for an invention’s patentability, it vacated the Federal Circuit’s ruling and ordered the court to issue a new one. Prometheus argued the Bilski decision did not merit a reversal, as the tests “involve a particular transformation of a patient’s body and bodily sample and use particular machines to determine metabolite concentrations in a bodily sample.” The court came back with the same decision—good news for companies wanting to develop and patent companion diagnostic tests. Michael Francisco

Accelerated approvals examined The US Food and Drug Administration (FDA) is seeking to improve the much-criticized accelerated approval program by reviewing six drugs approved under this pathway. The agency’s Oncologic Drugs Advisory Committee (ODAC) held a meeting on February 8 to scrutinize Eli Lilly’s Erbitux (cetuximab), GlaxoSmithKline’s Bexxar (tositumomab) and Arranon (nelarabine); Genzyme’s Clolar (clofarabine), Amgen’s Vectibix (panitumumab) and Novartis’ Gleevec (imatinib). The committee’s intention was to analyze the process that brought these drugs to market without full confirmation that they are safe and effective. ODAC concluded that to grant accelerated approval, the agency should require a randomized trial, which could measure a surrogate endpoint. The panel also proposed that at the time of gaining accelerated approval, two randomized controlled trials should be under way. “The real issue is that lots of drugs are approved that are not terribly efficacious,” argues Laurence Baker, chairman of the Southwest Oncology Group, Ann Arbor, Michigan, who was not on the panel. Recently, for instance, the agency withdrew the breast cancer indication for Avastin (bevacizumab), given accelerated approval in 2008, after studies found the drug did not provide a survival advantage (Nat. Biotechnol. 29, 3–15, 2011). Emma Dorey

176

kind, the most common side effect is a cough, Pfeffer says. “We see less of that with the new inhaler in the small, early studies that we’ve done so far.” Inhalers also produce a measurable reduction in lung function, though it is reversible and clinically insignificant. That side effect is also reduced with Dreamboat, he says. “It’s kind of intuitive that inhaling less powder can’t help but be a good thing.” If MannKind had gone to market with the earlier MedTone, “Bringing in a new inhaler would be very confusing,” says Pfeffer, explaining why the company decided to swap as soon as the pivotal clinical trials had been completed. Inhaled insulin has a checkered history. First developed in the mid-1990s, New York–based Pfizer’s Exubera was the first inhaled insulin to receive FDA approval in 2006. Original estimates predicted $2 billion in sales, but the inhaler was unpopular with patients. Perhaps the biggest strike against it was the large, awkward delivery device—so ungainly it was nicknamed ‘the bong’. In addition, lingering uncertainty over a putative association between inhaled insulin and lung cancer also compromised patient uptake. The lukewarm reception for the product and poor sales prompted Pfizer to pull the product from the market only two years later, citing lung cancer concerns (Nat. Biotechnol. 26, 479–480, 2008). Soon after, Novo Nordisk of Bagsvaerd, Denmark, also cancelled its phase 3 program for an inhaled insulin (Nat. Biotechnol. 26, 255, 2008). “When you go and look at the original data, I don’t know anybody who buys the argument that there was an increased lung cancer risk. The word on the street was that it wasn’t selling and they needed an excuse to pull out of it,” says Ananth Annapragada, a professor of entrepreneurial biomedical informatics and bioengineering at The University of Texas Health Science Center in Houston. Inhaled insulin has clear advantages: it is simpler than injections and hypoglycemia incidents are sharply reduced, “which is probably the biggest fear of both doctors and patients regarding the use of injected insulin,” says Pfeffer. Skyler also notes Afrezza’s extremely fast action. Peak insulin is achieved at 14 minutes, compared with 49 minutes for Exubera. Rapid-acting insulin analogs like Novo Nordisk’s Novolog, which peaks at 52 minutes, still don’t become available rapidly enough to deal with the spike in glucose during a meal, Skyler says. “[Afrezza] is the first really super-rapidacting insulin. I think this will allow much better control. The second advantage is that the thumb-sized device makes it really easy to use. Those are two very attractive features that should resonate with patients and doctors,”

Skyler says. Afrezza could also ease concerns about potential effects on the lung. “This is so rapidly absorbed that there’s little exposure in the lung. I don’t think [safety] is an issue, but one never knows for sure. I think eventually it [will be] approved,” says Skyler. Although Afrezza is on hold, eyes are turning toward Generex of Toronto. The company’s oral insulin Oral-lyn—an aerosolized, mixedmicelle liquid formulation, comprising recombinant insulin with excipients (alkali metal alkyl sulfate), absorption enhancers, phenol stabilizers and propellant—insulin spray is in phase 3 clinical trials. The product is delivered directly into the mouth and absorbed through the mucosa of the cheeks and the back of the throat. No product enters the lungs, according to the company. Like Afrezza, Oralin is rapidly absorbed, with a dose taken immediately before eating, followed by a second dose after the meal. That profile mimics insulin patterns in nondiabetics, who experience an insulin peak 30 to 60 seconds after beginning to eat, says James Anderson, a professor at Indiana University in Bloomington, and a member of Generex’s advisory board. “When you inject insulin, you don’t get that large spike early on. [With Oral-lyn], you can get better control of the glucose rise following a meal than you can with injected insulin.” Overall, the reviews for MannKind and Generex are mixed. Clinicians still worry that insulin could be linked to lung cancer. “The long-term results of giving inhaled insulin are still not very clear. I think primary-care physicians will be reluctant to use it,” says Joel Zonszein, director of the clinical diabetes center at Montefiore Medical Center, in the Bronx, New York, and a professor of clinical medicine at the Albert Einstein College of Medicine, also in the Bronx. Also, improvements in injected insulin have eroded the need for alternatives. “The needles are not very painful—it’s more painful to check blood sugar level. Patients don’t complain much. If [Afrezza] is approved it will be a niche product but maybe a small niche for individuals who have needle phobia,” says Zonszein. But others still hold out hope that inhaled insulin can transform insulin therapy. “The reality is that superiority in efficacy is pretty much a given because of compliance. With injected insulin, compliance is terrible,” says Annapragada. People are often reluctant to inject themselves in front of others, whereas Afrezza’s Dreamboat is more like an asthma inhaler. “I don’t know any [asthma sufferer] who won’t take a puff in the middle of a meeting,” he adds. Jim Kling, Bellingham, Washington

volume 29 number 3 MARCH 2011 nature biotechnology

news

Sandoz, the generic drugs unit of Basel’s 2014, and US coverage in 2015. The second has to do with new guidelines Novartis, announced in January that it has begun a phase 2 rheumatoid arthritis trial with for biosimilar mAbs released by European its own version of blockbuster monoclonal regulators last November (Nat. Biotechnol. antibody (mAb) Rituxan (MabThera, ritux- 29, 10, 2011). To gain approval, the demands imab). It joins Teva Pharmaceuticals, of Petach placed on biosimilar manufacturers are Tikva, Israel, and Spectrum Pharmaceuticals, less onerous than anticipated. The requireof Irvine, California, both of whom are also ments for approval outlined by the European Medicines agency working on verinclude pharmasions of the anticodynamic and CD20 chimeric pharmacokinetic mAb approved for studies, an equivchronic lymphoalency margin of cytic leukemia, 80% to 125%, a non-Ho dg k in’s full comparative lymphoma and clinical trial or rheumatoid an interim endarthritis. The point for approval progress of these followed by a biosimilar vertraditional endsions of Rituxan point for postwill be closely approval as well monitored by bio- Biosimilars producers are keen to bite into Rituxan’s as safety data tech innovators, $6.6 billion global market. gained through particularly the rapidity with which they proceed through the sufficient patient exposure. In essence, bioreview process. At $6.6 billion in 2010 sales, similars makers need to show only similarity Rituxan is the largest revenue-producing bio- and conduct clinical trials with only a small logic yet to come into the crosshairs of bio- number of patients. That has lowered the bar further than some biosimilars developers had similar developers. Besides the allure of its billion-dollar mar- dared hope. The decisiveness in Europe has made the ket, Rituxan has been prioritized by biosimilar manufacturers for two simple reasons. US’s inability to agree on its own pathway more The first is intellectual property. The drug, conspicuous. (Europe already has cleared for marketed in Europe by Roche of Basel, and marketing several recombinant small proteins in the US by Biogen Idec, of Cambridge, as biosimilars, such as growth hormones and Massachusetts, and Genentech, of South San insulin.) US Congress gave the Food and Drug Francisco, was first approved in 1997, and Administration (FDA) authority to approve is due to lose European patent protection in biosimilars for the US market as part of the Genentech

© 2011 Nature America, Inc. All rights reserved.

Biosimilars encircle Rituxan, US debates innovator exclusivity Obama administration’s Patient Protection and Affordable Care Act passed a year ago. But it’s unclear what the FDA will do. Some observers predict the agency could take a few years to draft its plan. The possibilities run the gamut from replicating the EU model to adopting more stringent guidelines that require larger, separate trials with efficacy objectives. With several biosimilars already approved in Europe (Table 1), there will be “pressure” on the agency, says Rajesh Shrotriya, Spectrum’s chairman, CEO and president “to provide guidance to individual companies while still developing guidance documents on biosimilars.” Some of that pressure is already here, as major stakeholders are voicing their varied concerns with the FDA. Robert K. Coughlin, the Massachusetts Biotechnology Council’s president and CEO, stresses that a follow-on biologic pathway in the US “must rely strongly on robust clinical testing to ensure patient safety and efficacy.” The Washington, DC–based Biotechnology Industry Organization’s 34-page commentary issued late in December urges the FDA to take a class-by-class approach to determine the scope of clinical trials. It supports a biosimilarity objective, allowing lower patient enrollments, but says that safety requirements may ultimately dictate a trial’s size. On the flip side, the Generic Pharmaceuticals Association, also of Washington, DC, has argued that it is appropriate to rely on a reference product’s safety and efficacy profile if a biosimilar meets similarity standards. Perhaps the most contentious issue is exclusivity. The Patient Protection and Affordable Care Act not only gives the FDA authority to

Table 1 Biosimilar drugs approved by the EU Biosimilar

Reference product (originator company)

Sponsor

Abseamed (epoetin alfa)

Eprex (Janssen-Cilag, Saunderton, UK)

Medice Arzneimittel Pütter (Iserlolm, Germany) August 2007

Date of approval

Biograstim (filgrastim; G-CSF)

Neupogen (Amgen)

CT Arzneimittel

September 2008

Binocrit (epoetin alfa)

Eprex (Janssen-Cilag)

Sandoz (unit of Novartis)

August 2007

Epoetin alfa Hexal (epoetin alfa)

Eprex/Erypo (Janssen-Cilag)

Hexal Biotech (owned by Novartis)

August 2007

Filgrastim Hexal (filgrastim; G-CSF)

Neupogen (Amgen)

Hexal Biotech

February 2009

Nivestim (filgrastim; G-CSF)

Neupogen (Amgen)

Hospira Enterprises

June 2010

Omnitrope (somatropin; human growth hormone)

Genotropin (Pfizer, New York)

Sandoz

April 2006

Ratiograstim (filgrastim; G-CSF) and Filgrastim Ratiopharm

Neupogen (Amgen)

Ratiopharm (acquired by Teva)

September 2008

Retacrit (epoetin zeta)

Eprex (Janssen-Cilag)

Hospira

December 2007

Silapo (epoetin zeta)

Eprex (Janssen-Cilag)

STADA Arzneimittel (Bad Vilbel, Germany)

December 2007

Tevagrastim (filgrastim; G-CSF)

Neupogen (Amgen)

Teva Pharmaceuticals

September 2008

Valtropin (somatropin; human growth hormone)

Humatrope (Eli Lilly, Indianapolis)

BioPartners (Baar, Switzerland)

April 2006

Zarzio (filgrastim; G-CSF)

Neupogen (Amgen)

Sandoz

February 2009

Source: Company websites. G-CSF, granulocyte colony-stimulating factor.

nature biotechnology volume 29 number 3 MARCH 2011

177

NEWS

in brief

© 2011 Nature America, Inc. All rights reserved.

NCI revamps trials The National Cancer Institute (NCI) is restructuring its long-established clinical trials program to take advantage of new understanding in molecular oncology and improvements in clinical trial design. The NCI’s clinical trial Cooperative Group program’s nine groups will be consolidated into four entities. “As we start defining illness based on molecular or genetic signatures, we start homing into more specific patient populations, which require screening for larger populations,” says Jan Buckner, professor of oncology at Mayo Clinic in Rochester, Minnesota, and the chair of the North Central Cancer Treatment Group. The NCI’s Cooperative Group program was founded over 50 years ago and involves more than 3,100 institutions. The organizational changes follow a NCI-requested report released last April by the Institute of Medicine (IOM), of Washington, DC. Efficiency will be boosted by revamping informational technology infrastructure, outfitting all groups with a uniform information system and seamless sharing of information, sample banks and databases. One of the major goals is to speed up the time taken to approve and initiate phase 2 and phase 3 clinical trials. “We desperately want to get new treatments out to cancer patients, and do this in the most expeditious and safe way possible,” says James Doroshow, director of the Division of Cancer Treatment and Diagnosis at NCI in Bethesda, Maryland. Nidhi Subbaraman

Yardsticks for R&D Two nonprofits—the Critical Path Institute (C-Path), based in Tucson, and the Clinical Data Interchange Standards Consortium (C-DISC) of Round Rock, Texas—are teaming up to set common standards for companies to report clinical data on diseases considered major public health challenges. The aim is to quicken R&D efforts and potentially facilitate the evaluation of new therapies at the US Food and Drug Administration (FDA). “Most companies are recognizing greater efficiency when we all call an apple an apple,” says Raymond Woosley, C-Path’s president and CEO. The data standards are intended as useful guidelines rather than mandates. C-Path and C-DISC built a database for Alzheimer’s disease, launched in June 2010, as part of C-Path’s Coalition Against Major Diseases project, and data from 4,000 patients have now been mapped to the standard. The joint effort will now be expanded to include data on amyotrophic lateral sclerosis, Huntington’s disease, multiple sclerosis, lung cancer and diabetes. Standardized data would allow regulators to compare clinical data results across trials and across companies. ShaAvhree Buckman, Director of the Office of Translational Sciences at the FDA’s Center for Drug Evaluation and Research (CDER) welcomes these data standards, as they capitalize on work already set in motion by existing groups. Nidhi Subbaraman

178

approve biosimilars, but states that a developer must wait 4 years after a brand product is approved before filing an application. It also says a developer of a follow-on biologic must wait 12 years before it can receive approval for a drug made relying on innovator data. Also, an innovator can receive an additional 12 years of exclusivity for a modified product that produces changes in safety, purity or potency. That has sparked a debate over the meaning of exclusivity and has cornered FDA Commissioner Margaret Hamburg. Senators Kay Hagan (D-NC), Orrin Hatch (R-UT), Michael Enzi (R-WY) and John Kerry (D-MA) sent her a letter on January 7 saying the law provides for data exclusivity. This means that a biosimilar developer that relies on its own data need not wait 12 years to file a biologics license application. Thirteen follow-on biologic supporters, including the American Association of Retired Persons of Washington, DC, health insurance company Aetna of Hartford, Connecticut, specialty pharma Hospira of Lake Forest, Illinois, and Teva, also wrote to Hamburg in a January 20 letter suggesting that if data exclusivity expires after 4 years, it clears the way for developers to file applications relying on innovator data, even though the approval cannot come until the marketing exclusivity (the full 12 years) ends. Clarification finally came from sponsors of the law, Representatives Anna G. Eshoo (D-CA), Jan Inslee (D-WA) and Joe Barton (R-TX). The intent is to give companies 12 years of data exclusivity—not market exclusivity. That means biosimilar companies may not rely on innovator data but could still develop their own data for a similar product that could be marketed alongside the original. The bill does prohibit evergreening (a process whereby innovator companies make trivial or minor improvements to a drug in an effort to extend patent life), although brand manufacturers that launch next-generation products could gain their own exclusivity period, if they can be sufficiently differentiated from the originator molecule. The money at stake is staggering. New York–based consultants IMS Health’s most recent data show that biologic drugs generated $130 billion worldwide in 2009. If the US follow-on biologic pathway were in place, several large biotechs would be facing a patent cliff similar to what big pharma is currently facing. For instance, 74% of Thousand Oaks, California–based Amgen’s 2010 revenue (~$11 billion) and 57% of Boston-based Genzyme’s 2010 revenue (~$2.3 billion) would be exposed through patent expiry by 2015.

With Rituxan’s patents facing expiration, the drug has become a prime target for biosimilar makers. Teva has launched clinical trials in both rheumatoid arthritis and non-Hodgkin’s lymphoma with its biosimilar, TL011. The rheumatoid arthritis trial is enrolling 60 patients and will be completed by August. Two Irvine, California–based companies, Spectrum and Viropro, are collaborating to produce another version of Rituxan. Dr. Reddy’s Laboratories, of Hyderabad, India, already launched its own copy, Reditux, in India in 2007. How much it costs to develop a biosimilar is hard to pin down. Estimates range from $40 million on up to $250 million and far beyond, depending on the complexity of the molecule, and the premarket work can take up to seven years. Still, if a biosimilar steals a conservative 20% share of a $1 billion product, at the 20% price discount that some analysts predict, it reaps $160 million in annual revenue. But despite the rewards, the business of developing new versions of brand biologics still has much higher barriers to entry than chemical generics, even assuming that the EU pathway works smoothly and the FDA finally hammers out a guidance. Companies hoping to get into the follow-on biologic game will need ample resources to comply with manufacturing requirements, run clinical trials and go to court over biotech patents. Even cleared drugs may bump against patient prejudice and physicians reluctant to replace a proven therapy with a cheaper one. Sandoz’s copy of Rituxan was developed at its facilities in Schaftenau, Austria. The company spent years using different combinations of manufacturing processes and media components with the same gene sequence to identify its biosimilar. Intensive characterization followed using different methods to ensure its version fit within the normal variability of the original product. “It’s really, really tough to do this,” Sandoz’s global head of biopharmaceuticals, Ameet Mallik, says. “You basically need innovator capabilities.” Yet the list of interested parties grows. Teva signed a deal two years ago with Basel’s Lonza Group to develop biosimilars, and Whitehouse Station, New Jersey–based Merck BioVentures spent $130 million setting up its biosimilar capabilities through a buyout of Richmond, Virginia–based Insmed (Nat. Biotechnol. 27, 299–301, 2009). Even those at risk of losing share to biosimilars are looking to get into the game: Amgen and Biogen Idec both expressed interest in the space at the recent JP Morgan healthcare conference. Karen Carey, York, Pennsylvania

volume 29 number 3 MARCH 2011 nature biotechnology

NEWS

in brief

© 2011 Nature America, Inc. All rights reserved.

NCI revamps trials The National Cancer Institute (NCI) is restructuring its long-established clinical trials program to take advantage of new understanding in molecular oncology and improvements in clinical trial design. The NCI’s clinical trial Cooperative Group program’s nine groups will be consolidated into four entities. “As we start defining illness based on molecular or genetic signatures, we start homing into more specific patient populations, which require screening for larger populations,” says Jan Buckner, professor of oncology at Mayo Clinic in Rochester, Minnesota, and the chair of the North Central Cancer Treatment Group. The NCI’s Cooperative Group program was founded over 50 years ago and involves more than 3,100 institutions. The organizational changes follow a NCI-requested report released last April by the Institute of Medicine (IOM), of Washington, DC. Efficiency will be boosted by revamping informational technology infrastructure, outfitting all groups with a uniform information system and seamless sharing of information, sample banks and databases. One of the major goals is to speed up the time taken to approve and initiate phase 2 and phase 3 clinical trials. “We desperately want to get new treatments out to cancer patients, and do this in the most expeditious and safe way possible,” says James Doroshow, director of the Division of Cancer Treatment and Diagnosis at NCI in Bethesda, Maryland. Nidhi Subbaraman

Yardsticks for R&D Two nonprofits—the Critical Path Institute (C-Path), based in Tucson, and the Clinical Data Interchange Standards Consortium (C-DISC) of Round Rock, Texas—are teaming up to set common standards for companies to report clinical data on diseases considered major public health challenges. The aim is to quicken R&D efforts and potentially facilitate the evaluation of new therapies at the US Food and Drug Administration (FDA). “Most companies are recognizing greater efficiency when we all call an apple an apple,” says Raymond Woosley, C-Path’s president and CEO. The data standards are intended as useful guidelines rather than mandates. C-Path and C-DISC built a database for Alzheimer’s disease, launched in June 2010, as part of C-Path’s Coalition Against Major Diseases project, and data from 4,000 patients have now been mapped to the standard. The joint effort will now be expanded to include data on amyotrophic lateral sclerosis, Huntington’s disease, multiple sclerosis, lung cancer and diabetes. Standardized data would allow regulators to compare clinical data results across trials and across companies. ShaAvhree Buckman, Director of the Office of Translational Sciences at the FDA’s Center for Drug Evaluation and Research (CDER) welcomes these data standards, as they capitalize on work already set in motion by existing groups. Nidhi Subbaraman

178

approve biosimilars, but states that a developer must wait 4 years after a brand product is approved before filing an application. It also says a developer of a follow-on biologic must wait 12 years before it can receive approval for a drug made relying on innovator data. Also, an innovator can receive an additional 12 years of exclusivity for a modified product that produces changes in safety, purity or potency. That has sparked a debate over the meaning of exclusivity and has cornered FDA Commissioner Margaret Hamburg. Senators Kay Hagan (D-NC), Orrin Hatch (R-UT), Michael Enzi (R-WY) and John Kerry (D-MA) sent her a letter on January 7 saying the law provides for data exclusivity. This means that a biosimilar developer that relies on its own data need not wait 12 years to file a biologics license application. Thirteen follow-on biologic supporters, including the American Association of Retired Persons of Washington, DC, health insurance company Aetna of Hartford, Connecticut, specialty pharma Hospira of Lake Forest, Illinois, and Teva, also wrote to Hamburg in a January 20 letter suggesting that if data exclusivity expires after 4 years, it clears the way for developers to file applications relying on innovator data, even though the approval cannot come until the marketing exclusivity (the full 12 years) ends. Clarification finally came from sponsors of the law, Representatives Anna G. Eshoo (D-CA), Jan Inslee (D-WA) and Joe Barton (R-TX). The intent is to give companies 12 years of data exclusivity—not market exclusivity. That means biosimilar companies may not rely on innovator data but could still develop their own data for a similar product that could be marketed alongside the original. The bill does prohibit evergreening (a process whereby innovator companies make trivial or minor improvements to a drug in an effort to extend patent life), although brand manufacturers that launch next-generation products could gain their own exclusivity period, if they can be sufficiently differentiated from the originator molecule. The money at stake is staggering. New York–based consultants IMS Health’s most recent data show that biologic drugs generated $130 billion worldwide in 2009. If the US follow-on biologic pathway were in place, several large biotechs would be facing a patent cliff similar to what big pharma is currently facing. For instance, 74% of Thousand Oaks, California–based Amgen’s 2010 revenue (~$11 billion) and 57% of Boston-based Genzyme’s 2010 revenue (~$2.3 billion) would be exposed through patent expiry by 2015.

With Rituxan’s patents facing expiration, the drug has become a prime target for biosimilar makers. Teva has launched clinical trials in both rheumatoid arthritis and non-Hodgkin’s lymphoma with its biosimilar, TL011. The rheumatoid arthritis trial is enrolling 60 patients and will be completed by August. Two Irvine, California–based companies, Spectrum and Viropro, are collaborating to produce another version of Rituxan. Dr. Reddy’s Laboratories, of Hyderabad, India, already launched its own copy, Reditux, in India in 2007. How much it costs to develop a biosimilar is hard to pin down. Estimates range from $40 million on up to $250 million and far beyond, depending on the complexity of the molecule, and the premarket work can take up to seven years. Still, if a biosimilar steals a conservative 20% share of a $1 billion product, at the 20% price discount that some analysts predict, it reaps $160 million in annual revenue. But despite the rewards, the business of developing new versions of brand biologics still has much higher barriers to entry than chemical generics, even assuming that the EU pathway works smoothly and the FDA finally hammers out a guidance. Companies hoping to get into the follow-on biologic game will need ample resources to comply with manufacturing requirements, run clinical trials and go to court over biotech patents. Even cleared drugs may bump against patient prejudice and physicians reluctant to replace a proven therapy with a cheaper one. Sandoz’s copy of Rituxan was developed at its facilities in Schaftenau, Austria. The company spent years using different combinations of manufacturing processes and media components with the same gene sequence to identify its biosimilar. Intensive characterization followed using different methods to ensure its version fit within the normal variability of the original product. “It’s really, really tough to do this,” Sandoz’s global head of biopharmaceuticals, Ameet Mallik, says. “You basically need innovator capabilities.” Yet the list of interested parties grows. Teva signed a deal two years ago with Basel’s Lonza Group to develop biosimilars, and Whitehouse Station, New Jersey–based Merck BioVentures spent $130 million setting up its biosimilar capabilities through a buyout of Richmond, Virginia–based Insmed (Nat. Biotechnol. 27, 299–301, 2009). Even those at risk of losing share to biosimilars are looking to get into the game: Amgen and Biogen Idec both expressed interest in the space at the recent JP Morgan healthcare conference. Karen Carey, York, Pennsylvania

volume 29 number 3 MARCH 2011 nature biotechnology

NEWS

in brief

© 2011 Nature America, Inc. All rights reserved.

NCI revamps trials The National Cancer Institute (NCI) is restructuring its long-established clinical trials program to take advantage of new understanding in molecular oncology and improvements in clinical trial design. The NCI’s clinical trial Cooperative Group program’s nine groups will be consolidated into four entities. “As we start defining illness based on molecular or genetic signatures, we start homing into more specific patient populations, which require screening for larger populations,” says Jan Buckner, professor of oncology at Mayo Clinic in Rochester, Minnesota, and the chair of the North Central Cancer Treatment Group. The NCI’s Cooperative Group program was founded over 50 years ago and involves more than 3,100 institutions. The organizational changes follow a NCI-requested report released last April by the Institute of Medicine (IOM), of Washington, DC. Efficiency will be boosted by revamping informational technology infrastructure, outfitting all groups with a uniform information system and seamless sharing of information, sample banks and databases. One of the major goals is to speed up the time taken to approve and initiate phase 2 and phase 3 clinical trials. “We desperately want to get new treatments out to cancer patients, and do this in the most expeditious and safe way possible,” says James Doroshow, director of the Division of Cancer Treatment and Diagnosis at NCI in Bethesda, Maryland. Nidhi Subbaraman

Yardsticks for R&D Two nonprofits—the Critical Path Institute (C-Path), based in Tucson, and the Clinical Data Interchange Standards Consortium (C-DISC) of Round Rock, Texas—are teaming up to set common standards for companies to report clinical data on diseases considered major public health challenges. The aim is to quicken R&D efforts and potentially facilitate the evaluation of new therapies at the US Food and Drug Administration (FDA). “Most companies are recognizing greater efficiency when we all call an apple an apple,” says Raymond Woosley, C-Path’s president and CEO. The data standards are intended as useful guidelines rather than mandates. C-Path and C-DISC built a database for Alzheimer’s disease, launched in June 2010, as part of C-Path’s Coalition Against Major Diseases project, and data from 4,000 patients have now been mapped to the standard. The joint effort will now be expanded to include data on amyotrophic lateral sclerosis, Huntington’s disease, multiple sclerosis, lung cancer and diabetes. Standardized data would allow regulators to compare clinical data results across trials and across companies. ShaAvhree Buckman, Director of the Office of Translational Sciences at the FDA’s Center for Drug Evaluation and Research (CDER) welcomes these data standards, as they capitalize on work already set in motion by existing groups. Nidhi Subbaraman

178

approve biosimilars, but states that a developer must wait 4 years after a brand product is approved before filing an application. It also says a developer of a follow-on biologic must wait 12 years before it can receive approval for a drug made relying on innovator data. Also, an innovator can receive an additional 12 years of exclusivity for a modified product that produces changes in safety, purity or potency. That has sparked a debate over the meaning of exclusivity and has cornered FDA Commissioner Margaret Hamburg. Senators Kay Hagan (D-NC), Orrin Hatch (R-UT), Michael Enzi (R-WY) and John Kerry (D-MA) sent her a letter on January 7 saying the law provides for data exclusivity. This means that a biosimilar developer that relies on its own data need not wait 12 years to file a biologics license application. Thirteen follow-on biologic supporters, including the American Association of Retired Persons of Washington, DC, health insurance company Aetna of Hartford, Connecticut, specialty pharma Hospira of Lake Forest, Illinois, and Teva, also wrote to Hamburg in a January 20 letter suggesting that if data exclusivity expires after 4 years, it clears the way for developers to file applications relying on innovator data, even though the approval cannot come until the marketing exclusivity (the full 12 years) ends. Clarification finally came from sponsors of the law, Representatives Anna G. Eshoo (D-CA), Jan Inslee (D-WA) and Joe Barton (R-TX). The intent is to give companies 12 years of data exclusivity—not market exclusivity. That means biosimilar companies may not rely on innovator data but could still develop their own data for a similar product that could be marketed alongside the original. The bill does prohibit evergreening (a process whereby innovator companies make trivial or minor improvements to a drug in an effort to extend patent life), although brand manufacturers that launch next-generation products could gain their own exclusivity period, if they can be sufficiently differentiated from the originator molecule. The money at stake is staggering. New York–based consultants IMS Health’s most recent data show that biologic drugs generated $130 billion worldwide in 2009. If the US follow-on biologic pathway were in place, several large biotechs would be facing a patent cliff similar to what big pharma is currently facing. For instance, 74% of Thousand Oaks, California–based Amgen’s 2010 revenue (~$11 billion) and 57% of Boston-based Genzyme’s 2010 revenue (~$2.3 billion) would be exposed through patent expiry by 2015.

With Rituxan’s patents facing expiration, the drug has become a prime target for biosimilar makers. Teva has launched clinical trials in both rheumatoid arthritis and non-Hodgkin’s lymphoma with its biosimilar, TL011. The rheumatoid arthritis trial is enrolling 60 patients and will be completed by August. Two Irvine, California–based companies, Spectrum and Viropro, are collaborating to produce another version of Rituxan. Dr. Reddy’s Laboratories, of Hyderabad, India, already launched its own copy, Reditux, in India in 2007. How much it costs to develop a biosimilar is hard to pin down. Estimates range from $40 million on up to $250 million and far beyond, depending on the complexity of the molecule, and the premarket work can take up to seven years. Still, if a biosimilar steals a conservative 20% share of a $1 billion product, at the 20% price discount that some analysts predict, it reaps $160 million in annual revenue. But despite the rewards, the business of developing new versions of brand biologics still has much higher barriers to entry than chemical generics, even assuming that the EU pathway works smoothly and the FDA finally hammers out a guidance. Companies hoping to get into the follow-on biologic game will need ample resources to comply with manufacturing requirements, run clinical trials and go to court over biotech patents. Even cleared drugs may bump against patient prejudice and physicians reluctant to replace a proven therapy with a cheaper one. Sandoz’s copy of Rituxan was developed at its facilities in Schaftenau, Austria. The company spent years using different combinations of manufacturing processes and media components with the same gene sequence to identify its biosimilar. Intensive characterization followed using different methods to ensure its version fit within the normal variability of the original product. “It’s really, really tough to do this,” Sandoz’s global head of biopharmaceuticals, Ameet Mallik, says. “You basically need innovator capabilities.” Yet the list of interested parties grows. Teva signed a deal two years ago with Basel’s Lonza Group to develop biosimilars, and Whitehouse Station, New Jersey–based Merck BioVentures spent $130 million setting up its biosimilar capabilities through a buyout of Richmond, Virginia–based Insmed (Nat. Biotechnol. 27, 299–301, 2009). Even those at risk of losing share to biosimilars are looking to get into the game: Amgen and Biogen Idec both expressed interest in the space at the recent JP Morgan healthcare conference. Karen Carey, York, Pennsylvania

volume 29 number 3 MARCH 2011 nature biotechnology

news

US farmers can again plant genetically engi- ing at the heart of the organic community,” says neered alfalfa following a decision in January Doug Gurian-Sherman, a senior scientist at the by the US Department of Agriculture (USDA). Union of Concerned Scientists in Cambridge, The ruling, which follows a tumultuous debate Massachusetts. “The biggest single use for alfalfa and four-year US court-imposed ban, comes as is dairy, and organic milk is a premium product.” a relief to the agricultural biotech industry. The Although there is no validated mechanism in the literature clarifyagency was proposing ing how transgenic to place geographic EPSPS sequences in restrictions on plantalfalfa would make ing in response to their way into cow’s organic growers’ milk, the issue is that requests. This alternaorganic products tive was only narrowly claim to avoid GM averted and could have products in any shape set sweeping regulaor form; thus, transtory precedents. genic alfalfa presents “There was proba problem to organic ably a collective sigh dairy farmers. of relief that the In 2006, a group of agency stuck with the organic alfalfa growprecedent that it has ers and nonprofit been relying on since organizations, such as it started reviewthe Center for Food ing and approving Safety in Washington, biotech traits,” says Planting glyphosate-resistant alfalfa has resumed following the USDA’s January decision. DC, sued the USDA Jeff Rowe, vice presifor approving the GM dent of biotech affairs and regulatory at Pioneer in Des Moines, Iowa. alfalfa, arguing that the agency had not fully But the events that led up to the USDA’s deci- considered its environmental and economic sion have left leaders in industry rattled. They impacts. A US federal court agreed and in 2007 are concerned that the agency will begin making ordered the agency to conduct a more thorough non–science-based concessions to the organic environmental analysis. In the meantime, crop community at the expense of biotech crop planting and sales were halted. USDA worked on the court-ordered envidevelopers and growers. Some expect litigation delays and longer regulatory timelines for crop ronmental impact statement (EIS), for nearly four years. After receiving about 244,000 approvals. Alfalfa is a high-protein forage crop for live- public comments and holding four public stock. On one side of the debate are those seeking meetings, the agency produced a final EIS on to sell and grow the biotech variety, genetically December 16, 2010. The 2,300-page review engineered to tolerate the herbicide glyphosate acknowledged the potential for genes from through expression of the Agrobacterium tume- EPSPS transgenic alfalfa to find their way into faciens transgene 5-enolpyruvylshikimate-3- nontransgenic varieties but noted that the probphosphate synthase (EPSPS) and brought to ability was “low” and depended on several conmarket in 2005 by St. Louis–based Monsanto ditions. USDA maintained its conclusion that and Nampa, Idaho–based Forage Genetics EPSPS transgenic alfalfa is safe for food and feed International. On the other, are those who mar- purposes and poses no plant pest risk. On the basis of the EIS, the agency at first ket organic alfalfa. Leaders of the organic lobby fear that proposed one of two actions: either to approve Monsanto’s alfalfa containing the EPSPS trans- the GM alfalfa fully or approve the crop in part, gene will outcross or admix with their organic with restrictions on where it could be planted. varieties. One of several reasons why consum- For instance, to segregate the transgenic alfalfa ers buy organic products is specifically to avoid from organic alfalfa, farmers would have to set transgenes in their food; thus, the presence (or up an exclusion zone of at least 5 miles. The agency said upon filing the EIS in the ‘contamination,’ as it is commonly branded) of transgenic material in organic food is viewed Federal Register December 23 it would decide as a threat to both the domestic and export after 30 days which of the two actions it would markets of organic producers. “This is strik- follow. “This final EIS is a first step toward lookJason Lugo/istockphoto

© 2011 Nature America, Inc. All rights reserved.

Industry exhales as USDA okays glyphosateresistant alfalfa

nature biotechnology volume 29 number 3 MARCH 2011

in brief DuPont swallows Danisco Early in January, agricultural biotech giant DuPont of Wilmington, Delaware, agreed to purchase Danish enzyme maker Danisco, based in Copenhagen, for $5.8 billion. The deal has not been finalized, but speculation about the potential consequences of this buyout is rippling through the Danish biotech sector. “We’ve sold one of our national treasures,” says Claus Felby, a professor of wood and biomass technology at the University of Copenhagen. Biotech researchers like Birger Moller, professor of plant biochemistry at the University of Copenhagen, fear that if DuPont decides to move Danisco’s manufacturing to the US, this may put an end to an era of fruitful collaboration between industry and basic research in the country. Equally, DuPont’s interest in Danisco could send a message about the value of Danish biotech. “It indicates we’re sitting on a gold mine here,” says Moller. In another recent transaction, Danish enzyme manufacturer Novozymes bought Darmstadt, Germany–based Merck’s bioagricultural science unit for $275 million. Merck’s divested Crop Bioscience, which makes inoculants for plant health, is a strong strategic fit for the Danish biotech located in Bagsvaerd. The companies expect to close the deal by May, pending regulatory approval. Nidhi Subbaraman

Alzheimer’s genetic map Research groups across France, the UK and US are pooling their resources to create the biggest genetic information bank on Alzheimer’s disease. Researchers participating in the International Genomics of Alzheimer’s Project (IGAP) will compare the genomic data of 20,000 individuals with 30,000 controls. Members of the project include the European Alzheimer’s Disease Initiative, led by the Institute Pasteur de Lille and Lille University, the Genetic and Environmental Risk in Alzheimer’s Disease group from Cardiff, UK, the Heart and Aging Research in Genomic Epidemiology, Boston University and the Alzheimer’s Disease Genetics Consortium at the University of Pennsylvania School of Medicine, Philadelphia. “This is the first time, internationally, we’ve all gotten together,” says Gerard Schellenberg, director of the Philadelphia-based team and professor of pathology and laboratory medicine, University of Pennsylvania Medical School. Each institute will carry out its own association analysis, and those statistics pooled into a meta analysis, says Schellenberg. With almost 50,000 individuals, and drawing on results from the 1000 Genome Project, the IGAP aims to deepen understanding of the molecular basis of rare variants of the disease, Schellenberg says, and identify genetic risk factors for the disease. IGAP’s meeting and analysis costs are currently supported by the Alzheimer’s Association of Chicago, and Foundation Plan Alzheimer, of Paris. Nidhi Subbaraman

179

NEWS Table 1 USDA sued for insufficient environmental reviews of GM crops Crop (event name) Developer (location)

USDA approval status

Lawsuit

Outcome

GT alfalfa

Granted: 2005, 2011

USDA was sued in 2006 for failing to fully examine environmental effects of GT alfalfa

USDA completes EIS in Dec 2010

Federal court orders USDA to conduct an EIS, and later halts planting

Activist groups say they will sue again

USDA was sued in 2008 for failing to fully examine environmental effects of GT sugar beets

USDA expects to complete EIS by May 2012

(J101 and J163) Monsanto and Forage Genetics GT sugar beets (H7-1) Monsanto and KWS SAAT AG (Einbeck, Germany) GT creeping bentgrass

Granted: 2005, 2011 (partial)

Federal court orders USDA to conduct an EIS and later halts planting Pending

(ASR368) Monsanto and Scotts Co. (Marysville, Ohio) Freeze-tolerant eucalyptus

USDA was sued in 2003 for allowing field trial planting of GT creeping bentgrass without first properly examining environmental effects

USDA again approves GT alfalfa

USDA in Feb 2011 partially approves GT sugar beets to allow planting while it completes EIS; growers must meet strict planting conditions USDA voluntarily initiates work on an EIS but has yet to complete it

Federal court agrees in part Pending

(FTE 427, FTE 435) ArborGen (Summerville, South Carolina)

USDA was sued in July 2010 for allowing field trial planting of freeze-tolerant eucalyptus without properly examining environmental effects

Case pending

© 2011 Nature America, Inc. All rights reserved.

EIS, environmental impact statement. GT, glyphosate tolerant.

ing at the ways we can achieve effective coexistence between all sectors of agriculture,” USDA Secretary Tom Vilsack commented upon his announcement of the proposal. “It’s a conversation that needs to happen now and we are not going to shy away from having it” (Box 1). The reaction from the biotech industry was immediate. USDA had concluded that the transgenic alfalfa was safe with a low probability of gene transmission to other varieties, yet was proposing to restrict planting because of its potential to harm a group of growers’ economic interests. Such a move, the industry feared, would set a precedent where commercial motives would prevail over science-based decisions from the USDA. “By attempting to use the regulatory process...as a mechanism for achieving broader coexistence between growers, USDA is over-reaching its authority and defying legal precedence and the science that has said this product is safe,” Jim Greenwood, president of Biotechnology Industry Organization (BIO) said in a statement. In a January 5 letter to the White House Office of Science and Technology Policy, farm groups said their international trade efforts would be undermined “if USDA moves forward with injecting non–science-based criteria into the regulatory process.” And a January 18 letter to USDA sent by US Representatives Saxby Chambliss (R-GA), Pat Roberts (R-KS) and Frank Lucas (R-OK) called the proposal “disturbing” because it “politicizes the regulatory process.” In an attempt to foster better dialog between the biotech and organic communities, USDA on December 20 held a closed meeting for stakeholders to discuss the EIS. “There was a nervous tone” at the meeting “from people trying to understand the process and where it was 180

leading,” says Rowe at Pioneer, who attended the meeting. On January 27, USDA announced it would fully deregulate alfalfa without restrictions. But the saga has fomented confusion in the minds of researchers. “I don’t know what to think,” says Bruce Chassy, a food safety professor at the University of Illinois at Urbana-Champaign. “What is mildly encouraging is that after being savaged by the agricultural organizations and congress, [USDA] had the uncommon good sense to drop the issue for now.” In both the EIS and official deregulation document for GM alfalfa, USDA described co-existence as a goal of the agency—a first for that kind of language in a regulatory document. “We have concerns with that type of language,” says Sharon Bomer Lauritsen, an executive vice president at BIO. “Even though we have a nice clean deregulation of alfalfa, there is obviously some interest on the part of the Department to be looking at coexistence issues.” Those issues, BIO has argued, should be addressed by growers and their neighbors, not the regulatory process. Until this recent spat involving alfalfa in the US courts, no transgenic crop registered by the USDA had required an EIS in the approval process. Normally, before it deregulates a biotech crop, the USDA must prepare an “environmental assessment” explaining why a crop may or may not significantly affect the “quality of the human environment.” Only if the impacts are “significant,” must the agency conduct a more detailed EIS, according to statutory requirements under the National Environmental Policy Act (NEPA). In the case of the GM alfalfa, the federal court found USDA’s environmental assessment unconvincing and ordered an EIS. The agency faced a similar legal battle in the US courts with

sugar beets in 2009. After the USDA deregulated glyphosate-tolerant sugar beets the same federal court found the agency’s environmental assessment “cursory” and ordered an EIS. The agency is still working on that document and announced in February that in the interim, it would allow planting of sugar beets transgenic for EPSPS, as long as they are grown under certain conditions. USDA has faced at least two other lawsuits challenging its environmental reviews of biotech crops (Table 1). In an attempt to comply with NEPA, and to avoid further litigation and court-ordered EISs, USDA and crop developers have been devoting more time and resources to improve environmental assessments. Both Monsanto and Pioneer, for example, have been providing the USDA with additional information on their products under review. As a result of that additional work, and to account for potential litigation delays and other factors, Pioneer recently increased by “multiple months,” its estimate for the amount of time it will take to get a crop approved, Rowe says. “There were several negative court decisions against the USDA and based on those catalysts, that certainly has changed our strategy,” Rowe says. “It has made us much more sensitive to NEPA exposure and the environmental assessments that come out of the USDA.” Whether the USDA’s improved environmental assessments will hold up to judicial review has not yet been tested. What is clear is that the industry hasn’t seen the end of litigation over biotech crop approvals. Within hours of USDA’s announced decision to deregulate GM alfalfa, the Center for Food Safety said it would again sue the agency over its alfalfa decision. Emily Waltz, Nashville, Tennessee

volume 29 number 3 MARCH 2011 nature biotechnology

news

© 2011 Nature America, Inc. All rights reserved.

Box 1 Can biotech and organic farming co-exist? Co-existence—the idea that transgenic and non-GM crops of the same species can be successfully grown near each other—has for years been part of the European regulatory discussion. For instance, EU member states have imposed isolation distances on GM crop growers (Nat. Biotechnol. 28, 133–136, 2010). But it is only in the past 18 months or so that the concept has made headlines in the US with legal tussles at the state and federal level between the organic sector and seed companies or farmers growing biotech crops. In December, as a response to the fierce debate following the USDA’s proposals for approving EPSPS transgenic alfalfa, USDA Secretary Tom Vilsack wrote an open letter urging co-existence among biotech, organic and conventional crop stakeholders, suggesting that the agency wants to “forge a new paradigm based on co-existence.” The clash between biotech crops and organic agriculture has “led to litigation and uncertainty. Such litigation will potentially lead to the courts deciding who gets to farm their way and who will be prevented from doing so…. Surely, there is a better way,” Vilsack wrote. Three months ago, the agency proposed to place European-style isolation distances on EPSPS transgenic alfalfa. It later backed down from that proposal and instead announced in January it would implement nonregulatory measures to address co-existence. For instance, the agency plans to revive its advisory committee on biotech and 21st century agriculture (AC21), which in 2008 produced a report on the issues surrounding coexistence. This time, the committee will be charged with guiding the agency on ways to strengthen co-existence. But the agency’s attempt to address the issue through federal regulations has met resistance from the biotech community. “Farmers have always had disputes and farmers have always been able to work things out without getting a great deal of government involvement in the middle of it,” says Alan McHughen, a plant biotechnologist at the University of California at Riverside. Left to the growers, disputes tend to get settled “with a compromise on both sides,” he says. EW

in their words “In defense of the coffee and doughnuts, I would say there are probably some areas you can cut but I am not sure we should be inviting the most knowledgeable scientists in the world to talk about a cure for cancer and not offer them coffee.” Democratic lawmaker Chaka Fattah echoed calls to cut overblown budgets but drew the line at food. (AFP, 11 February 2011) “What disturbs me a little is that a more-than-vivid imagination is needed to think that genealogy research in Iceland is being spied on if there is nothing behind it.” Kari Stefánsson, executive chairman and president of research at deCODE, comments on last year’s WikiLeaks disclosure that Chinese authorities may be spying on the company’s genealogy and medical research database. (Iceland Review_Online, 6 December 2010) “In recent years, big pharma, through a combination of internal research and acquisitions, has accumulated as much biotechnology capabilities as any biotech. What is left to differentiate big pharma from big biotech?” Columnist Jason Chew highlights the blurring line between pharma and biotech. (Seeking Alpha, 20 January 2011) “It took us 30 years to get to where we are. So it will take some time to understand the next step, to understand the dynamics and the value.” Genzyme CEO Henri Termeer comments on takeover talks between the biotech and Sanofi-aventis. (Boston Globe, 20 January 2011) “You have to ask yourself why are physicians involved with these organizations at all? What’s the benefit to society and medicine? In my mind, the answer is little or none.” Jerome Kassirer, Tufts University Medical School professor and former editor-in-chief of the New England Journal of Medicine, supports a proposal from over a dozen medical journals to require authors of submissions to disclose payments from hedge funds or other investors. (Bloomberg, 19 January 2011) “As a politician, I tend to listen to the emotions of the people. And yet as an engineer, I also have to listen to the scientific results.” Emmanuel Pinol, vice governor of North Cotabato, Philippines, after the local government uprooted Bt eggplant field trials. (Crop Biotech Update, 14 January 2011). “We need new tools. Nothing we’ve done in the past 40 years has had an impact.” Duane Gubler, professor in infectious diseases at Singapore’s DukeNUS Graduate Medical School, comments on the release of about 6,000 sterile Oxitec OX513A male mosquitoes into an uninhabited forest area in eastern Malaysia, part of a trial aimed at controlling dengue fever. (Associated Press, 26 January 2011).

Video games played with live organisms ‘Biotic games’ that mimic classic video games have been devised by Ingmar Riedel-Kruse and his team at Stanford. Single-celled organisms are placed in a microfluidics chamber with a microscope camera to track their movements. The image is overlaid on a game board. In PAC-mecium (pictured) the player guides paramecia up and down by changing the chamber’s electrical field with a joystick. Paramecia gain points for gobbling yeast cells, and avoiding a computer-animated fish. There’s Biotic Pinball, POND PONG and Ciliaball. Riedel-Kruse hopes these biotech games could become part of biology studies and contribute to crowd-sourcing and research. http://news.stanford.edu/news/2011/january/ biotic-video-games-011211.html

nature biotechnology volume 29 number 3 MARCH 2011

181

news

US farmers can again plant genetically engi- ing at the heart of the organic community,” says neered alfalfa following a decision in January Doug Gurian-Sherman, a senior scientist at the by the US Department of Agriculture (USDA). Union of Concerned Scientists in Cambridge, The ruling, which follows a tumultuous debate Massachusetts. “The biggest single use for alfalfa and four-year US court-imposed ban, comes as is dairy, and organic milk is a premium product.” a relief to the agricultural biotech industry. The Although there is no validated mechanism in the literature clarifyagency was proposing ing how transgenic to place geographic EPSPS sequences in restrictions on plantalfalfa would make ing in response to their way into cow’s organic growers’ milk, the issue is that requests. This alternaorganic products tive was only narrowly claim to avoid GM averted and could have products in any shape set sweeping regulaor form; thus, transtory precedents. genic alfalfa presents “There was proba problem to organic ably a collective sigh dairy farmers. of relief that the In 2006, a group of agency stuck with the organic alfalfa growprecedent that it has ers and nonprofit been relying on since organizations, such as it started reviewthe Center for Food ing and approving Safety in Washington, biotech traits,” says Planting glyphosate-resistant alfalfa has resumed following the USDA’s January decision. DC, sued the USDA Jeff Rowe, vice presifor approving the GM dent of biotech affairs and regulatory at Pioneer in Des Moines, Iowa. alfalfa, arguing that the agency had not fully But the events that led up to the USDA’s deci- considered its environmental and economic sion have left leaders in industry rattled. They impacts. A US federal court agreed and in 2007 are concerned that the agency will begin making ordered the agency to conduct a more thorough non–science-based concessions to the organic environmental analysis. In the meantime, crop community at the expense of biotech crop planting and sales were halted. USDA worked on the court-ordered envidevelopers and growers. Some expect litigation delays and longer regulatory timelines for crop ronmental impact statement (EIS), for nearly four years. After receiving about 244,000 approvals. Alfalfa is a high-protein forage crop for live- public comments and holding four public stock. On one side of the debate are those seeking meetings, the agency produced a final EIS on to sell and grow the biotech variety, genetically December 16, 2010. The 2,300-page review engineered to tolerate the herbicide glyphosate acknowledged the potential for genes from through expression of the Agrobacterium tume- EPSPS transgenic alfalfa to find their way into faciens transgene 5-enolpyruvylshikimate-3- nontransgenic varieties but noted that the probphosphate synthase (EPSPS) and brought to ability was “low” and depended on several conmarket in 2005 by St. Louis–based Monsanto ditions. USDA maintained its conclusion that and Nampa, Idaho–based Forage Genetics EPSPS transgenic alfalfa is safe for food and feed International. On the other, are those who mar- purposes and poses no plant pest risk. On the basis of the EIS, the agency at first ket organic alfalfa. Leaders of the organic lobby fear that proposed one of two actions: either to approve Monsanto’s alfalfa containing the EPSPS trans- the GM alfalfa fully or approve the crop in part, gene will outcross or admix with their organic with restrictions on where it could be planted. varieties. One of several reasons why consum- For instance, to segregate the transgenic alfalfa ers buy organic products is specifically to avoid from organic alfalfa, farmers would have to set transgenes in their food; thus, the presence (or up an exclusion zone of at least 5 miles. The agency said upon filing the EIS in the ‘contamination,’ as it is commonly branded) of transgenic material in organic food is viewed Federal Register December 23 it would decide as a threat to both the domestic and export after 30 days which of the two actions it would markets of organic producers. “This is strik- follow. “This final EIS is a first step toward lookJason Lugo/istockphoto

© 2011 Nature America, Inc. All rights reserved.

Industry exhales as USDA okays glyphosateresistant alfalfa

nature biotechnology volume 29 number 3 MARCH 2011

in brief DuPont swallows Danisco Early in January, agricultural biotech giant DuPont of Wilmington, Delaware, agreed to purchase Danish enzyme maker Danisco, based in Copenhagen, for $5.8 billion. The deal has not been finalized, but speculation about the potential consequences of this buyout is rippling through the Danish biotech sector. “We’ve sold one of our national treasures,” says Claus Felby, a professor of wood and biomass technology at the University of Copenhagen. Biotech researchers like Birger Moller, professor of plant biochemistry at the University of Copenhagen, fear that if DuPont decides to move Danisco’s manufacturing to the US, this may put an end to an era of fruitful collaboration between industry and basic research in the country. Equally, DuPont’s interest in Danisco could send a message about the value of Danish biotech. “It indicates we’re sitting on a gold mine here,” says Moller. In another recent transaction, Danish enzyme manufacturer Novozymes bought Darmstadt, Germany–based Merck’s bioagricultural science unit for $275 million. Merck’s divested Crop Bioscience, which makes inoculants for plant health, is a strong strategic fit for the Danish biotech located in Bagsvaerd. The companies expect to close the deal by May, pending regulatory approval. Nidhi Subbaraman

Alzheimer’s genetic map Research groups across France, the UK and US are pooling their resources to create the biggest genetic information bank on Alzheimer’s disease. Researchers participating in the International Genomics of Alzheimer’s Project (IGAP) will compare the genomic data of 20,000 individuals with 30,000 controls. Members of the project include the European Alzheimer’s Disease Initiative, led by the Institute Pasteur de Lille and Lille University, the Genetic and Environmental Risk in Alzheimer’s Disease group from Cardiff, UK, the Heart and Aging Research in Genomic Epidemiology, Boston University and the Alzheimer’s Disease Genetics Consortium at the University of Pennsylvania School of Medicine, Philadelphia. “This is the first time, internationally, we’ve all gotten together,” says Gerard Schellenberg, director of the Philadelphia-based team and professor of pathology and laboratory medicine, University of Pennsylvania Medical School. Each institute will carry out its own association analysis, and those statistics pooled into a meta analysis, says Schellenberg. With almost 50,000 individuals, and drawing on results from the 1000 Genome Project, the IGAP aims to deepen understanding of the molecular basis of rare variants of the disease, Schellenberg says, and identify genetic risk factors for the disease. IGAP’s meeting and analysis costs are currently supported by the Alzheimer’s Association of Chicago, and Foundation Plan Alzheimer, of Paris. Nidhi Subbaraman

179

news

US farmers can again plant genetically engi- ing at the heart of the organic community,” says neered alfalfa following a decision in January Doug Gurian-Sherman, a senior scientist at the by the US Department of Agriculture (USDA). Union of Concerned Scientists in Cambridge, The ruling, which follows a tumultuous debate Massachusetts. “The biggest single use for alfalfa and four-year US court-imposed ban, comes as is dairy, and organic milk is a premium product.” a relief to the agricultural biotech industry. The Although there is no validated mechanism in the literature clarifyagency was proposing ing how transgenic to place geographic EPSPS sequences in restrictions on plantalfalfa would make ing in response to their way into cow’s organic growers’ milk, the issue is that requests. This alternaorganic products tive was only narrowly claim to avoid GM averted and could have products in any shape set sweeping regulaor form; thus, transtory precedents. genic alfalfa presents “There was proba problem to organic ably a collective sigh dairy farmers. of relief that the In 2006, a group of agency stuck with the organic alfalfa growprecedent that it has ers and nonprofit been relying on since organizations, such as it started reviewthe Center for Food ing and approving Safety in Washington, biotech traits,” says Planting glyphosate-resistant alfalfa has resumed following the USDA’s January decision. DC, sued the USDA Jeff Rowe, vice presifor approving the GM dent of biotech affairs and regulatory at Pioneer in Des Moines, Iowa. alfalfa, arguing that the agency had not fully But the events that led up to the USDA’s deci- considered its environmental and economic sion have left leaders in industry rattled. They impacts. A US federal court agreed and in 2007 are concerned that the agency will begin making ordered the agency to conduct a more thorough non–science-based concessions to the organic environmental analysis. In the meantime, crop community at the expense of biotech crop planting and sales were halted. USDA worked on the court-ordered envidevelopers and growers. Some expect litigation delays and longer regulatory timelines for crop ronmental impact statement (EIS), for nearly four years. After receiving about 244,000 approvals. Alfalfa is a high-protein forage crop for live- public comments and holding four public stock. On one side of the debate are those seeking meetings, the agency produced a final EIS on to sell and grow the biotech variety, genetically December 16, 2010. The 2,300-page review engineered to tolerate the herbicide glyphosate acknowledged the potential for genes from through expression of the Agrobacterium tume- EPSPS transgenic alfalfa to find their way into faciens transgene 5-enolpyruvylshikimate-3- nontransgenic varieties but noted that the probphosphate synthase (EPSPS) and brought to ability was “low” and depended on several conmarket in 2005 by St. Louis–based Monsanto ditions. USDA maintained its conclusion that and Nampa, Idaho–based Forage Genetics EPSPS transgenic alfalfa is safe for food and feed International. On the other, are those who mar- purposes and poses no plant pest risk. On the basis of the EIS, the agency at first ket organic alfalfa. Leaders of the organic lobby fear that proposed one of two actions: either to approve Monsanto’s alfalfa containing the EPSPS trans- the GM alfalfa fully or approve the crop in part, gene will outcross or admix with their organic with restrictions on where it could be planted. varieties. One of several reasons why consum- For instance, to segregate the transgenic alfalfa ers buy organic products is specifically to avoid from organic alfalfa, farmers would have to set transgenes in their food; thus, the presence (or up an exclusion zone of at least 5 miles. The agency said upon filing the EIS in the ‘contamination,’ as it is commonly branded) of transgenic material in organic food is viewed Federal Register December 23 it would decide as a threat to both the domestic and export after 30 days which of the two actions it would markets of organic producers. “This is strik- follow. “This final EIS is a first step toward lookJason Lugo/istockphoto

© 2011 Nature America, Inc. All rights reserved.

Industry exhales as USDA okays glyphosateresistant alfalfa

nature biotechnology volume 29 number 3 MARCH 2011

in brief DuPont swallows Danisco Early in January, agricultural biotech giant DuPont of Wilmington, Delaware, agreed to purchase Danish enzyme maker Danisco, based in Copenhagen, for $5.8 billion. The deal has not been finalized, but speculation about the potential consequences of this buyout is rippling through the Danish biotech sector. “We’ve sold one of our national treasures,” says Claus Felby, a professor of wood and biomass technology at the University of Copenhagen. Biotech researchers like Birger Moller, professor of plant biochemistry at the University of Copenhagen, fear that if DuPont decides to move Danisco’s manufacturing to the US, this may put an end to an era of fruitful collaboration between industry and basic research in the country. Equally, DuPont’s interest in Danisco could send a message about the value of Danish biotech. “It indicates we’re sitting on a gold mine here,” says Moller. In another recent transaction, Danish enzyme manufacturer Novozymes bought Darmstadt, Germany–based Merck’s bioagricultural science unit for $275 million. Merck’s divested Crop Bioscience, which makes inoculants for plant health, is a strong strategic fit for the Danish biotech located in Bagsvaerd. The companies expect to close the deal by May, pending regulatory approval. Nidhi Subbaraman

Alzheimer’s genetic map Research groups across France, the UK and US are pooling their resources to create the biggest genetic information bank on Alzheimer’s disease. Researchers participating in the International Genomics of Alzheimer’s Project (IGAP) will compare the genomic data of 20,000 individuals with 30,000 controls. Members of the project include the European Alzheimer’s Disease Initiative, led by the Institute Pasteur de Lille and Lille University, the Genetic and Environmental Risk in Alzheimer’s Disease group from Cardiff, UK, the Heart and Aging Research in Genomic Epidemiology, Boston University and the Alzheimer’s Disease Genetics Consortium at the University of Pennsylvania School of Medicine, Philadelphia. “This is the first time, internationally, we’ve all gotten together,” says Gerard Schellenberg, director of the Philadelphia-based team and professor of pathology and laboratory medicine, University of Pennsylvania Medical School. Each institute will carry out its own association analysis, and those statistics pooled into a meta analysis, says Schellenberg. With almost 50,000 individuals, and drawing on results from the 1000 Genome Project, the IGAP aims to deepen understanding of the molecular basis of rare variants of the disease, Schellenberg says, and identify genetic risk factors for the disease. IGAP’s meeting and analysis costs are currently supported by the Alzheimer’s Association of Chicago, and Foundation Plan Alzheimer, of Paris. Nidhi Subbaraman

179

news

© 2011 Nature America, Inc. All rights reserved.

Box 1 Can biotech and organic farming co-exist? Co-existence—the idea that transgenic and non-GM crops of the same species can be successfully grown near each other—has for years been part of the European regulatory discussion. For instance, EU member states have imposed isolation distances on GM crop growers (Nat. Biotechnol. 28, 133–136, 2010). But it is only in the past 18 months or so that the concept has made headlines in the US with legal tussles at the state and federal level between the organic sector and seed companies or farmers growing biotech crops. In December, as a response to the fierce debate following the USDA’s proposals for approving EPSPS transgenic alfalfa, USDA Secretary Tom Vilsack wrote an open letter urging co-existence among biotech, organic and conventional crop stakeholders, suggesting that the agency wants to “forge a new paradigm based on co-existence.” The clash between biotech crops and organic agriculture has “led to litigation and uncertainty. Such litigation will potentially lead to the courts deciding who gets to farm their way and who will be prevented from doing so…. Surely, there is a better way,” Vilsack wrote. Three months ago, the agency proposed to place European-style isolation distances on EPSPS transgenic alfalfa. It later backed down from that proposal and instead announced in January it would implement nonregulatory measures to address co-existence. For instance, the agency plans to revive its advisory committee on biotech and 21st century agriculture (AC21), which in 2008 produced a report on the issues surrounding coexistence. This time, the committee will be charged with guiding the agency on ways to strengthen co-existence. But the agency’s attempt to address the issue through federal regulations has met resistance from the biotech community. “Farmers have always had disputes and farmers have always been able to work things out without getting a great deal of government involvement in the middle of it,” says Alan McHughen, a plant biotechnologist at the University of California at Riverside. Left to the growers, disputes tend to get settled “with a compromise on both sides,” he says. EW

in their words “In defense of the coffee and doughnuts, I would say there are probably some areas you can cut but I am not sure we should be inviting the most knowledgeable scientists in the world to talk about a cure for cancer and not offer them coffee.” Democratic lawmaker Chaka Fattah echoed calls to cut overblown budgets but drew the line at food. (AFP, 11 February 2011) “What disturbs me a little is that a more-than-vivid imagination is needed to think that genealogy research in Iceland is being spied on if there is nothing behind it.” Kari Stefánsson, executive chairman and president of research at deCODE, comments on last year’s WikiLeaks disclosure that Chinese authorities may be spying on the company’s genealogy and medical research database. (Iceland Review_Online, 6 December 2010) “In recent years, big pharma, through a combination of internal research and acquisitions, has accumulated as much biotechnology capabilities as any biotech. What is left to differentiate big pharma from big biotech?” Columnist Jason Chew highlights the blurring line between pharma and biotech. (Seeking Alpha, 20 January 2011) “It took us 30 years to get to where we are. So it will take some time to understand the next step, to understand the dynamics and the value.” Genzyme CEO Henri Termeer comments on takeover talks between the biotech and Sanofi-aventis. (Boston Globe, 20 January 2011) “You have to ask yourself why are physicians involved with these organizations at all? What’s the benefit to society and medicine? In my mind, the answer is little or none.” Jerome Kassirer, Tufts University Medical School professor and former editor-in-chief of the New England Journal of Medicine, supports a proposal from over a dozen medical journals to require authors of submissions to disclose payments from hedge funds or other investors. (Bloomberg, 19 January 2011) “As a politician, I tend to listen to the emotions of the people. And yet as an engineer, I also have to listen to the scientific results.” Emmanuel Pinol, vice governor of North Cotabato, Philippines, after the local government uprooted Bt eggplant field trials. (Crop Biotech Update, 14 January 2011). “We need new tools. Nothing we’ve done in the past 40 years has had an impact.” Duane Gubler, professor in infectious diseases at Singapore’s DukeNUS Graduate Medical School, comments on the release of about 6,000 sterile Oxitec OX513A male mosquitoes into an uninhabited forest area in eastern Malaysia, part of a trial aimed at controlling dengue fever. (Associated Press, 26 January 2011).

Video games played with live organisms ‘Biotic games’ that mimic classic video games have been devised by Ingmar Riedel-Kruse and his team at Stanford. Single-celled organisms are placed in a microfluidics chamber with a microscope camera to track their movements. The image is overlaid on a game board. In PAC-mecium (pictured) the player guides paramecia up and down by changing the chamber’s electrical field with a joystick. Paramecia gain points for gobbling yeast cells, and avoiding a computer-animated fish. There’s Biotic Pinball, POND PONG and Ciliaball. Riedel-Kruse hopes these biotech games could become part of biology studies and contribute to crowd-sourcing and research. http://news.stanford.edu/news/2011/january/ biotic-video-games-011211.html

nature biotechnology volume 29 number 3 MARCH 2011

181

PROFILE

Gary Pisano The author of the landmark book Science Business: the Promise, the Reality and the Future of Biotech discusses key challenges in life science commercialization.

© 2011 Nature America, Inc. All rights reserved.

H

arvard Business School’s Gary Pisano has spent several decades studying business and management strategy in the biotech sector. Here he talks about the current challenges in commercializing life science. What kinds of problems does any biotech business face? Gary Pisano: There are essentially three challenges. First, you have to solve the problem that you have uncertainty surrounding the science that prevails over very long periods of time, much longer than we see in almost any other industry. And we don’t have good structures and models for doing that. This has meant that investors have tried pulling off-the-shelf business models from other high-tech arenas, like software and electronics, where the product cycles are much shorter. These are just inappropriate, given the time horizon. Second, this business is not a one-discipline sport, it’s multidisciplinary. We use the terms ‘the life sciences’ or ‘the biotech revolution’ for convenience, but it’s very misleading. So there’s a fundamental problem in integrating the constellation of tools needed across disciplines, particularly for young companies, because it requires a certain scale. And then finally, there’s the challenge of learning; at any time, the state of the art is evolving rapidly and there’s a lot of trial and error. Organizational experience is really important to learn over time and figure stuff out. But in the biotech model, many new entrants come in with the mission of what’s the fastest exit strategy. This is not a criticism of entrepreneurs or entrepreneurship, but at some point you have to start to accumulate, as an organization, experience in how to do things. How do you see the biotech model changing going forward? GP: Biotech and pharma are part of the same ecosystem. And pharma can play a powerful

role in driving integration. What’s been happening, though, is some are going in the opposite direction. They’re saying, “We’re getting out of the early-stage R&D. We’re not going to be the experts. We just want to buy from that market.” So they’re counting on biotech to generate drugs for them. But my data show that biotech is no more productive than pharma; the productivity problem is shared. So it’s very important for pharma to retain expertise in the science. If you want to play in the scientific ecosystem, you have to be good at it. The other thing that I was hoping for, but we’re not seeing it at all— again, we’re going in the opposite direction— is that leading biotechs could emerge with new models and new ways of doing business. I think both biotech and pharma have to

“First, you have uncertainty surrounding the science that prevails over very long periods of time. Second, this business is multidisciplinary. Finally, there’s the challenge of learning; at any time, the state of the art is evolving rapidly and there’s a lot of trial and error.” learn to work together differently. They don’t do collaborative development. They do a lot of short-term deals. And it’s actually gotten worse as pharma companies have faced pipeline pressures. What they want is a late asset, which I can completely understand. But it doesn’t build the fundamental organizational capabilities required to do drug discovery. Why haven’t larger biotech companies done more in this respect? GP: The large biotech firms have fallen into some of the same patterns as big pharma, partly because they face similar financial pressures. As a biotech company grows, it starts to look more like a big pharma from an investors’ perspective. Suddenly, they’re an earnings-per-share story. And, that’s a trap. I would argue they’re too small and have too much inherent volatility to play this game. I

nature biotechnology volume 29 number 3 MARCH 2011

think it is going to require a certain amount of guts for CEOs and leaders in the business to say, “What’s the model that we want to pursue that makes sense in the longer term for our investors?” And I don’t think there’s one answer. There’s certainly a ‘Pfizer model’ that suits some investors, but there are other investors who look longer term and are willing to accept more volatility. At the end of the day, there’s a mismatch between the business model in a publicly traded firm and what biotech needs. What’s the solution? GP: We need different capital structures that are more long-term private-equity oriented. And this is again where big pharma could play a role. Investing in companies, even buying them, but letting them run independently and where appropriate, preserving the culture. Operating almost as part of their own private-equity portfolios. At the moment, too often they buy companies, assimilate them, cut costs, and ultimately kill them. What other central problem does the industry face? GP: There is a huge mismatch between the drugs biotech entrants are developing and the drugs pharma companies want to commercialize. In essence, the supply side and the demand side of the market for know-how are out of balance, and it’s been that way for 30 years. As I argued in my book, the reason for this imbalance is that markets for know-how don’t work very efficiently in this context, because know-how is a very, very hard thing to transact. For buyers (pharma), there’s an asymmetry of information; no two sellers are alike, much of their value is hidden or proprietary. For sellers (biotech), it’s a struggle to know what buyers want (or what they’ll want in 10 years), and even if they do know, product timelines are so long it’s difficult to respond. This is what leads to such a huge mismatch on the supply side and the demand side. And when big pharma companies say, “We’re going to do less internal R&D and in-license external projects instead,” I say good luck to them. I predict they won’t find what they need; and in the meantime, their internal R&D capabilities will atrophy. This is probably the single biggest strategic blunder being committed in the industry.

183

PROFILE

Stelios Papadopoulos A longtime investor, company founder, investment banker and industry observer discusses the factors shaping biotech financing.

W © 2011 Nature America, Inc. All rights reserved.

ith several decades of experience, Stelios Papadopoulos has a unique view on biotech. Here he discusses the current challenges for biotech financing. Why is the stock market important to biotech? Stelios Papadopoulos: In every business there are those who set the tone, and there are those who respond to those who set the tone. In biotech, by far the most important group that sets the tone is public investors—those who invest in biotech companies that trade in the stock market. The second is pharma companies. They decide what they want to acquire. They like product and they like technology, and on occasion, they like footprint—maybe a Japanese company looking for a major footprint in Boston, for example. The last group is the regulatory agencies, because they determine what it takes to get something approved. Conversely, entrepreneurs and venture capitalists [VCs] do not set the tone. Entrepreneurs may think they set the tone. They may think they’re visionaries going around telling everybody how their technology is great. But if nobody funds it, we’ll never know whether it was great or not. And VCs are the most responsive people in the world. Whatever they do is in response to what the stock market or pharma wants to buy. Once we understand this, the dynamics of the biotech sector become clear. How have the markets changed over the years? SP: When Fidelity launched the Fidelity Select Biotech Fund in 1986, it was a $60 million fund. That was the typical size then. But the funds have grown so that today a typical fund manager controls not $100 million, but much more than that. And there’s a limit to the number of stocks any fund manager can

184

follow. Let’s say you can follow and invest in 50. That’s a huge number, right? A $100 million fund means you can buy, on average, $2 million per stock. But if it’s a billiondollar fund, it’s $20 million per stock. Most biotech IPOs [initial public offerings] nowadays need to be heavily discounted to attract buyers. As the valuation dips to $100 to $150 million, the typical offering is $30 to $50 million. How can you invest $20 million in such an IPO? You can’t. There just isn’t enough liquidity. The investment community is also much more sophisticated than it was. It’s not possible anymore to attract investors simply through the next hyped IPO story. Does the small market cap of biotechs have other consequences? SP: We understand already that investors are migrating to bigger stocks because these have much more liquidity. In addition, investment decisions today rarely center on the speculation or expectation that a company’s technology will give rise to a successful set of products or that the company will evolve into a significant enterprise. Rather, most investments center on handicapping the outcome of a particular event, typically a potential acquisition or a clinical trial. For instance, investors make bets, months before a phase 3 trial is unblinded, as to the outcome. And it becomes very much a binary game. It’s investing, but it’s not the sort of thing that enables companies to grow the way we did in the eighties and nineties through the steady infusion of capital. To what extent is the biotech financing model broken? SP: The model is broken in one place: IPOs. The problem is the stock market is not prepared to invest in early-stage companies. Twenty years ago you could find some interesting biology in a university setting, and that was enough to start companies that within the usually prescribed three to five years, after raising maybe $20 to $30 million in venture capital, would do an IPO, even if they didn’t have products in the clinic. Today, that is no longer the case—the typical mantra nowadays is you need phase 2 data. So now the choice is for VCs to keep on investing until they’ve put in $100 million or more

over five to ten years to get from first principles to phase 2 data. The VC community does not have that kind of capital or that kind of patience. What kinds of solutions do you envisage? SP: The objective is to save innovation from becoming extinct. You could argue that for most of the eighties and nineties, biotech was the intermediary between academia and pharma. And biotech did it by inviting risk capital from the stock market. I think in some ways the lack of interest in the stock market is leading pharma to essentially circumvent the biotech sector and go directly to the source—the academic community. So that’s one way by which innovation could be salvaged. The other is the potential that government and other not-for-profit sources

“VCs are the most responsive people in the world. Whatever they do is in response to what the stock market or pharma wants to buy. Once we understand this, the dynamics of the biotech sector become clear.” will choose to fill the gap, which, as I see it, is the distance between interesting biology and compelling biology. The distinction is important, because the latter clearly and directly leads to new product ideas that a VC will fund. Most academics don’t appreciate this subtle distinction. For interesting biology to become compelling biology, one needs a fair amount of translational research. Develop a bunch of animal models. Confirm that the biology you’ve observed is conserved across multiple model systems. Maybe create some probe molecules. Not necessarily drugs, but molecules that will probe the condition and give you insight into the biology. Perhaps the recently announced National Institutes of Health initiative to form a National Center for Advancing Translational Sciences might provide much-needed capital and guidance in that area.

volume 29 number 3 MARCH 2011 nature biotechnology

PROFILE

Paul Keckley The executive director of the Deloitte Center for Healthcare Solutions discusses how the changing policy and reimbursement environment is likely to affect the biotech and pharmaceutical sectors.

© 2011 Nature America, Inc. All rights reserved.

A

veteran in health services research with experience in both the private sector and academic medicine, Paul Keckley has a unique vantage point on how US legislative and political changes are affecting healthcare delivery. Here he highlights some of the most important developments in healthcare reform and policy and how they are likely to affect the industry. How will US healthcare reform affect the drug industry? Paul Keckley: I would say healthcare reform is challenging the industry in four major areas. One would be the new tax applied to drug companies, which started this January—a $2.5-billion‑plus increase that provides rebates to states for Medicaid prescription drugs and over ten years will close the so-called doughnut hole in Medicare Part D [the coverage gap between the initial coverage limit and the catastrophic coverage threshold]. In effect, the drug industry is contributing about $80 billion toward health reform. Of course, the trade-off was that if large numbers of people were newly insured, you’d get back what you pay because more revenue is flowing into the system. But the question of how well those mandates and how well the numbers of previously uninsured enter the industry is still an unknown. The second component in the law is the role of comparative effectiveness and the Patient‑Centered Outcomes Research Institute, which has a broad scope to both evaluate the evidence around diagnostics and therapeutics and, on an annual basis by April of each year, go to Congress and say, here’s how we see the strength of evidence for comparable therapies for comparable patient populations. Because there are so many variables in looking at efficacy and it’s difficult to predict how the worlds of policy and public opinion might influence this, I

think this could present challenges to drug development. A third problem is the law basically says that the delivery system, meaning doctors, hospitals and long-term care providers, is to consolidate into risk-bearing, clinically integrated entities. The law talks about accountable-care organizations, bundled payments, medical homes and value-based purchasing. If you look at all of those, it’s essentially saying that incentives for high quality and cost need to be shifted from volume-based payments to results. This puts a lot of pressure on the cost structure and the delivery system.

“The insurance industry [needs] to take on a lot of additional cost to be in compliance with the law, so the negotiations between the health insurance industry and the drug industry are going to become much more challenging.” So we should expect doctors and hospitals to become much more aggressive in their contracting and direct negotiations with the drug industry in the supply chain. And then, lastly, there’s a substantial amount of change in the insurance industry, which has to reengineer itself and operate very differently through health exchanges. They have to take on a lot of additional cost to be in compliance with the law, so the negotiations between the health insurance industry and the drug industry are going to become much more challenging. What factors might affect the prescription drug market following reform? PK: There are four major unknowns in the health reform legislation. The first is the individual mandate and whether that successfully convinces the 32 million people who are currently uninsured to become insured. Is the penalty stiff enough? Will the market swell with people who are already sick and high cost, or will it be younger or healthy people? That’s one big bet. The second is the employer pay-or-play mandate. Does the building up of these health exchanges and these insur-

nature biotechnology volume 29 number 3 MARCH 2011

ance industry changes, coupled with the employer pay-or-play provision of the law, mean that large numbers of employers will stop providing health insurance to employees and individuals buy insurance through these exchanges? In other words, do you see employers exiting health benefits? Third, each US state bears a substantial financial responsibility to implement the law, and yet last year they had [a] $126 billion deficit. So can the states do everything that’s expected, given their shortfalls? And then, lastly, the changes proposed in the healthcare delivery system, from more siloed medicine to a more consolidated model, does that have the result of improving care or does it add cost? What about prescription drug user fees? PK: We will see dramatic increases. Much more so than the past because, given the US economic situation and Congress’s appetite to curb spending, the Food and Drug Administration’s [FDA] budget will face big cuts. The only place the FDA can go is to industry and raise its industry fees. So companies should expect substantial spikes in industry fees. What do you see as the most profound changes to healthcare going forward? PK: One would be applying evidence to care—the alignment of the evidence with the process of delivering care—because we have a substantial gap between the evidence and the care we deliver. Second is resource allocation. We have finite resources. We can’t do everything, so where do you focus? Third would be, I think, public support. I think what is problematic in many markets is that consumers really don’t think about the health system. They react anecdotally to their own experiences with doctors, hospitals, drugs and insurance, but they don’t have a systemic view. They don’t step back and say maybe of the two treatment options where the evidence is equally strong, the cheaper of the two is the one I should pursue. They’re more inclined to basically default to the doctor’s judgment and go with whatever recommendation they’re given. I think we’ve got a gap in consumer engagement in the system. We’ve got a fundamental limit on resources forcing us to make tough decisions, and the alignment of the evidence with practice is maybe the biggest hurdle of all.

185

PROFILE

Merv Turner An industry veteran talks about the challenges facing a world-leading pharmaceutical corporation.

© 2011 Nature America, Inc. All rights reserved.

I

n his 25-year career at Merck (Whitehouse Station, NJ, USA), Merv Turner has held numerous posts in R&D as well as external licensing. He now holds dual positions as Chief Strategy Officer for Merck and Senior Vice President for Merck Research Laboratories. Here he talks about the importance of innovative research to Merck’s mission of developing novel therapeutics and looks ahead to some of the challenges facing the industry. How do you view the pharma–biotech mergers and acquisitions landscape going forward? Merv Turner: We always say we are driven by the quality of the opportunity. First, you have to know what the opportunities are out there; second, you have to know whether they fit your strategy, and third, you have to be prepared to move. If these three things come together, then certainly we would move to seek to harness the opportunity. Our favored route is licensing. Constructive licenses share the risk with the biotech partner. There are times when a license is not feasible and an acquisition makes more strategic sense. Why is there such a mismatch between biotech development programs and pharma’s needs? MT: Several reasons. Many represent market opportunities that do not reach the size needed to match the portfolio needs of the biggest companies, either because they do not adequately advance standard of care or because they address true niche-market opportunities that are too small to move the needle. Others have been developed in a way that does not meet growing payer demands for demonstration of true value, and thus are truly at an earlier stage in value creation than claimed. Yet others do not have the strength of IP [intellectual property] needed to resist challenge in today’s aggressive marketplace.

186

With the costs of late-stage development rapidly increasing, regulatory demands becoming more stringent and payers looking for true value, pharma companies have a fiduciary responsibility to make sure that their portfolios contain the best opportunities that both meet unmet medical needs and provide shareholder value. Thus the bar for in licensing or acquiring candidates has been raised. There will always be an appetite for attractive candidates, but generally candidates lacking a distinct value proposition will not command the R&D investment or the price; it is a basic tenet of the marketplace.

“We’re not expecting any great transformation in productivity in the industry any time soon, yet we still believe that innovation at the core is going to be critical for us.” What will be the impact of the ongoing healthcare debate in the United States? MT: Healthcare reform is here to stay. We know what the costs of healthcare reform are for us, and we built it into our financial models, [as part of] the cost of doing business. At a macro level, the issues for us are the general ones bedeviling the industry: it is becoming much harder and more expensive to discover and develop new medicines. This is putting pressures on the innovative R&D model. We’re not expecting any great transformation in productivity coming out of the industry anytime soon, yet we still believe that innovation at the core is going to be critical for us. We also have to think of other ways by which we can add services and solutions to wrap around our molecules, to find other ways to give them value. I can think of a couple of examples. In the diabetes space, where we have a very successful drug, payers are less concerned with the cost of any one particular drug. They’re more concerned about overall management of their diabetes caseload. In a holistic sense, what can we do to manage costs of diabetes of all of our patients? Several pharmaceutical companies are starting to think like that, more from the

standpoint of the customer. Drugs are only part of the solution. What about opportunities in generics and biosimilars? MT: In the emerging markets there’s a lot of value in branded generics, molecules that are off patent but are marketed under the original brand. Some of our leading products in China are off patent. There’s value in the brand recognition, because there’s quality that goes along with the brand in markets that have yet to establish real quality in their local generics businesses. That remains important. Of course, Merck made a commitment that we are going to enter the biosimilars space, where we believe the barrier to entry—particularly in the US—will be high, restricting the number of competitors. And we think there’s a lot of value in the Merck brand, such that if we carry a portfolio of biologics under the Merck name, we will be able to work with patients and physicians and [have them] switch to our branded biosimilar drugs at discount that will be attractive to the payer and valuable to us. What will be the major changes in healthcare provision going forward? MT: Our thinking revolves around three issues. First, health information technology and the high-tech end of the reforms being pushed by the Obama administration. If this takes hold, there’ll be a big emphasis on better quality electronic medical records. We’ll get a better longitudinal view of the patient, a contextual view of the patient, and eventually we’ll get a personalized view of the patient as genetic data become more accessible and overall provide a richer source of information. Eventually we’ll get interoperability between different kinds of electronic records, and then there could be interesting market development around aggregated data, which could be used to better define standards of care and guide treatments. We see that as a big change on the horizon. Second, there will be an acceleration toward outcomes. We’re going to have to provide much more information to our payers around the value proposition of our drugs. Third, the pace at which emerging markets develop and the rate at which they take up our innovative drugs. Those are three major changes that we have to understand and watch.

volume 29 number 3 MARCH 2011 nature biotechnology

PROFILE

Anthony Coyle The man spearheading Pfizer’s Centers for Therapeutic Innovation (CTI) initiative outlines how his company hopes to spur academia collaborations.

© 2011 Nature America, Inc. All rights reserved.

P

fizer recently announced it had extended its CTI initiative to include seven research hospitals in New York City. Biotech veteran Anthony Coyle, who heads the CTI program, describes the company’s plans to reach back into academia. How do you choose partnering academic institutions? Anthony Coyle: One of the main driving considerations is the science and the focus on translational biology. We’ve been targeting institutions where we felt there is the best blend of basic research and individuals that really have that aspiration to see a basic research concept translated to the clinic—centers where we felt there is this focus, where there is a clear desire from individuals to see that preclinical hypothesis/concept being substantiated. Not just in animal models, but working with the same individuals or teams of individuals, who can take the hypothesis and ask a mechanistic question in a small human population to obtain data-rich, phase 1 studies and achieve positive proof of mechanism [POM] in the clinic. Another consideration is institutions that have a history of an entrepreneurial culture, that really understand the nature of a true partnership where we can be creative in how we think about a very different type of collaborative model. What other factors do you look for in partners? AC: Our grand ambition is to scale CTIs, both in and outside the United States, and we’ve already started talking with several different academic centers other than those in San Francisco, New York and Boston. One key factor is whether an academic institution is large enough. As the whole premise relies on co-location of the CTI, there has to be a large enough community of scientists and clinicians in a given area to justify investment

in building a lab next door. Second, are there the right types of scientists across multiple therapeutic areas who are passionate about clinical translation? And do they want to partner with us? There has to be an appreciation of the type of close collaboration we’re looking for and the opportunities the partnership presents in terms of increasing the value of an idea or hypothesis. Why focus on biologics? AC: Essentially, we want to make these groups as self-autonomous as they can be, so that the targets are co-selected and the candidate drugs are co-developed in the CTI center nested in the academic medical center. To

“There is a strong desire not to make this process bureaucratic but to simply rely on scientific rigor to take forward the best programs locally. In this model we can also take advantage of the deep therapeutic-area knowledge and drug discovery expertise and leverage that through the local sites to enhance the partnership.” build this successful partnership, we did not want to be solely dependent on other parts of the broader Pfizer community, with teams of medicinal chemists and all that’s associated with small-molecule drug development. We wanted to allow the local groups working together under a steering-committee mechanism to make the best decisions about which programs should be selected, advanced and funded. There is a strong desire not to make this process bureaucratic but to simply rely on scientific rigor to take forward the best programs locally. In this model we can also take advantage of the deep therapeutic-area knowledge and drug discovery expertise and leverage that through the local sites to enhance the partnership. In a CTI we can explore more early-stage programs, and as the science develops, it means we’ll be there first to develop those programs with the indi-

nature biotechnology volume 29 number 3 MARCH 2011

viduals who are key experts in a given target, in a given mechanism, in a given disease, and drive those decisions at the local centers. How many projects will be funded per CTI? AC: Using the UCSF CTI as an example, we’ll select eight proposals per year. The funding mechanism is capital light, with additional funding dependent on meeting milestones. By year two, some of the eight original proposals will still be in the portfolio, whereas others will have fallen out. But we intend on an annual basis to bring in eight new programs (depending on attrition of the original eight). We want to be able to terminate programs where the science, as exciting as it is, doesn’t pan out in terms of translation and early-stage positive mechanistic data in the clinic. And we’ll only fund programs where the science continues to be exciting, so our academic partners have the right incentives. How will you benchmark success? AC: We’ve broken this down into three different periods of deliverables. In the first year we will establish a preclinical portfolio across multiple therapeutic areas complementing our internal pipeline. The programs coming into CTIs will be at various different stages. Some will be very, very early, based on a phenotype of a knockout mouse or human in vitro studies. Other programs will be more advanced; for example, a mouse anti-human antibody. Here we will leverage Pfizer knowhow in terms of humanization and affinity optimization. Alternatively, we might consider an antibody that is already fully humanized but has less-developed biology and a less clear path to the clinic. By 18 months some of these programs should be in the clinic. And after three years we will have several clinical programs. Importantly, we will focus on not only the therapeutic candidates themselves, but also understanding patient heterogeneity and segmentation to develop precision-medicine approaches to target the right patient with the right therapeutic. The agreements that we have with our partners focus on demonstration of positive POMs. And Pfizer will have the right to exercise an exclusive option to develop that molecule. In five years, our ambition is to have licensed the best POMs and to have developed them to the proof of concept in Pfizer.

187

PROFILE

Elias Zerhouni The former National Institutes of Health (NIH) director lays out the numerous challenges facing the translation of academic discoveries.

© 2011 Nature America, Inc. All rights reserved.

A

s an inventor of groundbreaking technology, a founder of several startups, a leader of major public research institutions and now head of research at Sanofi-aventis, Elias Zerhouni has a unique perspective on the complex process of bringing drugs and technologies to market. Here the former NIH director spells out some of the major problems facing translation. Is the current model for translation misaligned with the healthcare challenges facing society? Elias Zerhouni: First and foremost, there’s a fundamental gap of knowledge. Despite much progress in the biological sciences, we do not yet know how to interpret complex human biology to the point where we can reliably identify safe and effective therapies. This is leading to a misalignment between growing research spending and decreasing translational productivity. This has multiple downstream consequences. First, venture capital funding for academic startups and early-stage biotech companies is drying out due to the long development times and high failure rates. The second misalignment is that government policy is swinging to rationing reimbursement to control healthcare costs. Many payers see innovation as a main culprit of rising costs; it’s almost like an anti-innovation spirit in the policy makers. In addition, an increasingly stringent regulatory system is making it more difficult to develop new therapies, especially primary-care drugs for large populations. This means that industry is being pushed into what I call specialty-care products that can [be] more easily developed rather than the primary-care products aligned with the current and future public health needs, such as chronic diseases. Then the last misalignment is allocation of human resources. In other words, there are too few MD-PhD scientists able to bridge the gap between understanding basic science and

188

human disease. When you look at academia, it has moved away from translational research by necessity. Those who could do this type of work are either consumed by what clinical service demands or would rather go into the basic, rather than translational, side of things because you can get grants, publish papers and get promoted more easily. Eighty-five percent of US MD-PhDs are at the bench, not at the bedside. What can be done to correct the human resource problem? EZ: When I was at the NIH, I pushed very hard to establish the concept of translational medicine. And when I first started talking about the need to rebuild translational research as a new discipline so that it would

“I also see a major need for us to align research with patient organizations. If you’re going to solve the Alzheimer’s problem, you won’t be able to do it with no participation in research by patients.” have its mechanisms of promotion, recognition and funding—some people were outraged. They felt that this was not the role of NIH. Well, I think otherwise; it is the role of major government agencies to find solutions to the translation problem. That’s why I created the CTSA [Clinical and Translational Science Awards] program to provide a kernel of support that will be independent of clinical services, which eat up an enormous amount of our scientists’ creative time. The point is to fundamentally challenge academic institutions to come up with their own ecosystem that will encourage translational research and bridge the widening gap between basic research and clinical impact. In many topline medical schools today there’s no such thing as a pharmacology department, let alone research in chemistry or toxicology. So we’ve tried to incentivize such activities. Another important facet of course is to provide a high-tier journal to acknowledge excellence in translation research, which is why I supported the creation of Science Translational Medicine.

How can one encourage fruitful industryacademia collaborations and avoid duplication of R&D? EZ: It’s really important to stress that translational work should never be done at the expense of continuing fundamental research. That would be a huge mistake. We cannot slow down our efforts in understanding the behavior of complex biological systems— what I’ve called the fundamental gap of knowledge. My philosophy at NIH was 60% of the budget went to fundamental research, and then 25% to what I would call translational research and 15% to public health. Having said that, one area where academic centers can help is understanding the biology of disease in human populations as early as possible using whatever method— biomarkers, adaptive clinical trials, exploratory INDs [investigational new drugs] or phase-zero trials. If you look at the behavior of the industry, it used to be closed-in R&D shops that worked within themselves and really didn’t have access to external innovation or external centers of innovation. That is changing for the better. Every company you hear now is saying, “I want to be connected. I want to work with academia. I want to have problems posed to folks who have direct interactions with the diseases themselves in human populations.” The NIH will also have a role to play in helping in the biological validation of core therapeutic hypotheses. I also see a major need for us to align research with patient organizations. If you’re going to solve the Alzheimer’s problem, you won’t be able to do it with no participation in research by patients. What do you see as the way forward? EZ: It is imperative that we narrow the gap between regulatory science, the gap between financing and the gap between academic organizations, industry strategies and patient groups around one central concept, and that is that we cannot ignore the public health requirements. Basically, we have to realize that without innovation in our innovation ecosystem, we won’t achieve the innovation that will serve the public and that needs the leadership of government agencies for it to happen. So it’s a time of change. It’s a time of reform. It’s a time of not being timid about identifying the problem and allowing people to try different ways of changing the innovation system itself.

volume 29 number 3 MARCH 2011 nature biotechnology

PROFILE

Edison Liu The executive director of the Genome Institute of Singapore surveys the changing global landscape of healthcare provisioning.

© 2011 Nature America, Inc. All rights reserved.

E

dison Liu has spent his career charting the course of translational biomedical research programs, previously at the US National Cancer Institute and for the past decade at the Genome Institute of Singapore. In this edited interview, he shares his views on the rise of the healthcare market in Asia. How are the drug and diagnostic markets in Asia likely to change? Edison Liu: I think the most fundamental change will be in the combined effects of the dominant size of the Indian and Chinese markets and importance of their views on health financing. Both are actually developing nations with a sizeable underclass where access to even antibiotics may be problematic to some in rural regions. The public sector will not pay the prices that Western countries are willing to give for even the common drugs, perhaps unless the pharmaceutical producer is owned by a Chinese or Indian national concern. When this happens, then it will be to their benefit to maintain higher profit margins for what would then be considered indigenous companies. Even then, the pricing will be lower than in the west. In terms of products, there will be more sold at either smaller unit costs, or sold cheaply as stripped-down versions of the western counterparts. This is already happening with medical devices. The downstream effect is that Western countries will want these devices if they see that the same function can be accomplished at considerably lower cost. This will mean lower margins for health device companies as well. Which areas of biomedicine should Asian governments prioritize? EL: I think new medical devices will be generated from Asian R&D primarily because of the number of engineers being trained and the history of Asian nations to be major

producers of engineered devices. I believe that medical IT will be a key sector in the new biomedical economy and that Asian companies will probably gain significant market share quickly. They have the infrastructure and the know-how already. What they lack is domain knowledge in biology and medicine in these IT sectors, but that deficit is being remedied. In terms of translational medicine, Asian institutions have been, until recently, poorly funded to do investigator-initiated clinical trials. This will change dramatically in China, where plans are afoot to develop large clinical translational networks. But they may also have an advantage primarily for two reasons:

“There will be significant political pressure from governments for the products sold in China and India to be produced there.” first, because of demographics and second, because of culture. The demographic reason is that the numbers of treatment-naive patients for any disease are huge, and a clinical trial can complete its accrual quickly and cheaply. The cultural reason is that in traditional Asian societies, the physician holds an exalted position and is highly respected. For this reason, the acceptance rate for clinical trials is high. These are compelling reasons for clinical trials to be performed in Asia. Proof-of-concept [POC] studies require more sophistication, which may not be found currently in China or India, but this situation is quickly changing in that POC units are being formed in China with Western expertise and returning Chinese expatriates. How will the changing global landscape affect biotech and pharma companies? EL: Two opposing forces are at work. Price sensitivity explains to a large extent the development of the generic industry in both India and China. Opposing this, however, a middle class is emerging who is willing to pay premium dollars—in cash—for certain high-end drugs. In the long run, these two forces may induce biotech and pharma companies not only to modulate their prices for products purchased by the public sector (like HIV antivirals) but also to provide the higher

nature biotechnology volume 29 number 3 MARCH 2011

end products to niche submarkets for the private sector as cash transactions. In either case, there will be significant political pressure from governments for the products sold in China and India to be produced there. What this will mean is that pharmaceutical giants will progressively acquire a greater Asian flair just as much as the original European pharmas became more ‘American’ as the market power shifted to the United States. Expect to see Asians being CEOs of multinational pharmaceutical companies or for an Indian company take over one of the stalwart large drug companies. How can western companies compete in Asian markets? EL: Foreign companies can compete only if they join forces with local players who understand the market who will have a stake and will not tolerate being undercut. Pharmaceutical companies will need new business models— getting into generics (which they are) and into pharmaceutical logistics (which is still poorly developed in many emerging countries). Smaller biotech companies will have to go global early. The scientific plan and the founders may be from the West, but the production and product improvement might be distributed in different Asian countries for their specific competitive edge—India for IT, China for manufacturing, Singapore for the quality stamp and final assembly. How should biotech companies think about targeting R&D efforts to capture the emerging Asian market? EL: Current biotech leaders have little or no experience with the Asian market, which is heavily driven by pricing. I think that products related to Internet access to drugs, delivery of quality care, solutions for an aging population will be where Western service know-how is important and is an advantage. This is not dissimilar to other products: the West cannot compete with China on the production of mass consumer goods, but there is a high demand for luxury-brand items (fsuch as Gucci bags and Porsche cars) in Asia, especially China. So biotech will need to be more innovative; for example, drugs that require specialized tests to maximize their effectiveness or monitor drug response in a complete therapeutic package not found in China or India. We will need to redefine what biotech is.

189

PROFILE

Greg Winter The inventor of humanized monoclonal antibodies and cofounder of Cambridge Antibody Technology, Greg Winter, muses on the future of antibody therapeutics and UK life science innovation.

© 2011 Nature America, Inc. All rights reserved.

T

hough currently in vogue, monoclonal antibodies (mAbs) took a long time to earn their present status. They were initially hailed as ‘magic bullets,’ but the excitement soon abated with the realization that immunogenicity problems compromised the use of rodent mAbs. Through the use of proteinengineering approaches, first to humanize rodent mAbs and later to make human antibodies, Greg Winter and his collaborators helped facilitate the transition from rodent mAbs to the therapeutic antibodies in use today, such as Campath-1H, Herceptin, Avastin and Humira. In 1989, Winter cofounded Cambridge Antibody Technology

“The chances of making a blockbuster may become less as differentiation starts splitting up the market. I also expect to see increasing use of smaller antibodies.” and later went on to establish Domantis; his latest venture is Bicycle Therapeutics (Cambridge, UK), where he is attempting to produce small antibody mimics with covalently bonded hydrophobic cores. Why did it take so long for pharma to recognize the value of mAbs? Gregory Winter: Several companies like Genentech, Celltech and Behringwerke believed in antibodies. But large pharma didn’t really believe; they were suspicious because of the earlier hopes that hadn’t been realized. In particular they wanted evidence that antibodies could be given for a prolonged treatment—much longer than for a

190

rodent antibody. Also, antibodies weren’t the kind of drug that they were used to dealing with; it was out of their comfort zone, so they left it to biotechs. It was probably the prospect of making money that changed their attitudes, when in the mid-90s the first engineered antibodies received [US Food and Drug Administration] approval and started to sell. With the rise of biosimilars, what do prospects look like for innovator companies focusing on mAbs, their fragments and antibody-like scaffolds? GW: The rise of biosimilars will depend on the attitude of various regulatory bodies, and market pressures, and that’s difficult to predict. But biosimilars will be pushed hard by innovators making differentiated, and better, products. For instance, the addition of extra power to antibodies, whether it’s cytotoxic drugs, effector functions or enhanced pharmacokinetics, will give an edge over biosimilars, as will the use of bispecifics. Using combinations of antibodies to proven drug targets is particularly attractive. For example, given there is angiogenesis in rheumatoid joints, it may be advantageous to combine an anti‑tumor necrosis factor-α antibody [like Humira] for rheumatoid arthritis with an anti‑vascular endothelial growth factor antibody [like Avastin] that inhibits angiogenesis. An implication is that the chances of making a blockbuster may become less as differentiation starts splitting up the market. I also expect to see increasing use of smaller antibodies: fragments with, for example, appendages that could extend their serum half-life. In what directions do you anticipate future mAb therapies going? GW: I can see at least two directions for the technologies. For antibodies with a very long half-life in serum (as may be possible to achieve by mutation of the binding site for the recycling receptor), or where the biological response is prolonged, it is possible to envisage injections once a year or every 6 months, rather than a daily pill. That’s got to be a lot more convenient and efficient. Amgen’s denosumab is already injected every 6 months. Another route is to make antibodies

much smaller, to penetrate the tissues more effectively. With the startup Bicycle Therapeutics we are trying to make tiny antibody mimics with binding and effector activities, that can be cut and pasted together just as in antibodies. Perhaps these can be made orally available. As a founder of three companies, who remains in academia, what do you consider to be the best way of translating academic discoveries? GW: Dream while (just) keeping your feet on the ground. More specifically, from discovery to translation, keep an eye on the big picture as well as the nitty-gritty details, and be lucky! I was lucky, as the [Laboratory of Molecular Biology] with its block grant gave me carte blanche for getting on with my work without distinguishing between pure and applied science. It would have been very difficult to have made my inventions on classic grant funding (it would have been seen as too applied) or on industry money (it would have been seen as too early, and anyway most companies weren’t interested in antibodies at the beginning). I immersed myself fully in the process of translation and was lucky because the whole antibody field was about to expand greatly; with hindsight it was like pushing on an open door, but at the time it didn’t feel like that. To what extent is UK life science innovation under threat? GW: It is under serious threat. Life science innovation is an international business; the UK is in competition with the rest of the world. We do have good scientists here, but they are not that well paid, and we will certainly lose them if they can’t get the money to do their science here, or if the bureaucracy involved in running a scientific group continues to expand. We are also seeing a reduction in the number of pharmaceutical jobs in the UK. That’s very worrying, especially for the message it gives. If it is thought that the UK pharmaceutical industry is on the slide, it could become a self-fulfilling prophecy, and difficult to reverse. If we can’t hold onto our academic stars and our pharmaceutical industry, we are in danger of losing life science innovation and the industries of the future.

volume 29 number 3 MARCH 2011 nature biotechnology

PROFILE

Lee Hood Lee Hood outlines his vision of personalized medicine for the next 10 years.

© 2011 Nature America, Inc. All rights reserved.

A

s a pioneer in new technologies that affect genome science and systems biology, Lee Hood of the Institute for Systems Biology in Seattle has set his sights on transforming the practice of medicine. Here he provides his vision of how new approaches will change the way we view health and disease. What is P4 medicine? Lee Hood: P4 medicine stands for medicine that is predictive, preventive, personalized and participatory, but the basic idea is that P4 medicine looks at medicine from an informational point of view. In the next ten years, patients are likely to be surrounded by virtual clouds of billions of data points and we will use IT [information technology] to reduce this data dimensionality to form simple hypotheses about health and disease. Individuals differ by 6 million nucleotides on average and are exposed to different environmental stimuli. The old population-based methods gave you bell curves for traits and if you were on either tail of the curve, you were sick. In contrast, with P4 medicine one treats each person individually and not as a part of a group. Can you describe this approach in more detail? LH: There are basically two major types of biological information, the digital information of the genome and environmental signals arising from outside the genome. The information structures that connect these two types of information with the phenotype in health and disease are the biological networks that capture, integrate and modulate information and then pass it off to molecular machines that execute the information. The disease-perturbed dynamics of these networks lie at the heart of understanding disease mechanisms. For example, in our study on prion disease in mice (Mol. Syst. Biol. 5, 252, 2009), we identified about 300 disease-perturbed genes that are involved in four major and six minor biological networks. The dynamics of these networks explained virtually

every aspect of this neurodegenerative disease. The networks became disease-perturbed in a sequential fashion. If you want to think about early diagnosis and therapy, you should focus on the first disease‑perturbed network both to identify biologically relevant molecules secreted into the blood for diagnosis and to find drugs that can reengineer the disease-perturbed network to make it behave normally, thus abrogating the progression of the disease. We also identified organ-specific blood markers, that make blood a window through which we can distinguish health from disease. All organs have specific markers that are secreted into the blood and constitute a molecular fingerprint that reports, by concentration changes, shifts from a normal to a disease‑perturbed state. Why do so many published biomarkers never make it to the clinic? LH: Success can be improved by rationally choosing biomarkers rather than doing a random shotgun search to detect biomarker changes between disease and health. I would guess that 99% of those biomarkers that are discovered by random searches are not going to be very useful. Most likely they just represent biological noise.

“The real issue with large data sets is that the data sets have enormous signal-tonoise problems.” What technologies are needed to make P4 medicine a reality? LH: A key advance is that we are now able to do complete genome sequencing of families to identify genes that are involved in simple genetic diseases. We are now beginning to apply genome sequencing to families with more complicated genetic diseases. Those studies look promising as well. In addition, the genomes of individuals will increasingly provide insights into the future health trajectory of the individual. A second transformative technology is the development of targeted proteomic assays for essentially all human proteins that my institute has recently announced. A third area is the use of microfluidic chips to be able to quantify not tens of proteins, but eventually thousands of proteins from a droplet of blood in just a few minutes. Making these devices is relatively straightforward, except for one thing: we need better protein-capture agents, such as aptamers or peptide binders, for protein assays. Fourth, single-cell analysis will be incredibly important for assessing

nature biotechnology volume 29 number 3 MARCH 2011

distinct quantized populations of cells. The idea that you can learn a lot about biology by looking at the individual cell rather than averaging populations of cells will provide fundamental new insights into cancer, development and physiology. How will this translate into a shift from disease treatment to disease prevention? LH: Systems thinking about disease gives you an entirely new strategy for identifying drug targets. The drug companies are good at making drugs once they have the target, but they are really bad in choosing the target. If you understand the nature of disease-perturbed networks, you can reengineer disease-perturbed networks to be normal. In most cases, this is clearly going to take multiple drugs and we will need good biomarkers to detect the early changes in these networks. It is a short step to design drugs that prevent potentially disease-perturbed networks (predicted from your genome sequence) from ever becoming disease-perturbed—true preventive drugs. What will this mean for the provision of healthcare in the future? LH: The focus of healthcare will shift over the next ten years from disease to wellness. We are developing metrics for assessing an individual’s wellness. There will be a wellness industry that in time could dwarf the healthcare industry. Medicine will also be focused entirely on the individual in the future. We will all have the equivalent of iPods that will be recording enormous amounts of personal data and transmitting it to servers for analyses that will monitor your wellness status and report developments that are a cause for alarm by sending you a signal, such as “Slow down on eating.” Data analysis tends to lag behind data generation in biology—do you see this changing? LH: I think the real issue with large data sets is that the data sets have enormous signalto-noise problems. If you measure a given phenotype response, it could be the sum of a number of different biological phenomena. If you are only interested in one of them, you have got to be able to subtract away the others. Learning to do that biological subtraction is one of the grand challenges in P4 medicine. That is the reason the genome-wide association studies have only been marginally effective. The signal-to-noise issues are overwhelming. 191

PROFILE

Robert Weinberg A decade after publishing the seminal “The hallmarks of cancer” paper in Cell with Doug Hanahan, Robert Weinberg reflects on where we stand in the fight against cancer.

© 2011 Nature America, Inc. All rights reserved.

T

he discovery of the first oncogene, Ras, and the first tumor suppressor, Rb, are just two of many seminal contributions of Robert Weinberg to our understanding of cancer. The founding member of the Whitehead Institute for Biomedical Research and Daniel K. Ludwig Professor for Cancer Research at the Massachusetts Institute of Technology talks about the successes and failures of cancer research in recent years. How has your thinking changed since the ‘hallmarks’ paper was published in Cell? Robert Weinberg: The six hallmarks Doug Hanahan and I described in the year 2000 are, we think, still central. The question is whether there are additional properties of cancer cells that we need to consider. For example, there are enabling characteristics that make the acquisition of the six hallmarks possible. One of them is enhanced mutability and destabilization of the genome. Another may be chronic inflammation. There are also “emerging hallmarks,” specifically immune evasion and the altered metabolism of cancer cells that Otto Warburg first described. There has also been an explosion of information indicating unambiguously that the tumor microenvironment has strong effects on the behavior of tumors. Is this one of the reasons why most experiments done in animal models never really translate to humans? RW: It could be that the interactions with the stroma are indeed captured in the animal models, but there are more fundamental problems with the mouse models of cancer, which preclude them from being very effective at present. For one thing, the cancer cells that are used in most mouse models, most often from the NCI-60 cell line collection, are poor representatives of the cells in actual tumors.

192

Are you excited about the contributions of large-scale genomics cataloging initiatives to understanding cancer? RW: One consideration is how much bang we get for the buck. But as the cost of sequencing and data analysis comes down, that will be less of a consideration. We now begin to focus more on the intrinsic ability or the inability of sequencing to provide useful information. Are we learning more about human tumors? We have certainly learned a lot more about the sources and the extent of genetic instability within tumor cell genomes. Still, the question is whether these kinds of studies contribute to understanding the physiology of a cancer cell. The insights gained to date have been real but, relative to the effort expended, modest. What about the role of stem cells in cancer? RW: My lab works a lot on cancer stem cells, and the more we look the clearer their existence becomes. The controversy has already begun to subside. A question that remains is whether cancer stem and non-stem cells are interconvertible. We find the bidirectional interconvertibility between stem cells and non-stem cell populations to be real. The recognition of the existence of cancer stem cells is critical to developing new kinds of therapeutics, if only because cancer stem cells often turn out to be more resistant to conventional therapeutics. How soon before progress in understanding metastasis translates into advances in treatment? RW: We still don’t understand all of the fundamental properties of metastatic cells. We may nevertheless empirically stumble across antimetastatic drugs without understanding why they are working. But if we wish to embrace rational drug design, we’re still in the awkward position, because we don’t understand all of the biochemical distinctions between primary cancer cells and their metastatic offspring. On the positive side, metastasis research has exploded over the past five years and our understanding of the molecular mechanisms that enable the physical movement of cancer cells from the primary tumor to distant organs has improved enormously. The next major challenges concern the complex programs

that allow cancer cells to adapt to foreign tissues. What about the contribution of epigenetics and microRNAs in cancer? RW: At one level, epigenetics and microRNAs represent additional components in the already complex circuitry of signaling pathways. An interesting question, which I cannot answer, is whether the study of microRNAs and histone modifications is going to generate entirely new conceptual paradigms. I think there could be surprises in both areas, because certain microRNAs and histone modifications are surprisingly specialized in affecting very discrete processes. What are the other biggest breakthroughs in cancer research over the past five years? RW: I might be prejudiced, but I think the discoveries of the importance of the epithelial-mesenchymal transition, the EMT, and cancer stem cells have profoundly altered many peoples’ thinking about the way tumors arise and disseminate. Many other discoveries have been interesting, but to my mind in a conceptual sense incremental. If you could set the priorities for cancer research for the coming years, what would be on the top of your list? RW: Something many people would probably not be interested in hearing. The big advances in our understanding of cancer have come year after year, decade after decade from small, independent research groups, rather than large research consortia. I’m hoping that the people who run the funding agencies come to recognize that much of the monies that are spent on multicenter collaborative initiatives—I’m not talking about clinical research—are not spent very effectively. Funds should be diverted instead to fostering young investigators to start their own groups and to move out in their own directions. I feel very passionate about this. We are losing generation after generation of young researchers. On the technology side, I think developing better xenograft models is going to be critical. Right now, to my mind, the major logjam in moving drugs ahead is that the preclinical testing of drugs is still so primitive.

volume 29 number 3 MARCH 2011 nature biotechnology

PROFILE

Arnold Demain

A

unlike a drug for a chronic condition that is taken every day for the rest of your life. But keep in mind that in recent years, natural products that are large molecules have become a new part of the pharmaceutical industry. I consider monoclonal antibodies to be natural products. They’re booming and increasing every year. A lot of the natural product people that haven’t been let go by the pharmaceutical industry are availing themselves of microbes to make larger molecules or applying microbial technology to the use of mammalian cells to make such products.

What areas of natural product discovery and biotech are we not paying attention to that we should be? Arnold Demain: I would say the discovery of new small-molecule natural products like the antibiotics. The emergence of resistant organisms is a developing crisis. I don’t know what’s going to happen here. But if governments aren’t interested in doing this, I think it’s going to be a disaster. I realize this is not a time to talk about the government spending more money, but I think we need something like a national institute—a new NIH [National Institutes of Health] institute on antibiotic and natural product discovery— because nobody has the money to do this.

Can you talk more about the scientific challenges associated with natural product discovery? AD: There are several technologies that have been around for a few years but that have not really been pursued in the major way that the pharmaceutical industry used to pursue something. For example, genome mining, i.e., genetically examining the organisms like Streptomyces for example, that make a lot of products. You look at the genomes and discover genes encoding groups of enzymes that allow you to predict that they are making natural products. Another one is metagenomics, since 99% of the bacteria have never been cultured in the lab. You take the DNA out of the soil and you put them in E. coli, and you now find out what these environmental microbes make. There’s only one little company, NovoBiotic Pharmaceuticals (Cambridge, MA, USA), that is learning how to culture these microbes and finding new products being made. There’s so much technology out there that is not being exploited in a major way.

Why has natural product discovery hobbled behind other approaches to pharmaceutical discovery? AD: One reason is that once the low-hanging fruits were taken off the tree, companies kept discovering the same products. Then they invested a lot of money in failing technologies like combinatorial chemistry and things like that. And after companies got the cream of the crop, the industry felt it was too difficult. A second reason is that molecules, like antibiotics, don’t fit the business models of pharmaceutical companies. With most of our antibiotics, you take them for a couple of days and you get well. So the usage is limited,

Where in this area can genomic technologies be best applied? AD: The marine environment is one story I am very familiar with. The ocean is great, because there are all sorts of interesting organisms including Streptomyces that no one thought existed in the ocean, but they certainly do. Bill Fenical’s group at the Scripps Institution of Oceanography (La Jolla, CA, USA) has been discovering compounds for years. I was on the board of a company which he started called Nereus Pharmaceuticals (San Diego). The first two compounds that were given to Nereus by Fenical’s group are in clinical trials now for cancer.

© 2011 Nature America, Inc. All rights reserved.

A trailblazer in the field of antibiotics reflects on natural product discovery in the genomic age.

rny Demain’s storied career has spanned pioneering work on β-lactam antibiotics and fermentation microbiology at Merck (Whitehouse Station, NJ, USA), teaching and research at the Massachusetts Institute of Technology for over 30 years and being a founding scientific advisory board member of our predecessor, Bio/technology. In this edited interview, he discusses the prospects of new genomic technologies fueling a comeback for natural products as drugs.

nature biotechnology volume 29 number 3 MARCH 2011

To what extent is technology rather than business issues, such as an excessive focus on blockbusters like statins, the problem? AD: It’s interesting you mention statins, rather than some synthetic compound, because statins are microbial products. They inhibit the key enzyme in cholesterol synthesis. [Akira] Endo in Japan and Merck worked on the enzyme and came up with natural products that inhibited it. There are many enzymes in medicine that have to be inhibited, and the story of statins provides an example of how one could go after them. So I say there are current problems that are solvable by the [natural product] technology. But we need the dollars.

“India and China... have a real opportunity, because our pharmaceutical industry has dropped out. Europe’s pharmaceutical industry has dropped out. But if somebody doesn’t put a batch of money in there, nobody is going to isolate those compounds.” With R&D funding increasing in China and India, which have their own tradition in herbal medicines, are natural products likely to be a larger part of the drug pipeline going forward? AD: No, I don’t think so. India and China, of course, are busy in the pharmaceutical area. But as far as I know, they are not investing in discovery of molecules, even though they’re the ones who discovered these herbal medicines and there must be active compounds in there. And that’s a pity, because there’s no doubt that these herbal medicines work. But they’re just not investing or discovering molecules. They keep working on these same molecules that we in the West discovered years ago. They have a real opportunity, because our pharmaceutical industry has dropped out. Europe’s pharmaceutical industry has dropped out. But if somebody doesn’t put a batch of money in there, nobody is going to isolate those compounds.

193

PROFILE

Irv Weissman An authority on hematopoiesis talks about the difficulties encountered in commercializing stem cell therapies.

© 2011 Nature America, Inc. All rights reserved.

I

rv Weissman is Virginia and D.K. Ludwig Professor for Clinical Investigation and Cancer Research and Director of the Institute for Stem Cell Biology and Regenerative Medicine at the Stanford University Medical School. The first to isolate any stem cell, he has pioneered investigation of the hematopoietic system. What challenges have you faced in translating your research to the clinic? Irv Weissman: Shortly after 1988, when we isolated the mouse hematopoietic stem cell (HSC) and figured out how to make a mouse that had a human blood-forming and immune system, Mike McCune and I formed Systemix. Within two-and-a-half years we had isolated the human HSC. When we were about to go public in 1991, we started getting inquiries from big pharma. Soon after, Sandoz bought 60% of Systemix, and we started moving toward clinical trials. We wanted to isolate cancer-free HSCs from patients with breast cancer and non-Hodgkin’s lymphoma and transplant them back to patients after very high-dose chemotherapy, rather than using bone marrow or mobilized blood cells, which are contaminated with cancer cells. We began clinical trials in 1996, but in that year Sandoz merged with Ciba to form Novartis, and Novartis made a business decision to shut down the trials. They were not going to be a stem cell isolation company for service. So the lesson with the first company was that the culture of big pharma isn’t the culture of cell therapies. But the science was right. Fourteen years after the first patient got a stem cell transplant, we have just summarized our breast cancer experience, and I’ll just say that the outcomes in our small number of patients have exceeded expectations. If it were a small molecule, it would be out there.

194

What about the two other companies you founded? IW: Cellerant is still alive, but they don’t do HSCs because they never got the money for a trial. At Cellerant I wanted to repeat the breast cancer trials with more patients but the length of time required wasn’t acceptable to the VC [venture capitalist] investors. I also wanted to do SCID [severe combined immunodeficiency] with HLA [human leukocyte antigen]-matched sibling donors, or the mother as an HSC donor. Since the mid-50s, we knew that donor T cells, a component of bone marrow and blood, mediate an immune reaction against the host. But with pure HSCs there are no T cells, and there was no graftversus-host disease (GvHD). So if you were lucky enough to cure SCID, you might be able to do sickle cell, thalassemia, and then diseases like diabetes, multiple sclerosis, lupus, all of which we’ve shown in mouse models are diseases of blood-forming cells that we can cure with HSCs from a disease-resistant donor. We had all of that in the mid-90s. When I proposed to the board at Cellerant that we do SCID first, the CEO blocked it, saying there’s not enough money in SCID patients. I said, no, but we can do a world of good, providing HSCs without GvHD. I couldn’t convince them. So this was a second lesson: if you try to do a phase 1 trial with VC backers and CEOs who control your destiny and whose function it is to make a profit in five years, the timelines are wrong. Rusty Gage, David Anderson and I formed Stem Cells to look for stem cells outside the blood system, because with HSCs we could induce tolerance in mice to any organ transplant from the same donor, and it made sense that it would also induce tolerance to tissue or organ stem cells from the same donor. The company isolated human brain stem cells and worked on spinal cord injury and Batten disease. But the clinical trial for Batten disease, which is fatal in childhood, was turned down at a prominent medical center by its IRB [institutional review board] because it involved children, even though no adult exists with the disease. So Stem Cells did phase 1 trials elsewhere and now has approval to enroll early-stage patients. The point is that you don’t learn about these kinds of barriers until you’re actually trying to open a new field, the field of regenerative medicine with

tissue-appropriate stem cell transplants. No current pharmaceutical company, no group of VCs know what the field of stem cell transplants is, and when they apply their culture and timelines and business parameters, you don’t go forward. If you were to start a company today, what would you do differently? IW: I wouldn’t start a company now unless I had a pretty high degree of control and, much more importantly, had progressed in the university through at least phase 1/2 trials. We have a CIRM [California Institute for Regenerative Medicine] disease team grant to take an anti-CD47 antibody to clinical trials in acute myelogenous leukemia [AML]. We are collaborating with the UK AML trials group, who have taken advantage of universal healthcare to organize clinical trials. In the old days we would have formed a company, but now neither we nor our university will grant licenses to form a company until we get through phase 1/2 trials. Because CIRM and the UK fund through phase 1, we are taking the place of a biotech. We’ve put together a great team that is moving these efforts forward. But I’ve made mistakes trying to form a company, because I’m a scientist and a doctor in an institution that is trying to save people. Unfortunately, the VCs want a profit in five years, which excludes most of what I want to do. So how are we going to get around this dilemma, when the hospital and the medical school want to save lives, and the companies want to make a profit? Something new is needed. I can think of lots of reasonable business models that would charge appropriately for stem cell transplants that regenerate healthy tissues for life. For example, if Systemix had succeeded with its early plan to establish HSC separation units, it would have done this next to a hospital. So why not partner with the hospital to establish and run such units? The hospital and medical school could experiment with how to set up an efficient HSC isolation and transplant and clinical care service, and how to resolve issues of compensation. Should you do it in an outpatient setting? Should you have hospice units? As they explore these issues, I think a model will emerge. Somebody will then fund it when they see they can make money. I think it will happen first at places that have the whole collection of resources, and that will build the model.

volume 29 number 3 MARCH 2011 nature biotechnology

PROFILE

Barbara Mazur A research leader at a major agrochemical company comments on the application of biotechnologies in commercial crop science.

© 2011 Nature America, Inc. All rights reserved.

B

arbara Mazur, a long-time employee of Dupont and its Pioneer Hi-Bred business (Wilmington, DE), currently heads their biotech research strategy for Dupont Agricultural Biotechnology. In the following interview, she reflects on the technological advances from the past five years and emphasizes the synergies between conventional methods and modern molecular technologies in creating the crops of the future. What kinds of technological challenges remain in agbiotech? Barbara Mazur: For transgenic crops, we still use fairly blunt approaches to express genes. The more we can refine expression technologies, the better transgenic products can be. Another need is for a better understanding of gene networks and the gene interactions that can dampen transgene efficacies. Genomics advances have changed our understanding of the corn genome: corn lines can have deletions and insertions relative to each other, and the genome itself is dynamic. As we increase our use of nextgeneration sequencing technologies, we’ll be able to refine our understanding of the functional roles of genes, and of epigenetics and intergenic sequences in regulating gene expression. When will we see crops with complex traits that benefit consumers? BM: There’s widespread work on drought tolerance, which may require combinations or sets of genes to create effective traits. Groupings of native and transgenic genes could lead to improved traits. For example, we have developed a maize hybrid [Optimum AQUAmax hybrids], which carries a native trait for drought tolerance and can be combined with drought-tolerance transgenes. A soybean line that I’m excited about is Plenish [containing an extra copy of the soybean fatty

acid desaturase gene (gmFAD2-1)] high-oleic soybeans, which we anticipate being launched next year upon regulatory approvals. These soybeans have 75% oleic acid, which is a monounsaturated, healthy oil. Plenish soybeans have optimized functional properties for foods and can be used in environmentally friendly industrial applications. Why has the introduction of the high-oleic soybean taken so long? BM: The high-oleic soybeans were made with a single transgene, but trait deregulation took many years because regulatory requirements have been continually changing. Regulatory requirements are not harmonized globally, and developing a more standardized deregulation process would enable products to reach consumers faster. Which advances from the past five years do you regard as key? BM: The use of DNA markers has made a tremendous difference for native and transgenic trait introductions and for breeding. Often, a native trait of interest can be accompanied by yield drag. By identifying markers that flank the trait, breeders can move the trait into high-yielding germplasm without the flanking deleterious genes. Over the past decade there has been a thousand-fold increase in molecular marker use at Pioneer. Another breeding technology that’s been particularly important is the use of doubled haploids. Haploid seeds can be genetically induced from diploids and then chemically converted to genetically pure, or homozygous, diploid lines. Significant improvements in both the technologies to produce the haploids and to double the haploids have occurred, and have changed corn breeding timelines. The number of pure lines that we have created as doubled haploids in recent years equals all the lines that we developed in the previous 80 years. What other biotechnologies are being used in R&D? BM: Agriculture is benefiting from advances in the field of small RNAs. We can now use microRNAs to control gene expression and double-stranded RNAs for pest protection. We routinely use association genetics for native-trait identification, and a number of automated spectroscopy and imaging

nature biotechnology volume 29 number 3 MARCH 2011

technologies for functional-genomics studies. Information management systems, and bioinformatics and computational biology tools are critical to the successful application of all these technologies. We’ve also been using gene-shuffling technologies to introduce diversity into trait genes. Genes are shuffled and variants selected in parallel: the gat [glyphosate N-acetyltransferase] herbicide-tolerance gene had almost no activity before gene shuffling, but enzyme variants with increased detoxification activity were produced in successive shuffling rounds. Can technology address emerging insect resistance in Bt crops? BM: We have developed a ‘refuge in a bag’ product for insect control, which simplifies insect refuge requirements for the farmer by carrying a mixture of the trait seeds and refuge seeds in a single bag. With integrated

“Over the past decade, there has been a thousand-fold increase in molecular marker use at Pioneer.” seed refuge products, the grower doesn’t have to stop planting the Bt corn and switch to planting the non-Bt refuge product in a separate plot. How can we foster adoption of valuable agbiotech crops in developing countries? BM: I worked with the Gates Foundation [Seattle] in setting up the IMAS [Improved Maize for African Soil] Project. I was impressed with the way the Gates Foundation researched possible projects to ensure that they met important societal needs and could be successfully completed, and also ensured that program objectives and milestones were in place to track progress. I’d also like to comment on the importance of intellectual property [IP] protection, which allows companies to realize a return on their investments that can support future research programs. In countries with good IP protection, we feel more comfortable adding research and commercialization programs. India, for example, has strengthened IP protection, and we have recently established a large DuPont research center in Hyderabad.

195

PROFILE

Lee Lynd Reflecting on progress in the bioenergy sector, Lee Lynd considers the prospects of producing liquid biofuels on a scale sufficient to impact energy challenges.

© 2011 Nature America, Inc. All rights reserved.

I

n January the Obama administration pledged to increase funding for research to provide clean energy, building on the initiative started by President Bush in 2006. Lee Rybeck Lynd is the Paul and Joan Queneau Distinguished Professor of Environmental Engineering Design and adjunct professor of biology at Dartmouth College, cofounder and chief scientific officer of Mascoma (Lebanon, NH, USA), initiator and coordinator of the Global Sustainable Bioenergy Project, and focus area lead for biomass deconstruction and conversion at the US Department of Bioenergy Science Center in Oak Ridge, Tennessee. Here he shares his thoughts on the potential and future of liquid biofuels. Is it clear the world needs biofuels? Lee Lynd: Biomass is by far the most viable sustainable source of liquid fuels today, and liquid fuels are by far the simplest energystorage medium for transportation. Liquid fuels provide greater than 50% of US transportation energy in aggressive scenarios for electrification of light-duty vehicles, and batteries are impractical for aviation and probably also for long-haul trucks. Fuel distribution and storage are expected to more than double the cost of hydrogen production but are much cheaper for liquid fuels. Biofuels will likely be a significant part of the energy supply picture for the indefinite future if key obstacles can be overcome. Can costs of cellulosic fuels be competitive with fossil fuels? LL: Cost is a surmountable barrier. Refined products from petroleum costing $75 per barrel (about $13.5 per gigajoule) are worth about $100 per barrel. At $60 per dry ton, which is $4 per gigajoule, the price of cellulosic biomass is about a third the price of oil. Thus, the cost of biomass refining can be three times that of petroleum refining and still produce products at a competitive price. Cellulosic biomass is not inherently 196

more difficult to process than petroleum. Petroleum has the advantage of being a fluid, but biomass is more reactive and much more amenable to biotech. The scale of production has less impact on the relative cost of processing cellulosic biomass and petroleum than is often assumed. What about feedstock supply issues? LL: The land needed to provide for mobility using biofuels is influenced by the site range and productivity of biomass crops, the extent of integration with other agricultural activities, the conversion process yield, fuel demand, food production and diet. I see increasing evidence that some combinations of these variables allow graceful production of biofuels in very large amounts. In the Blue Map scenario of the International Energy Agency, based on reducing greenhouse gas emissions to 50% of current levels by 2050, biomass provides 23% of primary energy. I think it is probably possible to produce enough biofuels to meet the world’s transportation needs, perhaps several times over, while feeding humanity, not clearing wild land, and maintaining or enhancing environmental quality. Systematic analysis of this possibility, the focus of the Global Sustainable Bioenergy Project, is however in the relatively early stages. How important will biotech be? LL: Likely important and likely twice: once in feedstock production and again in biological conversion of feedstocks to fuels. I expect that biotech will play a central role in production of cellulosic biofuels. Regardless of processing technology, biotech is a powerful route to develop desired traits in feedstocks. Which microbes will be most important for cellulosic biofuels? LL: Development of microbes able to produce cellulosic biofuels without added enzymes—the holy grail for low-cost processing—can be pursued by two approaches. Start with established industrial microbes, which generally do not ferment cellulose, or start with cellulose-fermenting microbes from nature, which are not industrial microbes. My guess is that this development will occur first commercially using established industrial microbes but will ultimately work best with naturally cellulolytic microbes.

What about algae? LL: Algal biofuels should be investigated as an alternative to petroleum-based fuels. That said, light only penetrates about a centimeter into a thick cell suspension, and surface-to-volume ratios are huge for algal culture as a result. I am doubtful that this challenge can be overcome at the low costs required for bulk fuel production, but this is an important focus for research. How important will it be to produce a diverse range of biofuels? LL: Cost-competitive conversion of lignocellulose, the most important next step for biofuels in both societal and commercial contexts, will likely occur over the next few years and will likely involve production of ethanol in order to avoid compounded technical risk. I expect commercial production of infrastructural compatible fuels from readily fermented feedstocks (for example, corn, sugarcane) to also occur during this time. Thereafter, it is likely that fuel molecules in addition to ethanol will be produced from cellulosic biomass to accommodate a diversity of transportation modes. Cost and performance, including environmental performance, will be the key factors in determining the mix of biofuels, and will likely prove more important than compatibility with existing infrastructure. The idea that biofuels have to be compatible with existing infrastructure amounts to saying that a system we know has to change is not capable of changing. We wouldn’t get very far applying that approach to electric or hydrogen-powered vehicles. Why do experts reach such different conclusions about the merit of biofuels? LL: The optimists’ and pessimists’ views are widely interpreted as different answers to the same question but are actually answers to two different questions. Whereas optimists ask, “What could be the role for biofuels given innovation and change to achieve a sustainable outcome?”, pessimists ask, “What would be the consequences of expanding biofuels based on extrapolating current practices and trends?” The main criticism of the optimists’ view is that the changes they argue are possible may not occur. The main criticism of the pessimists’ view is that extrapolating current practices and trends does not lead to a desirable energy future. We need to develop a better understanding of the potential of biofuels unconstrained by current practices and trends, and then use that understanding to inform our path going forward.

volume 29 number 3 MARCH 2011 nature biotechnology

n e w s f e at u r e

Fresh from the biologic pipeline—2010

© 2011 Nature America, Inc. All rights reserved.

Apart from a drug produced in rabbits, 2010 was in some ways an unremarkable year for biologic drugs coming onto the market. Emphasis on safety may be keeping the reins on new product registrations for some time to come. Jim Kling reports. 2010 saw the approval of only six new biologic drugs (Table 1), which follows the precedent of the past few years (Fig. 1). “The number of biologics approved in 2010 has been pretty light,” says Bryant Fong, managing director of Burrill & Company of San Francisco. Yet, 2010 saw some milestone product registrations. Seattle-based Dendreon’s Provenge (sipuleucel-T) became the first therapeutic cancer vaccine approved when the US Food and Drug Administration (FDA) cleared it in April. Thousand Oaks, California-based Amgen scored twice with denosumab, a monoclonal antibody (mAb) that binds to the receptor activator of nuclear factor kappa B (RANK) ligand, which got the nod for osteoporosis in June (under Prolia brand) and for prevention of bone injuries in patients with bone metastases in November (under Xgeva brand). 2010 was also the first year that a biologic drug produced in rabbits was given the green light. REMS on the rise In the wake of market withdrawals and safety controversies over the past decade, the FDA continues to emphasize safety, with many drugs being approved only if risk evaluation and mitigation strategies (REMS) are put in place. REMS, first instituted in 2007, are designed to ensure that the benefits of a product outweigh any risks. The agency had been trending toward weighing risk and efficacy in new approvals, but “REMS really accelerated that. REMS became common for newly approved products, even classes of products,” says Andrew Emmett, managing director of science & regulatory affairs at the Biotechnology Industry Organization, based in Washington, DC. The program has led to some delays in applications. “It’s very important to communicate about REMS early in the process, at the end of phase 2 trials, as well as during the review itself. Many applications have gone through multiple review cycles because REMS [were] initiated late in the process,” says Emmett. Even with REMS as a tool, the agency may still refuse a drug when REMS could have alleviated risk. “I see in general that it’s harder to

get drugs approved, due to a lower tolerance for safety issues,” says Philip Katz, a partner at Hogan Lovells of Washington, DC. REMS requirements are likely to continue to be common, he says. “This is part of the perpetual swinging pendulum. If the agency is too easy, it gets beat[en] up by Congress and consumer protection groups. Then it becomes too tough, and [it gets complaints] that effective treatments don’t get to market. At the moment we’re swinging toward FDA getting tougher on drugs.” Others have noticed an upswing of safetyrelated rejections. “I was quite taken aback by the number of drugs that didn’t make it for safety reasons. That seems to be well on the increase,” says Richard Hendriks, senior analyst at Nerac in Tolland, Connecticut. “There’s a certain neuroticism [about safety concerns], so it’s understandable that there are going to be fewer approvals. That means that there are populations of patients that could be treated who are going to miss out. At the same time, patients are going to win because [of] the overall safety of the drugs that are coming out.” Along with safety issues, the agency is altering its standards for efficacy. “We’ve been seeing a continuing focus on safety issues, and related to that, a focus on ensuring more robust efficacy. I’m sensing, for example, disenchantment [from FDA] with noninferiority trials [which demonstrate equivalence to a standard therapy],” says Katz. Fong contends that many of the proxies that have been used in the past are no longer going

to be accepted by the agency. For example, in diabetes, some companies have used hemoglobin A1c as a biomarker. (When blood sugar is high for an extended period, it combines with hemoglobin to form hemoglobin A1c.) “But showing an impact on hemoglobin A1c is not sufficient. You have to show some added benefit. The worst case scenario is that [the FDA requires] an outcome study,” Fong says. Companies might become more creative by looking for added benefits, “for instance, if a diabetes drug lowered drug blood pressure. That might be an added benefit that other drugs don’t have,” says Fong. Follow-on biologics debated Last year, the debate over follow-on biologics began to heat up. The discussions were fueled by the authorization for an abbreviated approval pathway for follow-on biologics, created in March as part of Barack Obama’s healthcare reform package. At issue is the fact that even subtle manufacturing changes can influence the immunogenicity of a protein and possibly its efficacy and follow-on biologic manufacturers will not have access to the originator’s clone, cell bank and fermentation and purification protocols. Potential differences between innovator products and follow-on biologics include post-translational modifications, three-dimensional structure and tendencies toward protein aggregation. FDA held a 2-day hearing in November to discuss the issues, and analytical characterization was a hot topic of debate. Some attendees, including Mark McCamish, head of global biopharmaceutical development at Novartis subsidiary Sandoz in Holzkirchen, Germany, argued that current techniques are sophisticated enough to determine if additional clinical trials are needed to confirm biosimilarity under the European regulatory system. But others pointed to the biosimilar epoetin zeta, marketed by Hospira, of Lake Forest, Illinois, which was shown to be similar to its reference product epoetin alpha (Eprex; Amgen/ Johnson & Johnson), but had lower potency in clinical trials1.

Table 1 Biologics approved by the FDA in 2010 Trade name

Generic name

Indication

Actemra

Tocilizumab

Rheumatoid arthritis Humanized mAb

Type of drug

Krystexxa

Pegloticase

Company Genentech

Gout

Recombinant protein Savient

Prolia, Xgeva Denosumab

Osteoporosis/ osteopenia

Fully human mAb

Provenge

Prostate cancer

Dendreon Autologous, engineered, dendritic cell–enriched vaccine

Sipuleucel-T

Amgen

Ruconest

Conestat alpha

Hereditary angiodema Recombinant protein Sanofi-aventis

Vpriv

Velaglucerase alfa

Gaucher’s disease

nature biotechnology volume 29 number 3 march 2011

Recombinant protein Shire Human Genetics

197

NEW S f e at u r e 60

� New molecular entities

Number of drugs approved

53

� Biologics license applications

50 39

40

35 31

30

30

27

24

21

20

10 3

6

7 3

2

5

7

21

18

17

6

18

5 2

4

2010 success stories In addition to a small molecule for multiple sclerosis with a particularly interesting mechanism (Box 1), the year brought to market the first cancer vaccine, a mAb product approved for two different large indications in a single year, a drug for multiple sclerosis that is not a b-interferon, a mAb against arthritis with a mechanism other than tumor necrosis factor (TNF)-a inhibition, and replacemet proteins produced either in rabbits or by gene-activation technology.

19

16

15

2

3

6

6

0 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010

Year

© 2011 Nature America, Inc. All rights reserved.

Figure 1 FDA approvals 1996–2010.

No one is quite certain how FDA will decide which follow-on biologics, if any, will require efficacy trials. The agency is “kind of noodling about whether they’ll require a phase 3 type of study in biosimilars,” says Fong. FDA also has to decide if it will accept results from foreign trials that compare a follow-on molecule to the original. As companies are gaining experience running supporting trials for biosimilars in Europe, it is expected that these human data will help smooth the path when the same companies introduce the same follow-on products in the US. Another question for follow-on products will be the issue of interchangeability. Smallmolecule generics can be listed as interchangeable, which means that pharmacists can automatically substitute the generic for a prescription of the innovator product. Some states in fact require the pharmacist to do so if a physician doesn’t object. The situation for follow-on biologics is less clear. Biosimilars in Europe are not substituted at the level of national healthcare systems; they must be prescribed by a physician (rather than substituted by a pharmacist). “Assuming [interchangeability] means clinical trials —how would you set up trials to measure it?” asks Erika Lietzan, partner at Covington & Burling of Washington, DC. At one end of the debate, according to Lietzan, there are some who are still arguing that unlike European Medicines Agency (EMA), the FDA could agree to equivalence without clinical data. “There are people who say that the analytical techniques are there, that we can fully characterize a [follow-on product] and reach a level of confidence that it would have the same clinical impact no matter who you give it to, and that should be sufficient to meet the standard…[a decision is] a 198

Republican House, there’s going to be oversight that’s negative for industry—either because it goes after industry directly, or in going after FDA, industry becomes collateral damage,” says Katz.

few years down the pike, but this is a big one,” Lietzan says. No consensus emerged from the November meeting, and the agency isn’t likely to provide substantial guidance soon. “They’re not rushing. They want to feel their way a bit. In terms of a substantial guidance, that’s going to take some time,” says Daniel Kracov, a partner and chair of the FDA practice at Arnold and Porter in Washington, DC. Political gains and losses President Barack Obama’s administration is putting its imprint on the agency. “This is really the Obama FDA, and maybe 2010 was the year where we got a full year’s worth of data” on how the agency has changed under his administration, says Lietzan. Among the changes: management is more hesitant to overrule the objections of reviewers. “A line reviewer says, ‘I’m getting quashed, and my science and purity is being undermined by meddlesome management’,” says Katz, and management is backing down more often. “The line folks I think are more comfortable staking out a position and fighting for it, and management is I think a little bit less inclined to overturn them.” “The FDA prides itself on making decisions driven by science, but [under the Bush administration] there was concern that in some instances, science wasn’t in the driver’s seat,” says Katz. Politics figures to weigh heavily in the agency’s near-term future. “I think it’s going to be interesting to see the dynamic between a conservative Republican House of Representatives and the FDA of the Obama administration. One could expect oversight to be more adversarial to the agency, and maybe less adversarial to industry, but even with a conservative

Prolia and Xgeva (denosumab). Amgen’s denosumab was arguably the year’s biggest biotech success story, having gained approval for two different indications. In June, the human mAb was approved for the treatment of osteoporosis in postmenopausal women who are at high risk for fractures, under the trade name Prolia. In November, it was also approved for prevention of bone injuries (fractures, spinal cord compression and severe bone pain in cancer patients with bone metastases), under the trade name Xgeva. The drug operates on the RANK ligand pathway, first discovered by Amgen in the mid-1990s (ref. 2). This pathway plays a key role in the regulation of osteoclasts and osteoblasts, which resorb and build bone, respectively. Osteoclasts secrete an enzyme that breaks down bone and produces a resorption pit, which osteoblasts later fill with a new bone matrix that eventually mineralizes. RANK is a receptor found on the surface of osteoclasts and osteoclast progenitor cells. RANK ligand is expressed by osteoblasts and other cells. Heightened levels of RANK ligand drive excess resorption. Denosumab binds to RANK ligand, preventing it from reaching RANK and driving bone resorption. Ultimately, this nudges the system toward bone formation and away from resorption, which explains its activity in osteoporosis3. In bone metastases, tumor cells secrete cytokines and growth factors that lead to overexpression of the RANK ligand, which leads to excessive osteoclast activity and bone destruction. Bone resorption then leads to release of growth factors from the bone matrix, which can stimulate tumor growth. As much as 75% of individuals with advanced breast or prostate cancer will develop bone metastases, and 50–75% of them will develop severe bone injuries, according to Amgen.

volume 29 number 3 march 2011 nature biotechnology

n e w s f e at u r e

© 2011 Nature America, Inc. All rights reserved.

The current market leader for preventing bone injury is Novartis’s (Basel) Zometa (zoledronic acid). The drug inhibits farnesyl diphosphate synthase in the 3-hydroxy-3methylglutaryl-CoA (HMG-CoA) reductase pathway in bone-resorbing osteoclasts, leading to inactivation of the cells and suppression of bone resorption. It is approved for patients with solid tumors and multiple myeloma, but Xgeva is restricted to solid tumors. Amgen predicts peak sales of $5 billion within five years of approval in the US, Europe, Japan and other markets. Provenge (sipuleucel-T). Dendreon’s Provenge was approved last April for the treatment of asymptomatic or minimally symptomatic metastatic, refractory prostate cancer. It is an immunotherapy in which the patient’s white blood cells are collected and activated with the recombinant fusion protein PA2024, which consists of the prostate antigen prostatic acid phosphatase fused to the immune-cell activator granulocytemacrophage colony-stimulating factor. Provenge first went before the FDA in 2007, when an advisory committee recommended approval. But the agency requested more data to show a survival benefit, and the company provided it at the 2009 American Urological Association meeting, when it showed that the treatment extended life span by a median of 4.1 months and increased 3-year survival by 38% compared to placebo4. The autologous, engineered, dendritic cell– enriched preparation that Provenge comprises has a price tag of $93,000 for a course of three treatments, a figure that has raised eyebrows. The Centers for Medicare Services and Medicaid (CMS) has not yet decided whether to reimburse for Provenge, but last November, a 14-member advisory panel recommended that it do so. If CMS approves reimbursement as expected, Provenge could have yearly sales as high as $2 billion, according to Ziad Bakri, a biotech analyst at Cowen and Company in New York. “The Provenge approval was a landmark decision. It validated all of the work that had been done in the immunotherapy space. It finally showed some benefit in a late-stage clinical trial, and more importantly it affirmed a path for all of the other immunotherapy vaccines that are making their way through the clinic,” says Fong. Actemra (tocilizumab). Genentech’s Actemra, approved in January to treat severe rheumatoid arthritis, is the first approved inhibitor of interleukin (IL)-6. The drug had already been approved in Japan, Australia and Europe, under the trade name RoActemra.

Box 1 A landmark small-molecule approval The multiple sclerosis (MS) market was rocked by the approval of Gilenya (fingolimod), an oral small molecule that will compete with injected interferon-β drugs (Biogen Idec’s Avonex, Bayer AG’s Betaseron and Merck’s Rebif) or Teva’s Copaxone (glatiramer acetate), which is the acetate salt of synthetic polypeptides containing four naturally occurring amino acids (l-glutamic acid, l-alanine, l-lysine and l-tyrosine) found in myelin. Gilenya was approved as a first-line treatment for relapsing forms of MS. Compared with Avonex, one trial showed that patients taking Gilenya relapsed roughly half as often as those taking Avonex (16 versus 33). This small molecule is a sphingosine 1-phosphate receptor modulator. Its exact mechanism is unclear, but it is believed to increase retention of lymphocytes in lymph nodes, thus reducing inflammation by preventing them from joining the assault on the nerve fibers of the central nervous system. The drug is a synthetic derivative of a compound isolated from the natural fungal product myriocin, which was used as a traditional Chinese ‘eternal youth’ medicine8. It was originally tested as an immune suppressor for use in kidney transplants, but a phase 3 trial showed no benefit compared with standard regimens, and some renal side effects9. Safety concerns, including skin cancer, heart problems and infections, led the FDA to approve the drug with REMS. Novartis will also conduct a 5-year post-authorization safety study. A survey of physicians conducted by the investment bank Leerink Swann of Boston indicates that most would prescribe Gilenya, and the analyst group projects that existing drugs will lose 25–30% of their market share to Gilenya. MS patients will also have a new drug to help them walk. Accorda’s Ampyra (dalfampridine, a potassium channel blocker) was approved in January, a first-in-class approval for this indication.

FDA green-lighted the humanized antiIL-6 receptor mAb in individuals who have not responded to TNF-α inhibitors, such as Brussels-based UCB’s Cimzia (certolizumab pegol), New York–based Pfizer’s and Amgen’s Enbrel (etanercept), Abbott Park, Illinois– based Abbott’s Humira (adalimumab), Horsham, Pennsylvania–based Centocor Ortho Biotech’s Remicade (infliximab) and Simponi (golimumab). The causes of rheumatoid arthritis are not fully understood, but many researchers believe that it involves the overproduction of IL-6, which regulates immune response, inflammatory reaction and bone metabolism. Other companies are working on inhibitors of IL-6 or its receptor, including New York’s Bristol-Myers Squibb; Gaithersburg, Maryland’s AstraZeneca/MedImmune; and Bridgewater, New Jersey’s Sanofi-aventis5. Krystexxa (pegloticase). Another landmark approval was that of East Brunswick, New Jersey–based Savient Pharmaceuticals’ Krystexxa, a new treatment for gout. The uricase decorated with polyethylene glycol (PEG) metabolizes the uric acid that builds up into needle-like crystals causing swelling, pain, stiffness, heat and redness in joints or soft tissue. Conventional medications, including xanthine oxidase inhibitors Zyloprim (allopurinol; Prometheus, San Diego) and Uloric (febuxostat; Menarini, Florence, Italy) reduce uric acid levels in the blood. Krystexxa

nature biotechnology volume 29 number 3 march 2011

was shown to reduce uric acid blood levels and deposits in joints and soft tissue. Until recently, no new therapies for gout had been introduced in over 40 years. Krystexxa is the second new therapy, following on the heels of Takeda Pharmaceuticals’ (Osaka, Japan) Uloric, which was approved in 2009. In August, 2009, Krystexxa was the victim of a rare move on the part of the FDA, when the agency ignored the recommendation of an advisory panel that voted 14–1 to approve the drug6. Instead, it rejected Krystexxa on the grounds that the company had altered the manufacturing process after its clinical trials. Savient resubmitted a biological licensing application in 2010. Safety issues remain, as 25% of clinical trial participants experienced a severe allergic reaction to the drug. Physicians are advised to administer a corticosteroid and an antihistamine to minimize these reactions. The drug was approved subject to a REMS program. Ruconest (conestat alfa). In 2010, Parisbased Sanofi-aventis’s Ruconest became the first drug produced in transgenic rabbits to be approved. It was cleared by EMA in October. The Dutch company Pharming, of Leiden, manufactures the drug, which is produced in the animals’ milk. Ruconest, a recombinant version of the human C1 esterase inhibitor protein (C1INH), is approved for hereditary angioedema (HAE), 199

NEW S f e at u r e

Box 2 GM salmon—veterinary drug or human food?

© 2011 Nature America, Inc. All rights reserved.

2011 could be the year that a biologic produced in transgenic salmon finally gets the seal of approval from the FDA as a drug rather than as a food. Developed by AquaBounty Technologies of Waltham, Massachusetts, AquAdvantage is an Atlantic salmon that has been modified to include a Chinook salmon growth hormone gene and a promoter and terminator from ocean pout. The fish grow twice as fast as selectively bred fish, with most of the accelerated growth occurring in the first year of life. The company claims that the fish grow no larger than wild salmon and are all female with up to 99.8% triploid, making them sterile. To the surprise of some, the agency is regulating the product as a veterinary drug, as it considers the construct to be the equivalent of a drug that is used to treat the fish. Furthermore, officials decided that the fish was not materially changed by the gene sequence, so it need not be regulated as a food. The FDA has taken criticism because it included no food experts on the advisory panel10. AquaBounty was formed in 1989 and the company first applied for FDA approval in 1995. But it wasn’t until 2008 that the agency released draft guidance for genetically engineered animals. Concerns over food safety and environmental impact, should some fish escape into the wild, have been brewing since the fish were first developed.

which is characterized by acute attacks of swelling in the skin, intestine, mouth and throat. C1INH regulates several inflammatory pathways. Deficiency of the protein leads to overactivation of the complement system and other pathways, which in turn leads to anigioedema. HAE patients carry a mutation in the gene encoding C1INH, which leads to deficiency. It is the second transgenic animal–produced drug to be approved, following Atryn, a recombinant anti-thrombin-α produced in goat’s milk. Framingham, Massachusetts– based GTC Biotherapeutics received EMA approval in 2006 and an FDA nod in 2009. Vpriv (velaglucerase alfa). Though not a novel drug, Basingstoke, UK–based Shire’s recombinant human glucocerebrosidase—an enzyme replacement therapy for Gaucher’s disease patients, who have lipid deposits in various organs due to an inability to produce the enzyme—is produced by means of a different mechanism from its competitor, Genzyme’s Cerezyme (imiglucerase). Vpriv is produced by gene activation, a technology developed by TKT (now part of Shire) that had been the subject of litigation for decades7. Amgen and Genzyme, both producers of recombinant human proteins, tried to lay claim to the

200

intellectual property rights for manufacturing human proteins in human cell lines. Vpriv got the nod from both the EMA and the FDA in 2010, which was welcome news to Gaucher’s patients, whose supply of Cerezyme was limited owing to manufacturing problems. What’s ahead in 2011? Two biologics are creating a buzz as they move toward possible FDA approval in 2011. Benylsta (belimumab), manufactured by Human Genome Sciences of Rockville, Maryland, could be the first new lupus drug in over 50 years. The human mAb targets B-lymphocyte stimulator, which when elevated boosts production of autoantibodies in lupus, rheumatoid arthritis and some other autoimmune diseases. Lupus treatment has come a long way since the last drug, Sanofi-aventis’s Plaquenil (hydroxychloroquine), was approved in 1955 for the treatment of malaria and lupus, with vastly improved survival among patients over the years. But those improvements have come as a result of off-label use of drugs approved for other indications. Last June, Human Genome Sciences submitted regulatory applications to both FDA and

EMA, and the drug received priority review status from the US agency. A November advisory panel recommended it be approved. Its original Prescription Drug User Fee Act (PDUFA) date was last December, but FDA pushed it back to this month with no explanation. “It’s not just that it’s novel to approve an antibody for that patient population, it’s a tremendous area of unmet need. This is the first new application for that indication in quite some time. For 2011, all eyes are on Benlysta,” says Fong. Another potential breakthrough drug is brentuximab vedotin, co-developed by Seattle Genetics of Bothell, Washington, and Millennium’s Takeda Oncology Company of Cambridge, Massachusetts. The drug is an antibody-drug conjugate that targets the Hodgkin’s lymphoma marker CD30. The mAb is linked to the anti-cancer drug monomethyl auristatin by a linker that is stable in the bloodstream but releases the drug, once internalized, into tumor cells. Data released at the American Society of Hematology meeting last December showed that 94% of 102 lymphoma patients treated with the antibody conjugate in the trial saw at least some reduction of their cancer. 2011 is the year of the rabbit, and the year that the first rabbit-produced drug will be marketed. It might also be the year that genetically modified salmon leap onto the market after a decade-long review by the FDA (Box 2). Jim Kling, Bellingham, Washington

1. Schellekens, H. Drug Discov. Today 14, 495–499 (2009). 2. Simonet, W.S. et al. Cell 89, 309–319 (1997). 3. Elvidge, S. Nat. Biotechnol. 29, 4 (2011). 4. Kantoff, P.W. et al. N. Engl. J. Med. 363, 411–422 (2010). 5. Anonymous. Business Wire, 25 November 2010. 6. Pollock, A. The New York Times, 2 August 2009. 7. Moran, N. Nat. Biotechnol. 28, 1139–1140 (2010). 8. Adachi, K. & Chiba, K. Perspect. Medicin. Chem. 1, 11–23 (2007). 9. Brinkmann, V. et al. Nat. Rev. Drug Discov. 9, 883–897 (2010). 10. Fox, J. Nat. Biotechnol. 28, 1141–1142 (2010).

volume 29 number 3 march 2011 nature biotechnology

n e w s f e at u r e

The power of many

In December, Life Technologies announced a $7 million crowdsourcing initiative, the Life Grand Challenges Contest, to focus on its newly acquired Ion Personal Genome Machine (PGM). The sequencing technology, pioneered by the Guilford, Connecticut–based startup Ion Torrent, costs $50,000 to buy and can sequence a sample at a cost of $500 in just two hours. But that is apparently not good enough for Jonathan Rothberg, founder and CEO of Ion Torrent. The first three $1-million challenges in the contest ask innovators to devise ways to make Ion Torrent’s technology even faster, cheaper and more accurate. Life Technology’s competition is just one example of ‘crowdsourcing’, a portmanteau that refers to outsourcing tasks usually performed by people closely connected with an institution to a ‘crowd’ of people outside the institution. It differs from other types of open innovation in that members of the crowd generally expect some kind of incentive or reward—sometimes, as in the Ion Torrent case, a large one. It differs in turn from more traditional outsourcing in that the task is handed over to a large group with disparate skills and expertise, rather than to a single highly skilled individual or team. In terms of applications, particularly in the commercial biotech arena, the approach is still in its infancy. As Joanna Chataway, director of the innovation and policy team at the European branch of the global not-for-profit organization RAND, puts it: “We have seen plenty of anecdotal evidence that crowdsourcing can work, but there has been little research into how and where it works best.” Already, though, a scattering of startups and spinouts have coalesced around the approach. Passive and active crowdsourcing The history of crowdsourcing is short, the term itself having been coined (by Jeff Howe, writing in Wired magazine) less than five years ago. In one form, it relies on a simple idea. At any time, many of the billion or so personal computers (PCs) in use worldwide are idle, or engaged in mundane tasks, such as word processing. This represents a vast, largely untapped resource of processor power. Small-scale attempts to use home PCs in research first took place in the mid-1990s, but it was the SETI@home project, launched in 1999, that first caught the

iStockphoto

© 2011 Nature America, Inc. All rights reserved.

Applications of crowdsourcing in commercial biotech remain few and far between, but the approach is proving increasingly popular for solving challenges in basic research. Clare Sansom reports.

Crowdsourcing started out by enlisting underutilized computer time, but has morphed into tapping into the brain power of people from all over the world.

PC-using public’s imagination. This was an attempt to discern the presence of extraterrestrial intelligence by analyzing radio signals from space; it was the brainchild of computer scientist David Anderson and his student David Gedye at the University of California, Berkeley. The project had about a million active users at its peak in 2000 and is still going today. (No signs of extraterrestrial intelligence have been discerned yet.) A fraction of this unused PC power is now utilized for research and development through the so-called ‘volunteer computing’ movement. This can also be thought of as a passive type of crowdsourcing. With few exceptions, volunteers donate only their ‘spare’ processor power and are rewarded with a small stake in a research project. Another incarnation of the approach—active crowdsourcing—involves the participation of users and has had some success in the commercial biotech sector. Among life science companies, pharma giant Eli Lilly of Indianapolis has been a leader in the field of internet-led open innovation for the past decade. Spinning out crowdsourcing In 2000, Sidney Junell, then head of Lilly, organized a group of executives to explore new ways of working in an increasingly fluid and internetconnected business environment. No fewer than three successful open innovation companies— InnoCentive (http://www2.innocentive.com/), based in Waltham, Massachusetts, YourEncore (http://www.yourencore.com/), in Indianapolis

nature biotechnology volume 29 number 3 march 2011

and Cincinnati, and Collaborative Drug Discovery (http://www.collaborativedrug. com/), based in Burlingame, California—sprang from these discussions. InnoCentive’s distinctive business model involves using the internet to connect the seekers—companies with technically challenging research or management problems—with individuals and groups who may be able to solve those problems. The basic details of each challenge are posted publicly on InnoCentive’s website. Any individual may register as a solver. “We have solvers from all walks of life,” says Alph Bingham, founder of InnoCentive. “We’re trying to find Archimedes just as he lifts himself into the tub, and he may be anywhere.” Solvers pay no fees, but must formally register for a challenge before they receive the full, confidential outline of the project; seekers, on the other hand, pay to register on the site, and again to register each challenge. If the problem is solved, rewards are paid to one or more successful solvers out of this fee. Since 2001, InnoCentive has posted about 12,000 challenges on its site, and a reward has been paid in about half those cases; there are over 200,000 registered solvers and over 150 companies registered as seekers. The number of solutions received varies enormously among challenges, from a handful to hundreds. Seekers, solvers and challenges may come from any discipline, including management and design. The largest numbers of each, however, are found in chemistry and the life sciences, and InnoCentive’s very first seeker was a pharmaceutical company (not surprisingly, Lilly itself). Challenges are presented anonymously; companies can be reluctant to disclose their identity, even when they are satisfied with their results, as this might disclose more than they intend of their future plans. Few allow the solutions to challenges to be published, even informally. Bingham often quotes the experience of an anonymous pharmaceutical company that posted a challenge to find a synthetic route to a new cardiovascular drug. “A German solver came up with a novel synthesis that produced a high yield of pure compound, cheaply, and with no toxic by-products. The company is now taking their drug through phase 2 clinical trials,” he says. In 2009, Nature Biotechnology’s publisher, Nature Publishing Group (NPG), established a partnership with InnoCentive, launching the Open Innovation Pavilion that provides access for nature.com registrants to the InnoCentive challenges (http://www.nature.com/openinnovation/ index.html). “Seekers choose whether they wish to post their challenges on the Open Innovation Pavilion as well as on InnoCentive’s site, and so 201

© 2011 Nature America, Inc. All rights reserved.

NEWS f e at u r e take advantage of the talent and expertise of our readership,” says Veronique Kiermer, executive editor at NPG and the first manager of the partnership. ~10% of all solvers who submit solutions to InnoCentive have through nature.com. The InnoCentive platform has been offered to not-for-profit organizations, sometimes with sponsorship from NPG. For example, Kiermer approached Robert Don, director of the Geneva-based, not-for-profit Drugs for Neglected Diseases Initiative, to see if he would be interested in participating in an NPGsponsored challenge. Don says that he “discussed several options with InnoCentive and NPG, and fixed on a challenge: a list of validated drug targets for one tropical infectious disease.” This challenge produced a “healthy number” of solutions, from simple lists of proteins to comprehensive literature reviews. The second spinout from Lilly’s exploration of open innovations, YourEncore, also links individuals and companies to solve research and business challenges, but uses a model that is closer to traditional outsourcing. “We maintain a network of skilled and experienced retired or semiretired executives, and match these to the specific needs of companies we work with,” says Lisa Arbogast, director of life sciences at YourEncore. Experts are paid as consultants, generally on a daily or hourly basis, and YourEncore manages the paperwork. Companies can engage experts quickly, with minimal bureaucracy, and only employ them for as long as they are needed. In many sectors, including life sciences, the economic downturn has increased the numbers of “post-corporate, professionally active” individuals interested in this way of working. Similar to YourEncore, yet another Lilly spinout, Collaborative Drug Discovery, harnesses the web to promote collaborations between professionals and institutions. It provides a range of database solutions for scientists who wish to combine and analyze preclinical drug discovery data. These offer different levels of security, ranging from the CDD Vault where access is limited to a single group, through CDD Collaborate, which enables selective data sharing between collaborators in different companies, to CDD Public—a repository for drug discovery data that have been released into the public domain. Any interested company, notfor-profit organization or individual may log into the public vault and mine the available data. GlaxoSmithKline, for example, has used the Collaborative Drug Discovery platform, alongside the Hinxton, UK–based European Bioinformatics Institute and the US National Library of Medicine, to publicly release data on more than 13,500 compounds potentially valuable as antimalarial drugs. 202

Patients at the helm The stereotype of the internet-enabled expert patient is commonly cited anecdotally and in the medical literature. Some companies have begun to harness the knowledge and connectivity of patient communities, using patients as experts similar to the way YourEncore uses technical experts. PatientsLikeMe (http://www. patientslikeme.com/), based in Cambridge, Massachusetts, was cofounded in 2004 by Ben and Jamie Heywood, who were inspired by the difficulties experienced by their brother Stephen, diagnosed with amyotrophic lateral sclerosis, in the conventional healthcare system. The site allows patients to share details of symptoms and treatments with each other as well as with the research and medical communities. Patients receive no monetary reward; their reward is in learning more about their condition from the experiences of thousands of other patients. PatientsLikeMe now has more than 80,000 members. Another company that accumulates data from its customers through crowdsourcing is Mountain View, California–based company 23andMe. Customers of the personal genomics startup who submit samples of their saliva for genotyping have the opportunity to take part in surveys, which, when combined with their genetic information, can provide useful information to the wider group about genetic linkage. The company recently published a paper describing novel genes linked to some common traits1. This approach becomes more powerful still when genetic data are combined with contributions from patients. In what is, so far, a unique collaboration, 23andMe has partnered with PatientsLikeMe and the Michael J. Fox Foundation to collect genetic data from individuals with Parkinson’s disease. “The fee for genotyping Parkinson’s patients has been subsidized to only $25, which is a fraction of our normal charge,” says Nick Eriksson, a statistical geneticist at 23andMe. “We have about 4,000 patients genotyped so far, and we expect to be able to announce some novel genetic variants associated with increased risk of the disease very soon. With 80% of our patients contributing detailed information about their health and treatment, we are beginning to study genetic variations that can stratify the disease into subtypes.” Sharing potentially sensitive health-related data does, however, raise critical questions for companies working in this area. “Companies are required under law to keep personal data secure and depersonalized, and to keep subjects aware of exactly what their data will be used for,” says Nathan Lea of the Centre for Health Informatics and Multi-professional

Education at University College London, UK. “Particular difficulties may be raised if a company based in one country collects data from citizens of other countries.” Volunteer computing The first volunteer computing projects in life sciences were Screensaver Lifesaver and Folding@Home, both launched in 2000. Folding@Home, initially developed in Vijay Pande’s group at Stanford University, California, models the thermodynamics of protein folding. Still at the forefront of technical development in distributed computing technology, it has over 350,000 users. Screensaver Lifesaver involved a partnership between Graham Richards’ group at the University of Oxford, UK, and distributed computing company United Devices (now part of Univa UD), based in Austin, Texas. Volunteers—eventually three and a half million of them—set up United Devices’ software on their desktops and used it to run molecular modeling simulations, docking potential ligands into the binding sites of known drug targets for various diseases. “Using volunteers validated the concept that massively parallel processing could remove resource restraints on computational drug discovery,” says Richards. Nevertheless, the approach did generate its own challenges, not least in intellectual property (IP). “We were searching for lead compounds for potential blockbuster drugs,” says Richards. “These might come from any PC, anywhere on the planet.” Richards and his co-workers spun out an Oxford-based company, InhibOx, to further commercialize technology developed by the volunteer platform and to exploit the IP from any hits generated by the screensaver project. Until the end of 2009, the company was focused on methods development in ligand docking and virtual screening, using parallel processing on in-house and commercial clouds rather than a network of volunteers. “Our methods development started with the screensaver project and would not have been possible without it,” says Paul Davie, CEO of InhibOx. Interestingly, Davie—whose background includes database security as well as computational chemistry—is largely unconcerned about security issues with cloud computing. “Companies like Amazon and Google have security infrastructures that are at least as sophisticated as any big pharma company,” he says. Given the IP difficulties generated by the volunteer computing model, not surprisingly it has been embraced largely by the academic and notfor-profit sectors. Since 2002, Anderson has been funded by the US National Science Foundation to develop the software used in SETI@home into a generic open interface for network-based

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

n e w s f e at u r e grid computing, known as Berkeley Open Infrastructure for Network Computing (BOINC). There are variants of the BOINC code for the PC, Mac and Linux platforms, and for some quasi-computers, such as PlayStation games consoles; Anderson is interested in developing versions that will run on mobile devices. Users, many of whom donate time on more than one PC, install the BOINC client software and then choose which projects to donate time to and how that time should be divided up. These encompass a range of research topics, with life sciences projects—mainly in computational biophysics—among the most popular. “Volunteer computing can be useful in almost every area of science,” says Anderson. Some projects are purely curiosity-driven, whereas others have practical, beneficial aims: some of the latter, particularly in drug discovery, might even be termed precompetitive. Many users are undoubtedly motivated by projects’ humanitarian aims. Anderson, however, does not believe that this is the main reason for project choice. “Projects where the developers make an effort to interact with their volunteers, for example, through informative, up-to-date websites, tend to attract the most volunteers,” says Anderson. Over two million users have signed up to BOINC projects, donating time on over six million machines. The most popular BOINC-based life sciences application, with over 300,000 users, is the protein folding project Rosetta@home. This was set up by David Baker the author of the Rosetta program for ab initio protein structure prediction and his colleagues at the University of Washington, Seattle. Rosetta@home harnesses the power of hundreds of thousands of computers to run Rosetta in a massively parallel way. “Even though many of the computers running our code are small, old or turned on only a fraction of the time, the volunteer network offers us computing power that no public institute could imagine installing in-house,” says Baker. The code has recently been adapted so it can be used to design proteins with novel functions. “We could in principle design an enzyme to catalyze any chemical reaction,” he says. Gianni De Fabritiis at Universitat Pompeu Fabra in Barcelona, Spain, runs a volunteer computing project with relevance for drug discovery. Blockage of the cardiac potassium channel encoded by ERG (the human ether a-go-go-related gene) can cause the disorder known as long QT syndrome, a rare cause of sudden cardiac death. This protein is therefore an important ‘anti-target’ in drug development. Molecules that block this channel must be removed from the drug discovery pipeline as early as possible, preferably in silico. De Fabritiis and his colleagues are using parallel molecular dynamics to simulate drug binding to this

channel. “We hope to work with companies to develop a fast hERG channel drug screening tool,” he says. “We plan to develop a methodology that can be applied in industry by deploying the code on an in-house cluster.” Volunteer computing is also having an impact in the area of neglected disease. Malariacontrol. net (http://www.malariacontrol.net/) is a project run from the Swiss Tropical and Public Health Institute in Basel to model malaria epidemiology. It was set up by Nicolas Maire, the institute’s project leader in scientific computing. “I was recruited in 2003, initially to model the impact of a malaria vaccine then in phase 2b clinical trials,” he says. “When the project was broadened out to compare many interventions, we realized that we had insufficient in-house computing power and using volunteers seemed an ideal approach.” In collaboration with the University of Geneva and CERN, he set up a project that now has ~70,000 users. Corporate philanthropy Many academic and not-for-profit organizations have research projects that would benefit from using volunteers, but lack the resources or programming skills needed to set them up. The World Community Grid (WCG), part of Armonk, New York–based IBM’s corporate social responsibility program, provides exactly this support to humanitarian research projects worldwide. Computer professionals from IBM help scientists design and port their code to the BOINC platform. A third of the ~600,000 volunteers registered on WCG projects are themselves IBM staff, and many IBM employees also loan time on their in-house computers for more commercial applications. One of the most popular WCG projects, screening candidate drugs against HIV proteins, has so far used the equivalent of 122,000 years of single-processor CPU (central processing unit) time. “We called this project FightAIDS@home, a name that is more accessible to the general public and therefore probably attracts more volunteers than ‘antiretroviral drug screening’ would,” says IBM distinguished engineer Joe Jasinski. The Citizen Cyberscience Centre, based in Geneva, was set up in 2009 to help scientists in developing countries take full advantage of the opportunities offered by volunteer computing. Its founding partners were CERN, the UN Institute of Training and Research, and the University of Geneva. “Our work is complementary to the World Community Grid,” says Francis Grey, center coordinator. “While IBM helps with the technical side of developing applications, we promote the concept and provide ongoing support and training.” Last September, the center organized

nature biotechnology volume 29 number 3 march 2011

the first Citizen Cyberscience Summit at King’s College, London, bringing together researchers, computer scientists and interested citizens, many of whom were committed volunteers. All participants, including volunteers, helped create a “Manifesto for Citizen Cyberscience.” Game on The involvement of volunteers in the summit’s agenda indicates a link between volunteer computing and open innovation. The Foldit project, a protein-folding computer game developed from Baker’s Rosetta algorithms, further combines volunteer computing and volunteer thinking. “Foldit is an interactive version of Rosetta@ home, in which users can influence the way the calculations go,” says Baker. “Players use the time and energy they might spend on traditional computer games on solving a ‘real world’ scientific problem with potential humanitarian and commercial value.” Although successful folders cannot achieve the monetary rewards of InnoCentive’s or YourEncore’s solvers, many are motivated by progression up a ‘leader board’. And they all have a small share in the traditional rewards of scientific research. The last author of a Nature paper describing the project2 is recorded in MedLine as “F. Players”: in other words, ~57,000 Foldit players. Elsewhere, EteRNA, a joint project of Carnegie Mellon University in Pittsburgh, Pennsylvania, and Stanford University in Stanford, California, enlists game players to design RNA molecules, which are then tested in the laboratory for their ability to assemble. Intended for the nonexpert, it has training exercises and puzzles for developing the needed skills. The final arbiter is Mother Nature: the design is either built in the laboratory of a founder, Rhiju Das, a professor of biochemistry at Stanford, or is compared with known RNA sequences. These disparate examples show how what has been described as “the wisdom of crowds” is already leading to innovation in bioscience and medical research and, perhaps less often, in biotech. James Surowiecki, author of an influential book of the same name3, claims that a “wise” (and therefore effective) crowd is one that is diverse, independent and decentralized. These characteristics can be seen, at least to some extent, in, for example, InnoCentive’s network of solvers, PatientsLikeMe’s expert patients and the Foldit player community. Questions of data security and appropriate IP models may remain, but these ideas promise to spread further through the sector. Clare Sansom, London & Cambridge

1. Eriksson, N. et al. PLoS Genet. 6, e1000993 (2010). 2. Cooper, S. et al. Nature 466, 756–760 (2010). 3. Surowiecki, J. The Wisdom of Crowds (Doubleday, New York, USA, 2004).

203

building a business

Divining the path to a successful European exit Håkon Kirkeby Buch, Anna C Gustafsson, Viktor Drvota & Carl Johan Sundberg A gathering of biotech leaders in Sweden attempts to share candid opinions on the most successful strategies for building a biotech in Europe.

© 2011 Nature America, Inc. All rights reserved.

A

recent report in Harvard Business Review1 describes the global venture capital industry as in a state of distress: delivering too little, too late to investors and, at the same time, losing dominance over its most valuable asset— entrepreneurs and their innovations. The venture-backed life science industry, notoriously known for its sky-high development costs and heavy regulation, is no exception. In this context, SEB Venture Capital and Karolinska Institutet gathered 80 industry leaders in March 2010 for a ‘Life Science Exit Seminar—How to Promote and Exit Venture Backed Companies’. The seminar placed an emphasis on examining exits from European-based, venture-backed life science companies, which have become increasingly geographically disadvantaged by a US-dominated transaction market. The participants of the symposium included current and former CEOs, venture capitalists (VCs), advisors and representatives from multinational industrial buyers. In the following article, we summarize the main themes emerging from the meeting’s discussions and best practices for building a European biotech business and achieving a successful exit. Straightforward exits are scarce At our seminar, the general question was posed, “At what development stage is it optimal to exit life science companies?” The text-book answer is that if your product can Håkon Kirkeby Buch is an analyst, Anna Gustafsson is an investment manager and Viktor Drvota is head of life science investments at SEB Venture Capital in Stockholm, Sweden. Carl Johan Sundberg leads the Unit for Bioentrepreneurship at Department of Learning, Informatics, Management and Ethics at Karolinska Institutet in Stockholm, Sweden. e-mail: [email protected]

Exit window 2: Profitable company Value Exit window 1: Strategic asset

Revenue Investment capital

Development timeline

~10 years Stage of development:

Proof of concept

Proof of business

Profitability

Figure 1 Exit windows in a company’s life cycle. Exit window 1 is characterized by early acquisition of strategically important assets to a buyer. Exit window 2 is typically the revenuebased trade sale or public listing of a sustainable company. Venture-financed development from startup to the second exit window generally falls outside the scope of most venture capitalist mandates in the case of pharmaceuticals, except for orphan drugs. The dotted lines mean failure due to overspending and/or inability to become profitable.

transform a medical practice, someone will buy it for a lot of money before you have made one sale. Another typical answer, of course, is that if your company has scalable products turning over significant sales and bringing in sizable annual revenues, you will have no trouble finding a buyer and thus an exit. But such examples in the European biotech sector are scarce. Although it is true that new products are always in demand, it is also true that big pharma and biotech buyers are more sophisticated than they once were. On top of that, regulatory requirements are driving up development costs in both biotech/pharma and medical technology. Also, buyers have become very astute in determining the cost base of targets and are especially mindful of how much money further development is going to cost. Recent data now suggest it takes, on average, more than 15 years and $1 billion to develop a blockbuster drug through to regulatory approval2. Likewise,

nature biotechnology volume 29 number 3 march 2011

taking new medical technology products to market has become an exhausting exercise for both entrepreneurs and their investors. Some speakers at the seminar thought that exits must come after developing the technology for just a few years, and that these exits are basically an early trade sale of a pre-revenue asset, a so-called ‘tech sale.’ If that route is not taken, then you must be willing to build a sustainable company, enabling a cash flow–based exit, which generally takes 15–20 years. In other words, in Europe and elsewhere, bringing new life science innovations all the way from the idea stage to sustainable, initial public offering (IPO)–ready companies funded on venture capital alone might take a long time and require substantial resources (Fig. 1). Of course, not every successful exit will include all steps of the classic life science venture capital investment scheme, which is invest early, develop product(s), obtain regulatory approval, generate revenue growth, become 205

building a business

Table 1 Challenges and actions for venture capital syndicates Keeping consensus on exit strategy often proves difficult because the priorities of the various venture capital funds comprising your syndicate change over time. To head off problems in differing expectations of syndicate members, it may be necessary to take preemptive action. The types of different actions you should consider are listed. Challenge

Actions

Loss of business momentuma due to disagreements over corporate strategy

Leave owner discussions outside the company board room. Ensure that board and management focus is placed on core business, not investor agenda. Evaluate board composition and replace members to better reflect the new situation of the company.

Financial distress necessitating internal down round with severe dilution of nonparticipating owners

Mitigate pressure from the leaving investor that might force an early exit for all investors. Evaluate if a smaller investor base is sufficient to continue operations and adjust business plan accordingly.

© 2011 Nature America, Inc. All rights reserved.

aBusiness momentum is defined as partnering dialog at an advanced stage, sales taking off, proximity to a new external financing round and so on.

profitable and then exit. Because of skyrocketing development costs and time-to-market for new products in large indication areas such as diabetes and cardiovascular disease in recent years, the predominant type of successful exit has come from companies producing or promoting orphan drugs or delivery devices. These companies generally realized more successful exits than biopharmaceutical companies because the regulatory requirements for product approval are less stringent and faster and their market exclusivity longer. The hard truth, though, is that the ideal time to exit always depends on the particular situation, and generally you will be required to have some sort of proof of concept before you can consider a trade sale. Beyond proof of concept, valuation multiples tend to be closely linked to the development phase of the companies. Those that have achieved proof of business, profitability, or possibly even market leadership achieve higher valuations than those with only proof of concept. Moreover, younger companies all over the globe are very much dependent on how strategically important their particular asset is to the buyer to achieve good valuations3. Competition helps One way to help your valuation, no matter where you are located, is to build a real or perceived auction dynamic for your product. One of the seminar’s attendees, Otello Stampacchia of London-based Omega Funds, commented. “It is critical to create a perception among potential partners and buyers that there is someone else out there who really wants your product. Otherwise, you might get a ‘decent’ exit, but you will not get a home-run.” Besides having several suitors, a way to achieve that perception of competing forces is building a sustainable company because that gives you the flexibility to either continue business on your own, to out-license your 206

product, to do a trade sale or even an IPO. Those options mean you are not at the mercy of any one particular bidder. Indeed, a recent trade sale involved 20 initial offers, with five being invited to closing negotiations. Still, an IPO was kept open to the very end, with a prospectus filed four days before closing the trade sale. That helped keep up the momentum of the process. But while it is best to be prepared for all exit opportunities4, be aware that doing so can be both expensive and time consuming. For example, when referring to a recent exit, Stockholm-based investor Thomas Eklund noted, “In hindsight, preparing for an IPO was not worthwhile, given all the money and time it required. But then again, selling this company in Europe was quite simple since the story was well known. If it had ended up in the US, the value of IPO preparations might have been different.” Get out there Measured against other industries, VCs and their portfolio companies actually spend relatively little resources on selling their companies (their ‘products,’ in this case) compared with how much they invest in sourcing and developing them. That begs the question: Are venture-backed life science companies spending too little time and money on the exit process? Perhaps. One way to fight against this is to redefine what ‘business development’ is at your company. It should mean being ‘out there’ all the time—in other words, understanding where the big market players are heading, where medical practice is going, what happens at the industry meetings and who key opinion leaders are. You should also know who makes decisions about what assets are to be acquired by the industry. It is not enough to hire the best business developer you can find, and let him or her do their job. As a CEO, or otherwise a

member of the executive team, you need to be aware of these issues yourself. In general, skilled senior management with extensive international networks and repeated experience with building companies has proven much harder to find in Europe than in the US. In a trade sale–dominated exit environment, companies that are able to put themselves on the strategic agendas of as many potential buyers as possible have the largest chance of success. Not only do managements need to be committed to business development to achieve this, they must start their business development activities at an early stage. If you are in Europe, you need to visit the US and define your potential buyers. Spend the time to engage the decision makers at the companies that might purchase you. You cannot call up potential buyers the moment you want to sell and expect to get results. Another tip: use investment bankers to assist you. That doesn’t mean you won’t have to identify the buyers and attract their interest— you will—but you cannot manage the whole process yourself. Besides, advisors tend to add value to the final price. Align the syndicate You will also need a consensus on your exit strategy with the syndicate of investors. If you have some investors in love with the science, but others backing out, you will be in trouble. Or, when investors come in at differing times, the board of directors might lose control of where the company is going. Both might prove disastrous when planning for an exit. It is possible that management is conducting business as usual, which might not be what is best to achieve an optimal exit. That such issues arise, however, seems more like the norm than the exception because VCs tend to change priorities over time. What was a perfectly aligned syndicate six months ago might not be so today. If your syndicate fractures, you will need a plan (Table 1). Be prepared to put a lot of time and effort into it. VCs know that syndicates likely will change, and therefore engage all involved parties in a discussion on exit before making the investment. This allows them to check if there is a fundamental alignment of interest before investment happens. There are other issues of misalignment of management and boards, and in particular there is often conflict when shifting from a scientifically oriented leadership typical for startup companies to an industrially oriented management5. Mainly, this happens because entrepreneurs are characterized by curiosity, creativity, exploration, improvisation and energy. The industry side of things brings

volume 29 number 3 march 2011 nature biotechnology

building a business

© 2011 Nature America, Inc. All rights reserved.

structure, regulation, standardization, documentation, validation, legislation and commercialization. An entrepreneur often needs to be forced into a pre-defined, squared shape as the company grows. Our seminar participants warned about the dangers of attempting to force an exit strategy on a scientifically focused management. Ultimately, management will do what it believes in, and if that is not aligned with the board, you will have a major issue to resolve. US presence Whereas it is clear that proof of concept and proof of business are important for life science startups to be attractive targets, to what degree must European companies be approved and present in the United States to be attractive to US buyers? The answer varies. Sometimes, establishing a foothold in the home market is enough. One attendee at the symposium, former NeoPharma CEO Ulf Rosen, stated, “Our focus was to achieve a relatively high level of penetration on a small number of clinics in our home market. Then, in discussions with international buyers, we argued that if we could reach 8% penetration with our resources, a global organization should at least be able to do the same on the world market. In North America alone, this corresponds to $150 million in revenues.” You might also move your headquarters to the United States a few years before selling, while perhaps keeping clinical operations in the home country. The US team can focus on further development of the product and start preparations for an IPO with US investment banks, if needed. The idea would be to show that the product can be sold in Europe, and thus it’s easy for US buyers to extrapolate that into their market. After all, medical practice in the US and Europe is more or less the same. Although it is true that the United States dominates the transaction market, in principle geographical localization should not matter— a great product is great no matter where it is based. Even so, European biotech companies tend to interact with certain people in certain places within certain contexts, while in the United States, companies interact with a larger community of potential buyers and key opinion leaders on a much more constant basis. That is a massive competitive advantage that the management in a European startup needs

Off-business cycle peak

Investment by aligned syndicate, early or late

Startup company, based on no-brain-er treatment concept

In home market, with the world market in mind

Proof of concept

Proof of business ~10 years

Profitability (initial public offering (IPO)) Development timeline

(Try to) hit bull market Exit (window 2)

Align management with exit strategy Global business development work Focus on cost and profitability Build a sustainable company Exit (window 1)

Line up trade sale buyers IPO preparations?

Figure 2 Mapping the path.

to counteract. This can be done by traveling to US events often and also by establishing a very well-connected network of key opinion leaders and scientific advisors. But having their names associated with your company and listed on your website is not enough—you will need to make them visit your company, make them use your products. In other words, make them the best advocate for your product or company. The uncertainties associated with economic cycles are clearly imperative for venture backed exits, and best practice for exit promotion will need to be moderated for the general economic climate. Going public during a downturn is rarely a viable option and merger and acquisition opportunities will be reduced. Nonetheless, highly successful transactions have taken place in the midst of the recent financial meltdown. The difference there is that those companies were great, not just good. It is very difficult to plan for economic cycles, but if you’re planning on exiting when everyone else is, you can consider a higher payout. When transactions are down and the markets shut, certainly expect that to be reflected in your valuations and pricings. Conclusions Successful entrepreneurship in venture backed life science is founded on possessing the science and technology (as well as the adequate legal rights) to fill a clear market need with novel products. However, the level of development costs required to bring many drugs and devices to market limits the investment cases that fit the mandate of many EU venture

capital firms. The discussions of successful exits at our seminar focused on either trade sales at an early stage of development or late investments. In particular, the seminar concluded that orphan drugs and devices are a particularly suitable product class for venture investments in Europe. More than management excellence, scientific foundation and marketability, our seminar pointed out that a competitive sentiment among buyers is a necessary value driver to obtain premium returns from life science exits. Many factors come into play achieving this, such as the early initiation of extensive and international business development efforts directed toward exit, the ability to build a sustainable company and the alignment of investors and management on exit strategy. International business development activities were identified to be of particular importance for European companies. On the contrary, achieving regulatory approval and established business outside domestic markets were not preemptory for successful exits, with the exception of companies striving to achieve revenue based exits or reach the public markets (Fig. 2). 1. Ghalbouni, J. & Rouziès, D. Harvard Business Review 18, 21–23 (2010). 2. Kaitin, K.I. (ed). The Tufts CSDD Outlook 2010 (Tufts Center for the Study of Drug Development, Medford, MA, 2010). 3. Stewart, J.J. & Bonifant B. Nat. Biotechnol. 28, 178 (2010) 4. Buckel, P., Dauer, U., Frei, P. & Nothias, J. Nat. Biotechnol. 25, 1195–1197 (2006) 5. De Rubertis, F., Fleck, R. & Lanthaler, W. Nat. Biotechnol. 27, 595–597 (2009).

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 29 number 3 march 2011

207

correspondence

© 2011 Nature America, Inc. All rights reserved.

Strengths and limitations of the federal guidance on synthetic DNA To the Editor: The December issue included a report summarizing the first reactions of the gene synthesis industry to the publication of the US government Screening Framework Guidance for Providers of Synthetic DoubleStranded DNA1. Some of the questions raised by the federal guidance had already been exposed in your columns2,3, but none of these previous comments relied on a bioinformatics analysis of the screening protocol proposed by the US government. Here we present the preliminary results of an implementation of this protocol with the hope of documenting the strengths and limitations of the federal guidance. This document outlines a minimal DNA sequence screening protocol that providers of gene synthesis4 services are encouraged to follow before fulfilling an order. The objective of the protocol is to identify sequences of concern of any length that are specific to ‘select agents or toxins’ (SAT) listed on the National Select Agent Registry (http://www.selectagents.gov/). It starts by translating the nucleotide sequence ordered by the customers into each of six possible reading frames. Both the nucleotide and amino acid sequences must then be divided into fragments that are individually aligned against GenBank using a local sequence alignment algorithm. Alignment results are interpreted using the ‘best match’ criterion, a procedure designed to identify sequences specific to SATs without relying on a curated database of sequences of concern. Although the federal guidance gives a general method for the automatic identification of potentially dangerous sequences, few instructions are given concerning the exact implementation of the method. Here we describe an interpretation of the method that is amenable to implementation in software (Fig. 1). The input DNA sequence to be screened first undergoes a six-frame translation. The resulting six-amino-acid sequences and the two original DNA sequences corresponding 208

Query sequence DNA strands

Six-frame translation

Division 200 bp

Division 66 aa

For all subsequences Subsequence Extracts BLAST results that have: query coverage = 100% percent identity = max. percent identity of all BLAST results

BLAST BLAST results

Extract best matches

All best matches are SA

SA

NSA

Pass

SA

Some best matches are NSA

Hit

NSA SA

NSA

SA

Pass

No best matches

NSA

NSA

Extension method

SA

Pass

BLAST BLAST results

Hit

All best matches are SA

Extract best matches Else

Pass

If one subsequence leads to a hit, then the initial query sequence is a hit Figure 1 Sequence screening algorithm. The query sequence first undergoes a six-frame translation, then the amino acid sequences and nucleotide sequences are fragmented into the appropriate size. The subsequences are then aligned using BLAST against GenBank and the nature of the best matches is determined. If there is no best match but there are sequences of concern with query coverage >50%, then the alignment extension occurs. The algorithm is repeated on the extended sequences to determine whether original query sequence is a hit to a SAT.

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

c orr e s p ond e n c e to the two strands of the query sequence are then divided into 66 amino acids (aa) and 200-bp fragments, respectively. When the sequence length is not a multiple of 200 bp or 66 aa, a new subsequence is created using the last 200 bp or 66 aa of the sequence. This subsequence overlaps the last subsequence resulting from the initial fragmentation, but it ensures that the entire sequence is screened. All of these fragments are then analyzed individually to determine if they should be flagged. They are first aligned against GenBank using BLAST5. The best matches are extracted among the BLAST results by selecting the alignments with the highest percent identity over the entire 200-bp fragment (query coverage of 100%). To determine if a best match corresponds to a SAT, the information in the GenBank reference page is cross-referenced with a keyword list. For toxins, keywords include alternative names of the toxin, the names of enzymes that are associated with the production and function of the toxin, and the names of organisms that uniquely produce the toxin. For organisms and viruses, keywords include alternative species names, the names of diseases associated with the entries and any toxins or pathogenic agents uniquely produced by the entry. Two keyword lists were developed. The restricted keyword list has 86 records, whereas the extended keyword list has 340 keywords. If every best match is to a SAT, then the fragment is considered a hit. A sequence can be fragmented such that a 200-bp region of SAT could unequally straddle two contiguous fragments. To alleviate this issue, the algorithm creates a new 200-bp (of 66 aa) fragment when it detects the presence of an alignment to a SAT longer than 100 bp or 33 aa on either extremity of the subsequence. This new subsequence is composed of the SAT region from the initial fragment and a region from the appropriate adjacent fragment of a length such that the sum of both regions is equal to 200 bp or 66 aa. Every new extended subsequence is compared with GenBank to identify its best matches, as previously described. This thorough analysis is fairly computationally expensive because screening a 1-kb sequence requires at least 40 sequence alignments (two DNA and six protein alignments for each 200-bp fragment). Sequences of several kilobases can be analyzed in a few minutes on a dedicated server or high-end workstation, which should be compatible with the operational constraints of the gene synthesis industry. The draft guidance published in 2009

Table 1 Comparison of sequence screening protocols Recommendation Fragment double-stranded DNA sequence

IASB

IGSC

US

No

No

200 bp

Screen six-frame translation of DNA sequence

No

Yes

Yes

Screen against curated sequence database

No

Yes

Optional

Defined criteria to identify sequence as a hit

No

No

Best match

Requires human element in screening procedure

Yes

Yes

No

focused exclusively on sequences longer than 200 bp, but the final version has removed this exclusion. This decision is unfortunate. Screening short sequences creates all sorts of bioinformatics complications that can affect the quality of the results. The best-match method has been designed to screen long sequences and is not suitable for screening short sequences. Furthermore, by removing the 200-bp limit, the guidance is somewhat inconsistent. Short sequences are more likely to be ordered as oligonucleotides than double-stranded DNA, but screening oligonucleotide orders is outside the scope of the guidance. For all these reasons, we decided to keep the 200-bp restriction in our implementation of the guidance. To evaluate the performance of this protocol, we developed a test suite of sequences annotated as either SAT (75 sequences) or non-SAT (100 sequences) after manually reviewing alignment results for each sequence. The accuracy of the screen can be estimated by comparing the screen output with the test sequence annotations. Not surprisingly, the performance of the screening protocol depends on the content of the keyword database. The number of false negatives, sequences of concern that are undetected, is minimized when using the extended keyword list (25 false negatives with the limited keyword list versus 1 false negative with the extended keyword list). Because the outcome of the screen is so dependent on the keywords used to analyze alignment results, it would be useful to develop a standardized list of keywords acceptable to all constituencies. Beyond its application in this particular context, these keyword lists are a prerequisite to the development of a sequence-based classification system of SATs6. Moreover, we screened the GenoCAD7,8 parts database. This data set includes 1,258 sequences longer than 200 bp that mimic the order books of gene synthesis companies. The screen returned 32 hits (2.54%). For most hits, the human review did not uncover any significant relation to SATs beyond some local homology between one of many fragments and a SAT sequence. Even so, we found one GenoCAD part

nature biotechnology volume 29 number 3 march 2011

closely related to the YopH protein from Yersinia pestis (gi|14488772). This protocol is extremely effective at detecting sequences of concern embedded into larger sequences because each 200-bp fragment is analyzed individually. The six-frame translation also ensures that redesigned sequences which take advantage of the degeneracy of the genetic code are easily detected by the protocol. However, it proved difficult to design test sequences by introducing mutations in SAT sequences found in GenBank as there is no simple way to determine if such sequences should be detected or not as the biological activity of these sequences is unknown. It would therefore be useful to develop large and realistic training sets that could be used to assess the performance of software implementations of the guidelines recommended by the government. Before the publication of the federal guidelines, the International Association– Synthetic Biology (IASB; Heidelberg, Germany) published “Code of conduct for best practices in gene synthesis” and the International Gene Synthesis Consortium (IGSC; San Francisco) released their “Harmonized screening protocol.” Several important differences between the protocols can be confusing to the public and the gene synthesis industry3. Table 1 shows that the industry is advocating a global analysis of the sequence, leaving the responsibility of interpreting the results to a human operator. The federal protocol advocates a more granular approach that requires breaking down sequences into smaller fragments analyzed individually. This high-resolution screen can detect local features of a sequence that may be undetected if the sequence is analyzed globally in one pass. Since it is not practical to manually review the results of all the sequence alignments performed by the federal protocol, the federal document provides objective criteria to identify what should be further investigated. This automatic classification of sequences of concern is both a strength and a weakness. On the one hand, it makes it possible to objectively assess the performance of the screen, something that is not possible 209

© 2011 Nature America, Inc. All rights reserved.

c orr e s p ond e n c e when the results of sequence alignment are interpreted by human operators. On the other hand, the intrinsic limitations of the best-match method may overlook patterns that human operators would detect. Furthermore, determined individuals could take actions before placing an order to ensure that their order does not raise a red flag. In its defense, the government standard has always been described as a bare minimum that does not prevent the use of complementary approaches such as the ones proposed by the industry. In the long term, the security of gene synthesis may not lie as much in standards as in the availability of biosecurity software applications inspired by computer security solutions. Such biosecurity tools would rely on rapidly evolving models of biosecurity threats to provide human operators with the information they need to quickly and efficiently screen all synthetic DNA sequences at the different steps of the design and fabrication process. The wide adoption of such tools would be objective evidence that the community is developing a culture of responsibility, which is unanimously regarded as the best protection against this new biological threat2,9.

ACKNOWLEDGMENTS This work was supported by National Science Foundation Award no. 1060776 to Virginia Tech. L.A. was supported by a graduate fellowship from the Science Applications International Corporation (SAIC). M.K., A.S., O.M., G.L. and T.S. were supported by undergraduate research fellowships from the MITRE Corporation. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Laura Adam, Michael Kozar, Gaelle Letort, Olivier Mirat, Arunima Srivastava, Tyler Stewart, Mandy L Wilson & Jean Peccoud Virginia Bioinformatics Institute, Virginia Tech, Blacksburg, Virginia, USA. e-mail: [email protected] 1. Eisenstein, M. Nat. Biotechnol. 28, 1225–1226 (2010). 2. LaVan, D.A. & Marmon, L.M. Nat. Biotechnol. 28, 1010–1012 (2010). 3. Fischer, M. & Maurer, S.M. Nat. Biotechnol. 28, 20–22 (2010). 4. Czar, M.J., Anderson, J.C., Bader, J.S. & Peccoud, J. Trends Biotechnol. 27, 63–72 (2009). 5. Camacho, C. et al. BMC Bioinformatics 10, 421 (2009). 6. Wadman, M. Nature 466, 678 (2010). 7. Czar, M.J., Cai, Y. & Peccoud, J. Nucleic Acids Res. 37, W40–W47 (2009). 8. Cai, Y., Wilson, M.L. & Peccoud, J. Nucleic Acids Res. 38, 2637–2644 (2010). 9. Bennett, G., Gilman, N., Stavrianakis, A. & Rabinow, P. Nat. Biotechnol. 27, 1109–1111 (2009).

Partnering Brazilian biotech with the global pharmaceutical industry To the Editor: Previous descriptions of the Brazilian health biotech sector in this journal1,2 have highlighted several challenges to sustainable development, including inefficient interactions between the public and private sectors1, a lack of venture financing1 and a paucity of legal incentives to encourage commercialization of the region’s rich biodiversity2. Here we would like to emphasize the importance of another issue that prevents Brazilian biotech enterprises from successfully bringing innovative drugs to market—the lack of local partnerships between small and large companies and the poor level of collaboration between Brazilian companies and multinational pharmaceutical companies that can accelerate late-stage clinical development. 210

One illustration of the behavior of the local health biotech sector is the lack of interaction between the two main industry associations in the country—the National Association of Pharmaceutical Laboratories (ALANAC; http://www.alanac.org.br) and the Brazilian ResearchBased Pharmaceutical Manufacturers Association (Interfarma; http://www. interfarma.org.br). This weakens the Brazilian industry by preventing both collaboration and pooling of complementary scientific and financial resources that might otherwise bankroll innovative drug development. Most local companies are insufficiently capitalized to carry out innovative R&D activity in the area of biopharmaceuticals, let alone invest over a billion dollars to fund the core process

from target discovery to a regulatory approval or registration. As a result of the weakness of the pharmaceutical sector, not one blockbuster drug has been developed in Brazil throughout its history. Moreover, many ALANAC member companies are opting to produce less R&D-intensive products, such as generics, instead of innovative drugs. Against this background, the Brazilian government has implemented several initiatives to create a local environment that is more conducive to innovative product development, thereby enriching the pool of partnering opportunities for pharmaceutical companies. In 2004, the ‘Innovation Law’ (Law 10,973)1 was introduced to encourage the sharing of intellectual property and other resources between public and private entities and allow direct support of R&D activities in private enterprises. Although the number of Brazilian biomedical inventions licensed at the US Patent & Trademark Office (Washington, DC) has doubled over the past two decades, it is still only a small number (http://www.uspto. gov/web/offices/ac/ido/oeip/taf/cst_utl. pdf). The situation in Brazil is complicated further by the country’s cumbersome patenting process. Under Patent Law 9,279, the National Institute of Industrial Property can grant a pharmaceutical patent related to a product only after agreement has been obtained from Brazil’s National Health Surveillance Agency. This rule makes the Brazilian process longer and more unwieldy than that in any other territory in the world. Even so, progress in fostering an innovation- and enterprise-friendly environment is being made. Two laws for creating favorable fiscal incentives for R&D investment (the ‘Asset Law’; Law 11,196) and income tax exemptions for enterprises involved in R&D (Law 11,487) were introduced in 2005 and 2007, respectively. Although these laws had only a minor impact initially, in 2008 the income tax deduction derived from Law 11,196 amounted to ~0.05% of Brazilian gross domestic product (http://www.mct.gov.br). Even greater benefits could potentially be accrued if Law 11,487 could be extended to private enterprises, rather than applied solely to public research institutions, as it does at present. More recently, the launch of the Brazilian Technology System (SIBRATEC3) has facilitated the identification and development of promising compounds

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

c orr e s p ond e n c e when the results of sequence alignment are interpreted by human operators. On the other hand, the intrinsic limitations of the best-match method may overlook patterns that human operators would detect. Furthermore, determined individuals could take actions before placing an order to ensure that their order does not raise a red flag. In its defense, the government standard has always been described as a bare minimum that does not prevent the use of complementary approaches such as the ones proposed by the industry. In the long term, the security of gene synthesis may not lie as much in standards as in the availability of biosecurity software applications inspired by computer security solutions. Such biosecurity tools would rely on rapidly evolving models of biosecurity threats to provide human operators with the information they need to quickly and efficiently screen all synthetic DNA sequences at the different steps of the design and fabrication process. The wide adoption of such tools would be objective evidence that the community is developing a culture of responsibility, which is unanimously regarded as the best protection against this new biological threat2,9.

ACKNOWLEDGMENTS This work was supported by National Science Foundation Award no. 1060776 to Virginia Tech. L.A. was supported by a graduate fellowship from the Science Applications International Corporation (SAIC). M.K., A.S., O.M., G.L. and T.S. were supported by undergraduate research fellowships from the MITRE Corporation. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Laura Adam, Michael Kozar, Gaelle Letort, Olivier Mirat, Arunima Srivastava, Tyler Stewart, Mandy L Wilson & Jean Peccoud Virginia Bioinformatics Institute, Virginia Tech, Blacksburg, Virginia, USA. e-mail: [email protected] 1. Eisenstein, M. Nat. Biotechnol. 28, 1225–1226 (2010). 2. LaVan, D.A. & Marmon, L.M. Nat. Biotechnol. 28, 1010–1012 (2010). 3. Fischer, M. & Maurer, S.M. Nat. Biotechnol. 28, 20–22 (2010). 4. Czar, M.J., Anderson, J.C., Bader, J.S. & Peccoud, J. Trends Biotechnol. 27, 63–72 (2009). 5. Camacho, C. et al. BMC Bioinformatics 10, 421 (2009). 6. Wadman, M. Nature 466, 678 (2010). 7. Czar, M.J., Cai, Y. & Peccoud, J. Nucleic Acids Res. 37, W40–W47 (2009). 8. Cai, Y., Wilson, M.L. & Peccoud, J. Nucleic Acids Res. 38, 2637–2644 (2010). 9. Bennett, G., Gilman, N., Stavrianakis, A. & Rabinow, P. Nat. Biotechnol. 27, 1109–1111 (2009).

Partnering Brazilian biotech with the global pharmaceutical industry To the Editor: Previous descriptions of the Brazilian health biotech sector in this journal1,2 have highlighted several challenges to sustainable development, including inefficient interactions between the public and private sectors1, a lack of venture financing1 and a paucity of legal incentives to encourage commercialization of the region’s rich biodiversity2. Here we would like to emphasize the importance of another issue that prevents Brazilian biotech enterprises from successfully bringing innovative drugs to market—the lack of local partnerships between small and large companies and the poor level of collaboration between Brazilian companies and multinational pharmaceutical companies that can accelerate late-stage clinical development. 210

One illustration of the behavior of the local health biotech sector is the lack of interaction between the two main industry associations in the country—the National Association of Pharmaceutical Laboratories (ALANAC; http://www.alanac.org.br) and the Brazilian ResearchBased Pharmaceutical Manufacturers Association (Interfarma; http://www. interfarma.org.br). This weakens the Brazilian industry by preventing both collaboration and pooling of complementary scientific and financial resources that might otherwise bankroll innovative drug development. Most local companies are insufficiently capitalized to carry out innovative R&D activity in the area of biopharmaceuticals, let alone invest over a billion dollars to fund the core process

from target discovery to a regulatory approval or registration. As a result of the weakness of the pharmaceutical sector, not one blockbuster drug has been developed in Brazil throughout its history. Moreover, many ALANAC member companies are opting to produce less R&D-intensive products, such as generics, instead of innovative drugs. Against this background, the Brazilian government has implemented several initiatives to create a local environment that is more conducive to innovative product development, thereby enriching the pool of partnering opportunities for pharmaceutical companies. In 2004, the ‘Innovation Law’ (Law 10,973)1 was introduced to encourage the sharing of intellectual property and other resources between public and private entities and allow direct support of R&D activities in private enterprises. Although the number of Brazilian biomedical inventions licensed at the US Patent & Trademark Office (Washington, DC) has doubled over the past two decades, it is still only a small number (http://www.uspto. gov/web/offices/ac/ido/oeip/taf/cst_utl. pdf). The situation in Brazil is complicated further by the country’s cumbersome patenting process. Under Patent Law 9,279, the National Institute of Industrial Property can grant a pharmaceutical patent related to a product only after agreement has been obtained from Brazil’s National Health Surveillance Agency. This rule makes the Brazilian process longer and more unwieldy than that in any other territory in the world. Even so, progress in fostering an innovation- and enterprise-friendly environment is being made. Two laws for creating favorable fiscal incentives for R&D investment (the ‘Asset Law’; Law 11,196) and income tax exemptions for enterprises involved in R&D (Law 11,487) were introduced in 2005 and 2007, respectively. Although these laws had only a minor impact initially, in 2008 the income tax deduction derived from Law 11,196 amounted to ~0.05% of Brazilian gross domestic product (http://www.mct.gov.br). Even greater benefits could potentially be accrued if Law 11,487 could be extended to private enterprises, rather than applied solely to public research institutions, as it does at present. More recently, the launch of the Brazilian Technology System (SIBRATEC3) has facilitated the identification and development of promising compounds

volume 29 number 3 march 2011 nature biotechnology

c orr e s p ond e n c e Biodiversity Bioprospecting Extracts Chemical synthesis Combinational chemistry

Academia*, Sc&T, national institutes, universities + Brazilian pharmaceutical companies

CROs, national and international Brazilian and international companies and partnerships

$

$

$

$

Preclinical tests*

Clinical tests Phase 1

Clinical tests Phases 2 and 3

FDA and EMA

Private sector

*Public and private funds. Laws 10,973, 11,196 and 11,487 - SIBRATEC

© 2011 Nature America, Inc. All rights reserved.

Figure 1 Model system to foster partnership in the Brazilian pharmaceutical sector. CRO, Consultative Research Organization; Sc&T, Science and Technology.

in academia. In this scheme, government funding is used to support preclinical and phase 1 clinical research of certain compounds selected by specialists from academic laboratories. It is hoped that these activities will complement and synergize with the activities of the small number of private contract research organizations in Brazil that carry out clinical work. Indeed, there are clear examples of companies in the ALANAC group that are now attracted to developing new drugs. Such initiatives are critical to move lead molecules to a stage of validation where the pharmaceutical industry is willing to license in, or collaborate in the development of, a molecule. Domestic Brazilian pharmaceutical companies, as mentioned previously, are in general not financially capable of performing clinical testing alone, particularly phase 2 and 3 trials. To address this problem, the Brazilian government is actively funding pharmaceutical enterprises to propel compounds into early trials through a competitive system called ‘subvenção’. In addition, foreign venture capital funds specialized in biotech are now setting up in Brazil to further contribute to this role and grow the national health biotech industry. For example, San Francisco—headquartered Burrill and Company is now fully operational in the country, with a $150 million life science venture fund slated to launch early this year. To support basic scientific research and facilitate the translation of products from the bench to industry, Brazil has recently

invested over $200 million to establish 123 science and technology national institutes (virtual networks of linking individuals in different centers of excellence), 34 of which are in the area of human health. Half of these health institutes are involved with SIBRATEC. Finally, the Ministry of Health is funding partnerships between the private and public sectors in an attempt to reduce Brazil’s $7 billion accumulated deficit and ensure that the needs of its major public health program, SUS (Unified Health System; http://portal.saude.gov.br/portal/ saude/default.cfm), are met. Thus, more than ever before, the Brazilian biotech sector has an opportunity to develop innovative drugs. As many of the large pharmaceutical corporations have been operating in Brazil for many years, there is also a customer base for licensing such drugs—if only the bridges could be made. Fostering big pharma–Brazilian biotech partnerships will be crucial for the further development of the sector and its ability to successfully access the $7 trillion global pharmaceutical market. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Luiz A B de Castro Brazilian Academy of Sciences, Rio de Janeiro, Brazil. e-mail: [email protected] 1. Rezaie, R. et al. Nat. Biotechnol. 26, 627–644 (2008). 2. Castro, L.A.B. & Barros, A.K. Nat. Biotechnol. 27, 317–318 (2009). 3. Mota, R. Tecnologia & Inovação. Valor Especial. June (2010).

nature biotechnology volume 29 number 3 march 2011

211

c orr e s p ond e n c e

212

other related ones, including salt, heat and oxidative stresses, were considered as well. The final selection consisted of 25 genes, which we designate ‘stress tolerance genes’ (STGs), involved in diverse aspects of stress tolerance and in a wild-type Columbia-0 (Col-0) background (Table 1, Supplementary Table 1). We added two additional lines (MYB90 and tAPX) that had previously not been analyzed to 15 of the 25 STG lines that had already been demonstrated to survive better upon severe drought (Table 1, Fig. 1a and Supplementary Fig. 1). To quantify growth of the STG lines, we developed an assay mimicking relatively mild drought stress conditions in which the rosette size of plants grown in soil was followed over time (Supplementary Methods). To ensure test reproducibility, we also used a large number of plants in an automated platform, designated the ‘weighing imaging and watering machine’ (WIWAM; Fig. 1b and

a

d

Supplementary Fig. 2). WIWAM enabled the daily imaging and controlled watering of 216 plants. Plants were germinated and grown under control conditions until stage 1.04 (ref. 7), after which watering continued for the control plants, but was stopped for the stressed plants until the set stress level was reached and subsequently kept constant (Supplementary Figs. 3 and 4). In wild-type plants, progressive soil drying resulted in a gradual decrease of growth rates, with a final reduction of the rosette area of 30–40% as a consequence (Fig. 1c, Table 1 and Supplementary Fig. 4). To assess the performance of the STG lines in terms of genotype differences and genotype-specific responses to the drought stress, we analyzed genotype, environment, time effects and their interactions with a linear mixed model (Table 1, Supplementary Figs. 5 and 6, Supplementary Methods). Significant (P < 0.01) genotype differences were

b

50

e

R = 0.82*

30 10 –10 –30 –50 –50

c

20

Relative growth performance

To the Editor: Although drought tolerance is a central concern of plant research, the translatability for crop improvement is relatively low. Here we report on a major contributing factor to this lack of success. Drought tolerance is predominately scored based on an improved survival rate under lethal conditions that, as demonstrated by our study, does not predict superior growth performance and, thus, biomass yield gain, under moderate drought often encountered in the field. Drought tolerance is a major subject of trait research for agbiotech companies and thousands of academic papers have been published on the topic. Consequently, there is a plethora of reports on improved drought tolerance, mainly in the model plant Arabidopsis thaliana1. Classic genetic engineering approaches involve target genes that function in mechanisms used by plants to avoid and/or tolerate drought, such as stomatal conductance or osmolyte production2. Such genes, frequently identified through expression profiling, include signaling components and downstream effector genes. However, despite the apparent success of stress research on model plants, rarely are the findings applied to improve crops. Only a few genes have been characterized that enhance stress tolerance in model plants or crops leading to increased yields3–6 and the molecular mechanisms through which they work remain only partly understood. One of the key reasons relates to the genetic and physiological differences between model and crop species. In Arabidopsis research, drought tolerance is assessed predominantly under quite severe conditions in which plant survival is scored after a prolonged period of soil drying. However, in temperate climates, limited water availability rarely causes plant death, but restricts biomass and seed yield. To study the relation between survival and biomass gain under drought, we analyzed the growth of transgenic Arabidopsis plants with increased tolerance to lethal stress in a mild stress assay. An extensive literature screen was conducted to identify Arabidopsis genes that, in gain- or loss-offunction situations, confer stress tolerance in Arabidopsis, without growth penalty under control conditions. Although drought and osmotic stresses were prioritized,

Growth under stress conditions

© 2011 Nature America, Inc. All rights reserved.

Survival and growth of Arabidopsis plants given limited water are not equal

–30

–10

10

30

Growth under control conditions

50

R = 0.0037

0

–20 –40

–20

0

20

40

Growth under control conditions

Figure 1 Growth reduction caused by stress is independent of rosette size under control conditions. (a,b) Tolerance to severe stress was scored in the survival assay (a), whereas growth under mild drought was assessed with WIWAM in a drought stress regime (b) that reduced final rosette area by 30–40%. (c) The top panel shows a wild-type plant grown under control conditions. The bottom panel shows a plant grown under drought conditions. (d,e) End-time-point area measurements were used to calculate differences between STG and wild-type (WT) plants under control (area C) (1 – (area C STG/area C WT)) × 100 and drought (area D) conditions (1 – (area D STG/area D WT)) × 100, as well as the difference in droughtrelated growth inhibition (relative growth performance) ((1 – (area D WT/area C WT)) – (1 – (area D STG/ area C STG))) × 100. Asterisk marks significance (P < 0.01).

volume 29 number 3 march 2011 nature biotechnology

c orr e s p ond e n c e Table 1 STG lines tested show no significant genotype-specific responses to the imposed drought stress, either combined or across all the time points. Survival

Mean percent P-value (genotype P-value (genotype reduction ± control conditions) drought conditions) s.e.m.a

At1g01720

ATAF1

1

Loss of function Transcription

Improved

0.336

0.012

At3g06010

CHR12

1

Loss of function Transcription

Unchanged

0.119

0.724

36.15 ± 2.35

At1g30270

CIPK23

1

Loss of function Ca+ signaling

Improved

0.210

0.121

42.07 ± 1.87

At1g73660

MAPKKK

1

Loss of function Signaling

Unchanged

0.191

0.118

40.48 ± 3.70

At5g21100

aAAO

1

Loss of function Reactive oxygen species metabolism

Unchanged

0.836

0.240

42.35 ± 1.83

© 2011 Nature America, Inc. All rights reserved.

Gene identifier Gene symbol

Experiment

Biological function

Line

42.35 ± 1.83

At5g45340

CYP707A3

1

Loss of function Hormone metabolism

Improved

0.154

0.657

40.54 ± 2.08

–

–

1

Wild type

–

–

–

42.45 ± 2.19

At1g05260

RCI3

2

Gain of function Cell wall

Improved

0.004b

0.027

26.22 ± 6.46

At3g14440

NCED3

2

Gain of function Hormone metabolism

Improved

<0.001b

<0.001b

35.98 ± 6.26

At1g78290

SRK2C

2

Gain of function Signaling

Improved

0.043

0.196

39.38 ± 1.67

At1g74310

HSP101

2

Gain of function Protein stability

Unchanged

0.109

0.102

43.75 ± 6.00

At5g27150

NHX1

2

Gain of function Transport

Unchanged

0.002b

0.209

33.80 ± 2.48

At1g66390

MYB90

2

Gain of function Transcription

Improved

0.100

0.120

35.83 ± 2.49

At1g77490

tAPX

2

Gain of function Reactive oxygen species Improved metabolism

0.007b

0.001b

34.58 ± 2.58

–

–

2

Wild type

–

–

–

32.70 ± 3.93

At2g38880

NF-YB

3

Gain of function Transcription

Improved

0.810

0.858

23.37 ± 1.95

At4g17610

CBL1

3

Gain of function Ca+ signaling

Improved

0.001c

<0.001c

27.85 ± 1.69

–

Col

3

Wild type

–

–

–

29.98 ± 2.70

At5g13680

ELO2

4

Loss of function Transcription

Improved

<0.001b

0.007b

35.72 ± 3.64

At3g24500

MBF1c

4

Gain of function Transcription

Unchanged

0.772

0.290

45.47 ± 2.94

At1g08810

MYB60

4

Loss of function Transcription

Improved

0.367

0.362

40.03 ± 3.26

At3g15500

ANAC055

4

Gain of function Transcription

Improved

0.043

0.264

47.75 ± 4.58

At4g09570

CPK4

4

Gain of function Ca+ signaling

Improved

<0.001b

<0.001b

47.29 ± 3.71

At1g56600

GOLS2

4

Gain of function Osmoprotection

Improved

<0.001b

<0.001b

32.74 ± 2.32

–

–

4

Wild type

–

–

–

30.85 ± 5.52

At1g31970

STRS1

5

Loss of function Transcript stability

Unchanged

0.012

0.026

42.74 ± 5.01

At5g08620

STRS2

5

Loss of function Transcript stability

Unchanged

0.367

0.316

37.97 ± 3.31

–

–

5

Wild type

–

–

–

33.39 ± 1.32

At1g15690

AVP1

6

Gain of function Transport

Improved

<0.001c

<0.001c

33.01 ± 1.55

At1g05680

UGT

6

Gain of function Hormone metabolism

Improved

0.672

0.466

49.91 ± 2.51

–

–

6

Wild type

–

–

–

36.89 ± 4.05

–

–

–

–

–

–

aMean

± s.e.m. percent reduction of rosette size by drought measured 10 days into the treatment. bSignificant (P < 0.01) decrease of rosette area of the STG compared with the wild type under control or drought condition. cSignificant (P < 0.01) increase of rosette area of the STG compared with the wild type under control or drought conditions.

measured for nine lines and in seven cases the difference was significant in both the control and stress environments. The two lines that were significantly larger under control conditions, CBL1 and AVP1, were also larger during drought, and the reverse held true for the smaller lines GOLS2, CPK4, ELO2, NCED3, NHX1, tAPX and RCl3. However, none of the genotypes tested showed a significant specific response to the drought stress imposed either combined or across all the time points (genotype*environment or genotype*environment*time interaction; data not shown). In other words, growth reduction caused by drought was comparable for all genotypes tested. Overall, our data clearly show that enhanced survival under severe drought

is not a good indicator for improved growth performance under mild drought conditions. Superior survival under severe drought is often associated with constitutive activation of water-saving mechanisms, such as stomatal closure, that can, on the contrary, lead to growth penalty8. Here, a number of STG lines showed a growth reduction, albeit a subtle one (Table 1). Nevertheless, it is worth mentioning that the observed growth reduction could possibly be bypassed by the use of conditional or tissue-specific promoters8,9. Generally, plant size and survival, at least under laboratory conditions, are assumed to be negatively correlated, because small plants transpire and use less water. In contrast, in our study, growth reduction caused by mild drought was independent of the STG

nature biotechnology volume 29 number 3 march 2011

size measured under control conditions (Fig. 1d,e). Importantly, lines that were larger in the control environment kept their growth advantage under stress conditions, demonstrating that plants have enough resources to sustain both stress tolerance and improved growth. This observation is in line with the favorable carbon status measured in Arabidopsis leaves under both mild osmotic and drought stresses10,11 and consistent with the hypothesis that plants reduce their growth as a primary adaptation response to stress rather than as a secondary consequence of resource limitation. Under unpredictable environments, growth reduction enables plants to redistribute and save resources, ensuring reproduction even when the stress becomes extreme. However, from the agricultural point of view, when 213

© 2011 Nature America, Inc. All rights reserved.

c orr e s p ond e n c e the stress episode does not threaten plant survival, growth reduction can be counterproductive, leading to unnecessary yield loss. Thus, limiting growth reduction might provide a strategy to boost plant biomass productivity under stress. Biomass yield is of rising importance with the increasing demand for energy crops but has furthermore been shown to be relevant as one component determining seed yield12. Certainly, a better understanding of the mechanisms that regulate growth under stress conditions, such as those involved in shutting down meristem activity, will be vital in the development of new technologies to increase plant growth under stress13. Although stress responses of mature organs are relatively well characterized and it is now clear that stress responses are specific to the developmental stage, tissue and even the cell type11,14, the mechanisms that reduced growth under stress are poorly understood. From the technological perspective, automated growth phenotyping under variable environmental conditions with platforms, such as WIWAM, or Phenopsis15 will be essential. Finally, it is important to mention that whereas rosette area can be reliably used to assess biomass, it remains, however, a proxy for seed yield and thus analogous automated platforms focused on seed phenotypes will have to be developed. In summary, the results imply that indiscriminate selection for lines that survive better under severe stress might be a critical factor responsible for the low success rate by which academic research on drought stress translates to the field. As enhanced survival is largely a function of water-saving mechanisms rather than a net improvement in plant production, it will still be a trait of choice in arid regions, but will most probably not enhance plant yield in moderate climates. In our opinion, such mild, in contrast to severe, conditions will favor bolder plants maintaining more growth, photosynthesis and metabolism despite a water shortage, opening a new exciting paradigm for trait identification. Note: Supplementary information is available on the Nature Biotechnology website. AUTHOR CONTRIBUTIONS Experimental work was performed by A.S., K.V., P.C., K.M., A.P., N.G., F.H. and V.B.T. Unpublished transgenic lines were provided by M.G. and C.T.; B.D.M. and S.D. designed and programmed WIWAM; M.V. performed the statistical analysis; and F.V.B. and D.I. supervised the project.

214

ACKNOWLEDGMENTS We thank R. Mittler, V. Buchanan-Wollaston, K. Shinozaki, D.-P. Zhang, I. Murgia, J. Salinas, S. Lindquist, E. Blumwald and S. Barack for kindly providing transgenic lines, and M. De Cock for help in preparing the manuscript. This work was supported by grants from Ghent University (Bijzonder Onderzoeksfonds Methusalem project no. BOF08/01M00408) and Multidisciplinary Research Partnership (Biotechnology for a Sustainable Economy no. 01MRB510W), the Interuniversity Attraction Poles Programme (IUAP VI/33), initiated by the Belgian State, Science Policy Office and the European Community Grant FP6 IP AGRON-OMICS (contract LSHG-CT-2006-037704). S.D. is indebted to the Agency for Innovation through Science and Technology for a predoctoral fellowship. V.B.T. is the recipient of a Marie Curie Intra-European Fellowship for Career Development (PIEF-GA-2008-221427). COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

Aleksandra Skirycz1,2,5, Korneel Vandenbroucke1,2,4,5, Pieter Clauw1,2, Katrien Maleux1,2, Bjorn De Meyer1,2, Stijn Dhondt1,2, Anna Pucci1,2,4, Nathalie Gonzalez1,2, Frank Hoeberichts1,2, Vanesa B Tognetti1,2, Massimo Galbiati3, Chiara Tonelli3, Frank Van Breusegem1,2, Marnik Vuylsteke1,2 & Dirk Inzé1,2 1Department of Plant Systems Biology,

VIB, Ghent, Belgium. 2Department of Plant Biotechnology and Genetics, Ghent University, Ghent, Belgium. 3Dipartimento di Scienze Biomolecolari e Biotecnologie, Università degli Studi di Milano, Milano, Italy. 4Present address: Bayer Crop Science, Ghent, Belgium 5Current address: Dipartimento di Agrobiologia e Agrochimica, Università degli Studi della Tusc(K.V.); Dipartimento di Agrobiologia e Agrochimica, Universita degli Studi della Tuscia, Viterbo, Italy (A.P.). 5These authors contributed equally to this work. e-mail: [email protected] 1. Umezawa, T., Fujita, M., Fujita, Y., Yamaguchi-Shinozaki, K. & Shinozaki, K. Curr. Opin. Biotechnol. 17, 113–122 (2006). 2. Hirayama, T. & Shinozaki, K. Plant J. 61, 1041–1052 (2010). 3. Nelson, D.E. et al. Proc. Natl. Acad. Sci. USA 104, 16450–16455 (2007). 4. De Block, M., Verduyn, C., De Brouwer, D. & Cornelissen, M. Plant J. 41, 95–106 (2005). 5. Castiglioni, P. et al. Plant Physiol. 147, 446–455 (2008). 6. Li, B., Wei, A., Song, C., Li, N. & Zhang, J. Plant Biotechnol. J. 6, 146–159 (2008). 7. Boyes, D.C. et al. Plant Cell 13, 1499–1510 (2001). 8. Kasuga, M., Liu, Q., Miura, S., Yamaguchi-Shinozaki, K. & Shinozaki, K. Nat. Biotechnol. 17, 287–291 (1999). 9. Wang, Y. et al. Mol. Plant 2, 191–200 (2009). 10. Hummel, I. et al. Plant Physiol. 154, 357–372 (2010). 11. Skirycz, A. et al. Plant Physiol. 152, 226–244 (2010). 12. Hausmann, N.J. et al. Evolution 59, 81–96 (2005). 13. Tisné, S. et al. Plant Cell Environ. 33, 1875–1887 (2010). 14. Dinneny, J.R. et al. Science 320, 942–945 (2008). 15. Granier, C. et al. New Phytol. 169, 623–635 (2006).

volume 29 number 3 march 2011 nature biotechnology

c o m m e n ta r y

Biomedical technology and the clinic of the future © 2011 Nature America, Inc. All rights reserved.

Technology pioneers trade views with a clinician and an entrepreneur on the likely impact of large-scale systems technology in healthcare.

T

o date, large-scale ’omics data sets and systems approaches in biology have had a relatively minor impact on the practice of medicine. As new technology brings individual genome sequencing closer to reality and large-scale biology continues to progress, opportunities are likely to open up in disease prediction, prevention, diagnosis and treatment. Here the views of two researchers on the potential of disruptive biomedical technologies in clinical practice are contrasted with the perspectives of a clinician and an entrepreneur in commercial clinical information technology.

POINT: Are we prepared for the future doctor visit? Stephen H Friend & Trey Ideker Imagine the following visit to the doctor’s office, which, although fictitious, is based on technologies that are emerging or already available. A patient, Jane Doe, enters the clinic for a routine physical exam. Today, at least seven parameters would be registered upon her admittance: sex, age, height, weight, temperature, pulse rate and blood pressure (itself a pair of values). But in the future when Jane registers, this set of routine measurements will have expanded enormously (Table 1). Tomorrow’s routine checkup Either on this visit or a previous one, Jane’s full genome has been sequenced, noninvasively, using a buccal swab. At the same time, and optionally on every visit, the nurse has sampled and sequenced the metagenome of Stephen H. Friend is at Sage Bionetworks, Seattle, Washington, USA. Trey Ideker is in the Departments of Medicine and Bioengineering, University of California, La Jolla, California, USA, and at The Institute for Genomic Medicine, University of California, La Jolla, California, USA. e-mail: [email protected] or [email protected]

the microbiome pool resident in the patient’s mucosal and gastrointestinal cavities, providing a detailed characterization of the population of microbes commensal with the human host. Messenger RNA, microRNA, proteome and metabolome profiles may be gathered from urine and, if necessary, whole blood and other tissues. Finally, in addition to height and weight, a large panel of physiological parameters and images is monitored, capturing detailed information about respiration, endocrine function, cardiac and brain activity, and so on. Another key development that will transform Jane’s visit to the clinic is deeper data integration. All of the newly gathered information are banked in a unified electronic medical record, which uses a relational database to establish cross-references among the different data types. The new information augments the history of data gathered on previous visits, including all medical treatments and outcomes accumulated over the patient’s lifetime. Crucially, the new data are then integrated with a library of biological network models spanning multiple levels and scales (Fig. 1). First is the network of functional and molecular interactions—a.k.a. the molecular wiring diagram—providing a modular, hierarchical

nature biotechnology volume 29 number 3 march 2011

and executable view1 of the cellular processes underlying human health and disease. Such networks are being assembled from diverse large- and small-scale experiments performed over decades of systems biology and biomedical research, providing an up-to-date representation of current knowledge in the field2,3. A second type of network model will represent the relevant nosology, which maps relationships between diseases based on their similarities in etiology, pathogenesis and symptoms. Related to this will be another network—that of pharmacologic treatments, which provides rich information about the different protocols and drugs that are available along with their quantitative inter-relationships. One more important network will be the patient’s extended social network and pedigree, which will be available along with references to the integrated medical records of friends and relatives. This social network documents significant personal relationships in Jane’s life, weighted by importance and, subject to privacy concerns, gathered from social networking websites, personal address books, geographical co-location data, as well as cell phone and e-mail usage. The pedigree provides a complementary set of social relationships that have a genetic basis. The benefits of these network models to Jane are severalfold. First, they integrate an array of different lines of evidence for health or disease, enabling the formulation of compound biomarkers that are combinations or functions of many simultaneous readouts. Such compound biomarkers can be more robust than biomarkers based on individual genes, proteins or metabolites4. Second, the networks provide a natural interpretation of the mechanisms behind Jane’s present and future conditions, in contrast to current biomarkers that often have little relation to the actual cause of disease. Third, Jane’s data and outcomes can be dynamically analyzed and reintegrated to 215

C O M M E N TA R Y

Table 1 Current and emerging genomic technologies for network medicine Data space

Technologies

Information

Feasibility for patient testing

Genome

Next nth-generation instruments (e.g., reversible dye terminators, sequencing by ligation and pyrosequencing)

Whole genome, including single nucleotide polymorphisms and copy number variants

<$1,000 per patient within 2 years

Epigenome

Chromatin immunoprecipitation sequencing (ChIP-seq), methyl-seq, genome-wide DNase hypersensitivity assays

Chromatin modifications and structure

Distant future: currently used for basic research only

Transcriptome

RNA sequencing, DNA microarrays and bead arrays

Whole genome transcript abundances and translation rates

<$500 per patient

microRNA abundances

Noninvasive (urine, blood) or invasive (biopsy)

Available now

Proteome

Mass spectrometry, multiparameter fluorescence-activated cell sorting (FACS)

Protein abundances and modifications

Predominantly used in basic research

Metabolome

Mass spectrometry (electrospray ionization/ triple quadrupole)

Metabolic abundances and fluxes

<$200 per patient Well established for neonatal screening

NMR, isotope labeling

© 2011 Nature America, Inc. All rights reserved.

Noninvasive (buccal swab)

Characterization of patient microbiome

Noninvasive (urine, blood)

Protein binding and signaling networks

Immunoprecipitation, co-affinity purification and protein arrays

Protein-protein physical binding interactions, kinase-substrate targeting

Distant future: currently used for basic network assembly

Transcriptional networks

Genome-wide ChIP-seq and protein binding arrays

Protein-DNA, protein-RNA interactions

Distant future: currently used for basic network assembly

Forward and reverse genetic networks

Forward: gene linkage and association studies

Phenotypic profiling, epistatic interactions

Not applicable: networks inferred from populations of individuals

Reverse genetics: RNA interference screening and combinations, synthetic genetic analysis

improve the network models themselves. Thus, the impact of a network can increase over time along with the coverage and accuracy of the information it captures. For this reason, all of these network models have been developed using an online public ‘commons’, which is open-access, crowd-sourced and hosted by a neutral party. The commons serves as a platform for sharing biomedical data, models and tools, including results from extensive clinical trials, ample proteomic and genomic information, proper curation with standard annotations and full assurance that all of the information will remain in the public domain without the constraints of intellectual property (IP). The commons is also a portal by which federal regulators monitor drugs, since, in this future world, therapies are evaluated predominantly by patient-driven trials after their initial approval as safe compounds. On the basis of Jane’s integrated data, multiple indicators are triggered that she is at moderate risk factor 12.7 for breast cancer. The molecular network model indicates both common and rare variants in genes within module 3b.AF8001D, a tumor suppressor module involved in DNA repair and cell cycle checkpoints, resulting in a quantitative decrease in its simulated functional output, which is corroborated by the mRNA and protein expression profiling data. In addition, the system predicts greater than average activation of a key onco-module involved in cell proliferation, which triggers a warning on the nurse’s 216

information management console. The entire pattern of network module activity is crossreferenced to the nosology, highlighting a web of diseases for which Jane is at risk and with tubular carcinoma type IIa3 as the most likely outcome. Type IIa3 is a tumor substratification of the future, which can only be identified using molecular profiling data in conjunction with a network model. Jane’s integrated pedigree shows that, although no immediate family members have been diagnosed with similar diseases, two family members at network distances 2 and 3, respectively, have had breast and ovarian cancer. The history for these individuals shows that both were initially placed on preventative treatment with the compound ‘aleamed A’ but switched to ‘aleamed B’ after experiencing deleterious side effects, including severe depression. Although Jane’s genome sequence places her only at moderate risk for depression, this trait is strongly enriched among the social network of her immediate friends—a finding that raises Jane’s own depression risk factor5. Thus, aleamed B is recommended as the initial course of action for Jane, or related protocols as indicated by the network of treatments. Technological possibility or political and social pipe dream? What are the barriers to making this scenario a reality? Technologies, such as genome sequencing and molecular profiling, are here now (Table 1). The required network mod-

els—representing connections at the molecular, social, chemical and disease levels—are also available in various forms, although their coverage is far from complete. Clearly, using network maps to develop therapies will require representations of disease that go far beyond the classic biopathway maps so vaunted today. It will require pathophysiological maps that highlight the protein targets lacking in redundancy, such that when altered by drugs these targets modify disease. In turn, these maps will need to highlight unforeseen secondary effects of modifying each potential target. Assembling and interpreting such integrative network maps will also require that we populate patient records with genotypic and phenotypic changes at scales far beyond our capabilities today. It will require a new class of primary care physician who is proficient in biostatistics, the various data types, networks and modes of integration, and the contribution of each of these components to the overall disease risk and treatment plan. Presently, some of the most forward-looking tests are provided by direct-to-consumer personalized genetics companies6, but a key challenge faced by such companies is how to provide suitable education to the patient without physician guidance. However, the proposal we make here is that the most challenging hurdles that will keep this reality from occurring may not be related to technology or education but will be social and political in nature. We acknowledge that the complex technology and informatics meth-

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

C O M M E N TA R Y ods that will need to be developed will require massive efforts extending over more than just a few years. At the same time, we anticipate that overcoming the accompanying social and political hurdles will be the more vexing problems, as they will involve addressing issues such as how we will need to work together, how we will need to reward individuals and what we will value. First and foremost, the future of biomedicine will require that the data are generated and used in a sustainable way. Currently, we fund researchers to perform large clinical studies as if they were indigenous hunter-gatherers. The assumption is that these individuals must not only generate large data sets but should also zealously defend their right to use the data to deliver conclusions that develop the careers of themselves and their laboratories. The data, when finally made available, are often not formatted in a way that is accessible for other investigators to use further, other than as a conclusion. It is as if the patient, who is the actual donor and owner of their data, is sidelined by the biomedical institutions that take on a paternalistic ownership role. Should it be a surprise that this situation typically places the institution’s interests and incentives in control of how the data are distributed? Another driver of current behavior within our medical-industrial complex is the publisher, who wishes to charge for access to the results wrapped within the paper, because this paper is the main scientific currency with which authors are recognized. How can we expect researchers to share their insights before they have written papers, if there are no means to provide them recognition for the actual work itself, including their models and representations of disease? Because the models will require massive amounts of data, building these models will require data sharing in ways that issues of privacy and IP typically obstruct. Dealing with these issues effectively will require that the patients with disease be highly visible. If patients come to better understand the Byzantine cloistering of data that is prevalent today, they will likely demand a shift in culture to one that places the impact squarely on patients, not on the careers of academic investigators. In a more positive frame, there is enormous potential for the coming tsunami of clinicalgenomic data to fundamentally improve the process of developing therapies, which has been atrociously ineffective7. Most necessary, we posit, will be to establish a shared infrastructure for the data, tools and models needed to evolve our understanding of disease and its treatment (that is, the online public commons featured in Jane Doe’s visit to the doctor

Figure 1 Layers of genomic and network-based information in integrative healthcare. The future primary care physician may need to cope with a staggering array of integrated patient data including genome sequences and biological networks. Access to the full electronic medical record (far left) will provide data at the level of genome sequence (lower left), pedigree and social network (lower center), nosology of disease (far right) and molecular network modules (center). The module 3b.AF8001D is represented as a map of functional interactions among protein complexes, with red nodes indicating proteins for which significant genetic variants were identified. Integrative analysis of these data and model simulation yields a patient prognostic report (lower right). Sequence view is adapted from the UC Santa Clara Genome Browser (http://genome.ucsc.edu/). Network views are from Cytoscape (http:// www.cytoscape.org/).

described above). Such a platform must grant unrestricted use of data to develop therapies, and it will benefit greatly from public-private partnerships. One powerful example of such a partnership within the realm of drug discovery is the Structural Genomics Consortium (SGC) led by Aled Edwards and Chas Bountra8. Now 6 years old, this consortium has stimulated sharing of data and models to the extent that the majority of crystal structures solved today no longer have IP attached to them. This is an important example of how a domain of scientific discovery has been transformed—from the traditional assumption that solving structures of targets is a competitive proprietary benefit, to the modern realization that such competitive activities end up crippling all parties because each effort is only a small piece of the whole and has access to only a fraction of the data. Since its inception with a focus on crystal structures, SGC has diversified to tackle other components of basic drug discovery, such as the generation of chemical probes, guided by the same open-access, IP-free philosophy. A second example of real data sharing is the Coalition Against Major Diseases9, which has

nature biotechnology volume 29 number 3 march 2011

worked to provide open access to clinical trial data from Alzheimer’s and other neurological disorders. Patient-led clinical trials, such as those facilitated by PatientsLikeMe or the Life Raft Group, are also a promising direction, provided certain challenges can be met, such as the establishment of appropriate controls. Beyond these needs, it will be essential that information technology companies be shown what a key role they will have in hosting massive amounts of biomedical data and resources in ‘the cloud’. An additional interconnected hurdle relates to the legal friction that the integration of clinical and genomic data will spawn. The desire to capture economic benefits from potential discoveries associated with the data and resulting integrative network models will, if not kept in check, lead to layered legal ownership constraints that could cripple sharing. Avoiding this paralysis will require cooperation among academic institutions, nonprofit foundations, government funders and journals, which set many of the current research rules and reward structures. If we are going to be able to guide the future care of Jane Doe, we will need to engage in 217

© 2011 Nature America, Inc. All rights reserved.

C O M M E N TA R Y “institutional analysis” akin to that described by Elinor Ostrum, who won the 2009 Nobel Prize in Economic Sciences10. Within the institution of academic research, the most important cultural issues are recognition and reward. We will need to develop robust ways to recognize scientists for their work before, and independent of, publication of journal articles. For example, if we were able to publish models of disease that could be cited by others, then academic institutions might be willing to grant tenure based on the citation impact associated with the models themselves. Similarly, funding agencies might judge potential grantees by the impact of their disease models and, in parallel, set standards for how grantees should share data and models in publicly accessible ways. Such mechanisms could speed the transition to a world in which public access to data and models, as ingredients for future experiments, is not the exception but the rule. It also would greatly help if others were to follow the example set by the Wellcome Trust (London), which has opened discussions about standard legal tools that enable disease-to-therapy projects within an IP-free zone11. Here, too, patients as advocates will need to harness their energy and visibility as we navigate the delicate path to robust public clinical-genomic data access while protecting key issues of patient privacy. Conclusions In summary, the technologies are here that will entirely transform healthcare. For that reason, it is vitally important that we now focus on realigning the cultural and institutional incentives driving researchers, academic institutions and publishers. The way forward is at least threefold. First, to engage the patients, who must demand methods for data sharing that move past current privacy issues; second, to promote open-access platforms for sharing of data, models and tools; and third, to reward scientists for publication of models, not papers. If these challenges can be met, the future promises to be a world of healthcare honed by data collected from a vast majority of patients being treated in real time. At the same time, the world of drug discovery will no longer be filled by the top ten pharmaceutical giants of the present day. Instead, these titans will be complemented by a distributed chain of groups who each build a given tool, reagent or product—much closer to the archipelago of software engineers that currently provide applications for iPhones. It is indeed possible that certain forces—in pharma, in insurance or in hospital administrations—will be aligned against this view. Nonetheless, the tasks described are not impos218

sible, especially if we the people—as citizens, as scientists and as patients—are willing to experiment with how we work together. Don’t doubt that the technology will be powerful enough to provide deep understandings. Do doubt whether we are willing to take the cultural and institutional steps to fundamentally change how we work together, and how we share the data and models that will be needed to take advantage of the upcoming opportunities.

ACKNOWLEDGMENTS We are grateful to G. Siuzdak and S. Choi for helpful comments on the manuscript. T.I. is a David and Lucille Packard Fellow and was funded by a grant from the US National Institutes of Health (RR031228). S.H.F. is funded in part by the National Cancer Institute Centers for Cancer Systems Biology (CCSB) Program and the State of Washington Life Sciences Discovery Fund. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

COUNTERPOINT: Do not opine before it’s time Isaac S Kohane & David M Margulies Ms. Jane Janus stumbled into the office of Dr. Jill Askepulus pale and sweating. Before the administrative assistant could intercept the unfortunate woman, Dr. Askepulus took her friend by the arm and guided her to a soft landing on her office couch. When Jane had sufficiently recovered, Dr. Askepulus gently asked her what had happened. After a few quavering aborted attempts, she managed to whisper, “I know you warned me, but I went to the Network Integromics Clinic [NIC].” Jane was alternately glum and anxious. She explained to Jill that, of course, she knew she already had a risk of cancer because of her family history of ovarian and breast cancer, but then the NIC had shown her these complicated diagrams, which their physicians informed her demonstrated a high risk that required very close attention. They also had suggested a drug based on the genomic measurements taken at the NIC, which their models suggested could reduce her risk. Jill paused for a moment, then brought her electronic tablet over to the chair next to Jane and went over with her what appeared to be a prognostication of a track similar to those of hurricanes often seen on the video news. “Jane,” she started, “given that you are a professor of mathematics, I figured you could appreciate this. Here,” she said pointing to a 95% confidence interval, shaded in red, growing and broadening with age, “is the risk that we know you have and that increases with age for these various cancers. And here are the trajectories that are peeling Isaac S. Kohane is at the Harvard Medical School Center for Biomedical Informatics and Children’s Hospital Informatics Program, Boston, Massachusetts, USA. David M. Margulies is at Correlagen Diagnostics Inc., Waltham, Massachusetts, USA. e-mail: [email protected]

away from the main risk trajectory under the influence of lifestyle choices, which you and I have already discussed. This broad trajectory in green is the estimated effect of the drug that they suggested to you, and some variations based on different predictive models of cancer based on your genetic markers. Jane stared for a minute at this display and remarked, “I see that I can change my risk somewhat by lifestyle and I do see that this drug might be able to reduce the risk. But I was expecting that all these genomic and proteomic measurements were going to give me a much more accurate and personalized perspective of my medical future. They all seem to overlap a lot.” Jill nodded, “They might be much more accurate one day soon, but we have had considerable challenges integrating these various clinical and experimental databases and results from other high-throughput data types to come up with a more accurate prognosis and individualized therapeutic decision-making. We will get there eventually, but the science still has to be worked out and frankly we need more research to be sure our models are accurate. Right now, let’s make sure you understand the certainty or lack of it that comes from these various new data types. And let’s weigh, with common sense, the preponderance of evidence to date. I could bore you with an accounting of untold suffering that occurred as a result of an insufficiently informed use of tests such as prostate specific antigen, mammograms or urinary screening for neuroblastoma. But I won’t. Let’s talk about how we are going to make the right decision for you, with you.” An alternative view The above slightly tongue-in-cheek sequel to the scenario proposed by Friend and Ideker is provided to emphasize where we believe the

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

C O M M E N TA R Y current challenges lie. To be sure, increased openness, transparency, data sharing and academic rewards for team and multidisciplinary behavior are important ingredients in developing a vibrant and productive biomedical discovery establishment. However, they do not constitute structural impediments to the translation of genome-scale measurements into safe clinical practice. Moreover, although we have a long way to go, historical trends point to steady progress towards openness and collaboration. This includes an ever-widening fraction of open-access publications with steadily rising impact, the opening to a world of researchers of cohort studies (e.g., the Framingham Study and the Gene Expression Omnibus storing the data of over half a million microarrays), each measuring tens of thousands of genes. It includes the evidence of the increased impact and frequency of large multinational studies with hundreds of authors; historic achievements such as trial registries like clinicaltrials.gov, which even now are being upgraded to include more primary data; multiple consumer-driven, data-sharing efforts, from the corporate, such as PatientsLikeMe, to purely voluntary and extensive social network support groups. Already, biomedical research groups are discussing publication formats that follow the lead of our colleagues in astronomy that include the full data within the publication document itself 12. We can cheer on these efforts, but the translation of existing and future ‘massively parallel’ measurements to clinical-grade decision support and therapeutics remains a methodological and scientific challenge for which there has been far less progress than the sociological trends appropriately lauded by Friend and Ideker. More pressing challenges What are the components of this most pressing and thorny challenge in achieving meaningful, clinical-grade, integrative medicine that leverages the various data types enumerated by Friend and Ideker? First, we have to develop suitable technical methods and user interaction models to integrate the diverse data sources. Although there are isolated instances of integration of, for example, expression data with underlying pathways, or expression data in the context of specific somatic genome variation, there is

no general purpose architecture or model for integrating the complexity of data types with physiology and anatomy over time. Second, we have to ensure that what we know is accurate. That is, we have to clean up our existing evidentiary knowledge base. For example, of the at least 150,000 genomic variants documented to have some import to disease, a substantial minority have not been reproduced or have been contradicted by subsequent reports. Third, we have to ensure that we know what is known. In the context of a medical education system that is already straining to keep physicians informed of best practices using only a few thousand clinical variables,

Increased openness, transparency, data sharing and academic rewards for team and multidisciplinary behavior do not constitute the primary structural impediments to the translation of genomic measurements into safe clinical practice. the challenge of supporting sound and efficient decision-making in the context of millions of variants will require substantial progress in data reduction, user interfaces and automated support. Fourth, we have to know whether we can safely proceed to clinical decision-making from computer models that are not completely based on human clinical trials, randomized or observational. That is, can our models achieve the same mechanistic and predictive qualities as the Henderson-Hasselbalch equation for acid-base equilibrium, the Frank-Starling Curve for cardiac contractility or at least the Framingham cardiovascular risk scores? If not, are they only useful for hypothesis exploration rather than clinical care? Breakthroughs in both measurement and modeling technology may be required to achieve clinical-grade soundness of our models. Fifth, there will need to emerge regulatory clarity around the use of data displays of this complexity. Who will decide whether

nature biotechnology volume 29 number 3 march 2011

what we think we know is safe? What are the boundaries of the US Food and Drug Administration’s (FDA) so-called ‘IVDMIA’ (in vitro diagnostic multivariate assay) threshold? How will the regulatory framework of the FDA and its international analogs cope with models as complex as those of in silico airplane design? Finally, how much better is our new knowledge than older knowledge? When is the incremental benefit of a genomic variant(s) or gene expression profile relative to a family history or classic histopathology insufficient and when does it add rather than subtract variance? If we are able to rationalize the selection of cancer chemotherapeutic agents by integrating information about responsiveness of cells with specific cell expression profiles, that would be an important ‘emergent’ benefit of deep integration. But it is important that we identify potential transformative benefits to focus and prioritize data integration efforts. The clinical perspective exemplified by these questions poses substantial challenges. We do not doubt that our biomedical research community is up to successfully addressing them, some even in the very near term. Like our colleagues, we are excited to be able to collaborate in integromic research that we are convinced will benefit many who suffer from disease. And like Friend and Ideker, we are optimistic that the trends to collaboration and transparency, already underway, can only help. Competing Financial Interests The authors declare no competing financial interests. 1. Fisher, J. & Henzinger, T.A. Nat. Biotechnol. 25, 1239– 1249 (2007). 2. Friend, S.H. Clin. Pharmacol. Ther. 87, 536–539 (2010). 3. Chuang, H.Y., Hofree, M. & Ideker, T. Annu. Rev. Cell Dev. Biol. 26, 721–744 (2010). 4. Ideker, T. & Sharan, R. Genome Res. 18, 644–652 (2008). 5. Rosenquist, J.N., Fowler, J.H. & Christakis, N.A. Mol. Psych. published online, doi: 10.1038/mp.2010.13 (16 March 2010). 6. Wagner, J.K. Am. J. Hum. Genet. 87, 451–456 (2010). 7. David, E., Tramontin, T. & Zemmel, R. Nat. Rev. Drug Discov. 8, 609–610 (2009). 8. Weigelt, J. EMBO Rep. 10, 941–945 (2009). 9. Romero, K. et al. Clin. Pharmacol. Ther. 86, 365–367 (2009). 10. Ostrum, E. Understanding Institutional Diversity (Princeton University Press, Princeton, NJ, USA, 2005). 11. Friend, S.H. Scientist 24, 22–32 (2010). 12. Goodman, A.A. et al. Nature 457, 63–66 (2009).

219

c o m m e n ta r y

Biomedical technology and the clinic of the future © 2011 Nature America, Inc. All rights reserved.

Technology pioneers trade views with a clinician and an entrepreneur on the likely impact of large-scale systems technology in healthcare.

T

o date, large-scale ’omics data sets and systems approaches in biology have had a relatively minor impact on the practice of medicine. As new technology brings individual genome sequencing closer to reality and large-scale biology continues to progress, opportunities are likely to open up in disease prediction, prevention, diagnosis and treatment. Here the views of two researchers on the potential of disruptive biomedical technologies in clinical practice are contrasted with the perspectives of a clinician and an entrepreneur in commercial clinical information technology.

POINT: Are we prepared for the future doctor visit? Stephen H Friend & Trey Ideker Imagine the following visit to the doctor’s office, which, although fictitious, is based on technologies that are emerging or already available. A patient, Jane Doe, enters the clinic for a routine physical exam. Today, at least seven parameters would be registered upon her admittance: sex, age, height, weight, temperature, pulse rate and blood pressure (itself a pair of values). But in the future when Jane registers, this set of routine measurements will have expanded enormously (Table 1). Tomorrow’s routine checkup Either on this visit or a previous one, Jane’s full genome has been sequenced, noninvasively, using a buccal swab. At the same time, and optionally on every visit, the nurse has sampled and sequenced the metagenome of Stephen H. Friend is at Sage Bionetworks, Seattle, Washington, USA. Trey Ideker is in the Departments of Medicine and Bioengineering, University of California, La Jolla, California, USA, and at The Institute for Genomic Medicine, University of California, La Jolla, California, USA. e-mail: [email protected] or [email protected]

the microbiome pool resident in the patient’s mucosal and gastrointestinal cavities, providing a detailed characterization of the population of microbes commensal with the human host. Messenger RNA, microRNA, proteome and metabolome profiles may be gathered from urine and, if necessary, whole blood and other tissues. Finally, in addition to height and weight, a large panel of physiological parameters and images is monitored, capturing detailed information about respiration, endocrine function, cardiac and brain activity, and so on. Another key development that will transform Jane’s visit to the clinic is deeper data integration. All of the newly gathered information are banked in a unified electronic medical record, which uses a relational database to establish cross-references among the different data types. The new information augments the history of data gathered on previous visits, including all medical treatments and outcomes accumulated over the patient’s lifetime. Crucially, the new data are then integrated with a library of biological network models spanning multiple levels and scales (Fig. 1). First is the network of functional and molecular interactions—a.k.a. the molecular wiring diagram—providing a modular, hierarchical

nature biotechnology volume 29 number 3 march 2011

and executable view1 of the cellular processes underlying human health and disease. Such networks are being assembled from diverse large- and small-scale experiments performed over decades of systems biology and biomedical research, providing an up-to-date representation of current knowledge in the field2,3. A second type of network model will represent the relevant nosology, which maps relationships between diseases based on their similarities in etiology, pathogenesis and symptoms. Related to this will be another network—that of pharmacologic treatments, which provides rich information about the different protocols and drugs that are available along with their quantitative inter-relationships. One more important network will be the patient’s extended social network and pedigree, which will be available along with references to the integrated medical records of friends and relatives. This social network documents significant personal relationships in Jane’s life, weighted by importance and, subject to privacy concerns, gathered from social networking websites, personal address books, geographical co-location data, as well as cell phone and e-mail usage. The pedigree provides a complementary set of social relationships that have a genetic basis. The benefits of these network models to Jane are severalfold. First, they integrate an array of different lines of evidence for health or disease, enabling the formulation of compound biomarkers that are combinations or functions of many simultaneous readouts. Such compound biomarkers can be more robust than biomarkers based on individual genes, proteins or metabolites4. Second, the networks provide a natural interpretation of the mechanisms behind Jane’s present and future conditions, in contrast to current biomarkers that often have little relation to the actual cause of disease. Third, Jane’s data and outcomes can be dynamically analyzed and reintegrated to 215

© 2011 Nature America, Inc. All rights reserved.

C O M M E N TA R Y “institutional analysis” akin to that described by Elinor Ostrum, who won the 2009 Nobel Prize in Economic Sciences10. Within the institution of academic research, the most important cultural issues are recognition and reward. We will need to develop robust ways to recognize scientists for their work before, and independent of, publication of journal articles. For example, if we were able to publish models of disease that could be cited by others, then academic institutions might be willing to grant tenure based on the citation impact associated with the models themselves. Similarly, funding agencies might judge potential grantees by the impact of their disease models and, in parallel, set standards for how grantees should share data and models in publicly accessible ways. Such mechanisms could speed the transition to a world in which public access to data and models, as ingredients for future experiments, is not the exception but the rule. It also would greatly help if others were to follow the example set by the Wellcome Trust (London), which has opened discussions about standard legal tools that enable disease-to-therapy projects within an IP-free zone11. Here, too, patients as advocates will need to harness their energy and visibility as we navigate the delicate path to robust public clinical-genomic data access while protecting key issues of patient privacy. Conclusions In summary, the technologies are here that will entirely transform healthcare. For that reason, it is vitally important that we now focus on realigning the cultural and institutional incentives driving researchers, academic institutions and publishers. The way forward is at least threefold. First, to engage the patients, who must demand methods for data sharing that move past current privacy issues; second, to promote open-access platforms for sharing of data, models and tools; and third, to reward scientists for publication of models, not papers. If these challenges can be met, the future promises to be a world of healthcare honed by data collected from a vast majority of patients being treated in real time. At the same time, the world of drug discovery will no longer be filled by the top ten pharmaceutical giants of the present day. Instead, these titans will be complemented by a distributed chain of groups who each build a given tool, reagent or product—much closer to the archipelago of software engineers that currently provide applications for iPhones. It is indeed possible that certain forces—in pharma, in insurance or in hospital administrations—will be aligned against this view. Nonetheless, the tasks described are not impos218

sible, especially if we the people—as citizens, as scientists and as patients—are willing to experiment with how we work together. Don’t doubt that the technology will be powerful enough to provide deep understandings. Do doubt whether we are willing to take the cultural and institutional steps to fundamentally change how we work together, and how we share the data and models that will be needed to take advantage of the upcoming opportunities.

ACKNOWLEDGMENTS We are grateful to G. Siuzdak and S. Choi for helpful comments on the manuscript. T.I. is a David and Lucille Packard Fellow and was funded by a grant from the US National Institutes of Health (RR031228). S.H.F. is funded in part by the National Cancer Institute Centers for Cancer Systems Biology (CCSB) Program and the State of Washington Life Sciences Discovery Fund. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests.

COUNTERPOINT: Do not opine before it’s time Isaac S Kohane & David M Margulies Ms. Jane Janus stumbled into the office of Dr. Jill Askepulus pale and sweating. Before the administrative assistant could intercept the unfortunate woman, Dr. Askepulus took her friend by the arm and guided her to a soft landing on her office couch. When Jane had sufficiently recovered, Dr. Askepulus gently asked her what had happened. After a few quavering aborted attempts, she managed to whisper, “I know you warned me, but I went to the Network Integromics Clinic [NIC].” Jane was alternately glum and anxious. She explained to Jill that, of course, she knew she already had a risk of cancer because of her family history of ovarian and breast cancer, but then the NIC had shown her these complicated diagrams, which their physicians informed her demonstrated a high risk that required very close attention. They also had suggested a drug based on the genomic measurements taken at the NIC, which their models suggested could reduce her risk. Jill paused for a moment, then brought her electronic tablet over to the chair next to Jane and went over with her what appeared to be a prognostication of a track similar to those of hurricanes often seen on the video news. “Jane,” she started, “given that you are a professor of mathematics, I figured you could appreciate this. Here,” she said pointing to a 95% confidence interval, shaded in red, growing and broadening with age, “is the risk that we know you have and that increases with age for these various cancers. And here are the trajectories that are peeling Isaac S. Kohane is at the Harvard Medical School Center for Biomedical Informatics and Children’s Hospital Informatics Program, Boston, Massachusetts, USA. David M. Margulies is at Correlagen Diagnostics Inc., Waltham, Massachusetts, USA. e-mail: [email protected]

away from the main risk trajectory under the influence of lifestyle choices, which you and I have already discussed. This broad trajectory in green is the estimated effect of the drug that they suggested to you, and some variations based on different predictive models of cancer based on your genetic markers. Jane stared for a minute at this display and remarked, “I see that I can change my risk somewhat by lifestyle and I do see that this drug might be able to reduce the risk. But I was expecting that all these genomic and proteomic measurements were going to give me a much more accurate and personalized perspective of my medical future. They all seem to overlap a lot.” Jill nodded, “They might be much more accurate one day soon, but we have had considerable challenges integrating these various clinical and experimental databases and results from other high-throughput data types to come up with a more accurate prognosis and individualized therapeutic decision-making. We will get there eventually, but the science still has to be worked out and frankly we need more research to be sure our models are accurate. Right now, let’s make sure you understand the certainty or lack of it that comes from these various new data types. And let’s weigh, with common sense, the preponderance of evidence to date. I could bore you with an accounting of untold suffering that occurred as a result of an insufficiently informed use of tests such as prostate specific antigen, mammograms or urinary screening for neuroblastoma. But I won’t. Let’s talk about how we are going to make the right decision for you, with you.” An alternative view The above slightly tongue-in-cheek sequel to the scenario proposed by Friend and Ideker is provided to emphasize where we believe the

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

C O M M E N TA R Y current challenges lie. To be sure, increased openness, transparency, data sharing and academic rewards for team and multidisciplinary behavior are important ingredients in developing a vibrant and productive biomedical discovery establishment. However, they do not constitute structural impediments to the translation of genome-scale measurements into safe clinical practice. Moreover, although we have a long way to go, historical trends point to steady progress towards openness and collaboration. This includes an ever-widening fraction of open-access publications with steadily rising impact, the opening to a world of researchers of cohort studies (e.g., the Framingham Study and the Gene Expression Omnibus storing the data of over half a million microarrays), each measuring tens of thousands of genes. It includes the evidence of the increased impact and frequency of large multinational studies with hundreds of authors; historic achievements such as trial registries like clinicaltrials.gov, which even now are being upgraded to include more primary data; multiple consumer-driven, data-sharing efforts, from the corporate, such as PatientsLikeMe, to purely voluntary and extensive social network support groups. Already, biomedical research groups are discussing publication formats that follow the lead of our colleagues in astronomy that include the full data within the publication document itself 12. We can cheer on these efforts, but the translation of existing and future ‘massively parallel’ measurements to clinical-grade decision support and therapeutics remains a methodological and scientific challenge for which there has been far less progress than the sociological trends appropriately lauded by Friend and Ideker. More pressing challenges What are the components of this most pressing and thorny challenge in achieving meaningful, clinical-grade, integrative medicine that leverages the various data types enumerated by Friend and Ideker? First, we have to develop suitable technical methods and user interaction models to integrate the diverse data sources. Although there are isolated instances of integration of, for example, expression data with underlying pathways, or expression data in the context of specific somatic genome variation, there is

no general purpose architecture or model for integrating the complexity of data types with physiology and anatomy over time. Second, we have to ensure that what we know is accurate. That is, we have to clean up our existing evidentiary knowledge base. For example, of the at least 150,000 genomic variants documented to have some import to disease, a substantial minority have not been reproduced or have been contradicted by subsequent reports. Third, we have to ensure that we know what is known. In the context of a medical education system that is already straining to keep physicians informed of best practices using only a few thousand clinical variables,

Increased openness, transparency, data sharing and academic rewards for team and multidisciplinary behavior do not constitute the primary structural impediments to the translation of genomic measurements into safe clinical practice. the challenge of supporting sound and efficient decision-making in the context of millions of variants will require substantial progress in data reduction, user interfaces and automated support. Fourth, we have to know whether we can safely proceed to clinical decision-making from computer models that are not completely based on human clinical trials, randomized or observational. That is, can our models achieve the same mechanistic and predictive qualities as the Henderson-Hasselbalch equation for acid-base equilibrium, the Frank-Starling Curve for cardiac contractility or at least the Framingham cardiovascular risk scores? If not, are they only useful for hypothesis exploration rather than clinical care? Breakthroughs in both measurement and modeling technology may be required to achieve clinical-grade soundness of our models. Fifth, there will need to emerge regulatory clarity around the use of data displays of this complexity. Who will decide whether

nature biotechnology volume 29 number 3 march 2011

what we think we know is safe? What are the boundaries of the US Food and Drug Administration’s (FDA) so-called ‘IVDMIA’ (in vitro diagnostic multivariate assay) threshold? How will the regulatory framework of the FDA and its international analogs cope with models as complex as those of in silico airplane design? Finally, how much better is our new knowledge than older knowledge? When is the incremental benefit of a genomic variant(s) or gene expression profile relative to a family history or classic histopathology insufficient and when does it add rather than subtract variance? If we are able to rationalize the selection of cancer chemotherapeutic agents by integrating information about responsiveness of cells with specific cell expression profiles, that would be an important ‘emergent’ benefit of deep integration. But it is important that we identify potential transformative benefits to focus and prioritize data integration efforts. The clinical perspective exemplified by these questions poses substantial challenges. We do not doubt that our biomedical research community is up to successfully addressing them, some even in the very near term. Like our colleagues, we are excited to be able to collaborate in integromic research that we are convinced will benefit many who suffer from disease. And like Friend and Ideker, we are optimistic that the trends to collaboration and transparency, already underway, can only help. Competing Financial Interests The authors declare no competing financial interests. 1. Fisher, J. & Henzinger, T.A. Nat. Biotechnol. 25, 1239– 1249 (2007). 2. Friend, S.H. Clin. Pharmacol. Ther. 87, 536–539 (2010). 3. Chuang, H.Y., Hofree, M. & Ideker, T. Annu. Rev. Cell Dev. Biol. 26, 721–744 (2010). 4. Ideker, T. & Sharan, R. Genome Res. 18, 644–652 (2008). 5. Rosenquist, J.N., Fowler, J.H. & Christakis, N.A. Mol. Psych. published online, doi: 10.1038/mp.2010.13 (16 March 2010). 6. Wagner, J.K. Am. J. Hum. Genet. 87, 451–456 (2010). 7. David, E., Tramontin, T. & Zemmel, R. Nat. Rev. Drug Discov. 8, 609–610 (2009). 8. Weigelt, J. EMBO Rep. 10, 941–945 (2009). 9. Romero, K. et al. Clin. Pharmacol. Ther. 86, 365–367 (2009). 10. Ostrum, E. Understanding Institutional Diversity (Princeton University Press, Princeton, NJ, USA, 2005). 11. Friend, S.H. Scientist 24, 22–32 (2010). 12. Goodman, A.A. et al. Nature 457, 63–66 (2009).

219

F E AT U R E

Five more years of Nature Biotechnology research Monya Baker & Laura DeFrancesco Authors of the past five years’ most highly cited research articles discuss their work and new directions in their respective areas.

© 2011 Nature America, Inc. All rights reserved.

F

ive years ago, one of the ways we celebrated our tenth anniversary as Nature Biotechnology was by looking back at the most highly cited papers of the previous decade. As we enter our fifteenth year, we take another look back, this time just five years. During this brief interval, assumptions about how stem cells can be generated and differentiated were revised, new technologies emerged in protein analysis and sequencing allowing ’omics approaches to move from species to individuals to differences between individuals, and computer models got better at predicting cellular behavior according to our knowledge of biological pathways. Because of space constraints, we do not cover all the biotech advances from the past five years; instead, the vignettes below provide a selection of the most important advances published in our pages with a nod to the implications and further applications of the work. Safer iPS cells One of the most important discoveries since the creation of induced pluripotent stem (iPS) cells began as a negative control experiment. To make iPS cells, Shinya Yamanaka of Kyoto University showed that inserting just four transcription factors into cultured mouse fibroblasts could make them behave like embryonic stem cells1. All four proteins were considered essential, but they also presented risks for any potential therapeutic use of reprogrammed cells. Indeed, one of them, c-Myc, is a well-known oncogene. Monya Baker is Technology Editor for Nature and Nature Methods. Laura DeFrancesco is Senior Editor at Nature Biotechnology.

In experiments aimed at understanding the role of Myc in reprogramming, Yamanaka’s laboratory set up experiments to study c-Myc compared with other members of the gene family and included a negative control without Myc. “To our surprise, we obtained iPS cells, even without Myc, albeit with a very low efficiency,” Shinya Yamanaka says Yamanaka. says the Myc genes Yamanaka went may hold the key on to make Myc-free to other aspects of reprogramming. iPS cells from both mouse and human cells2. To determine whether the presence of c-Myc increases the chance that the iPS cells will form tumors, Yamanaka’s group mixed mouse iPS cells into early-stage mouse embryos, which grew into chimeric mice with many tissues derived from the iPS cells. Six of 37 mice made from Myc-containing iPS cells died from tumors, whereas none of the 26 mice generated from Myc-free iPS cells did. Since this work was done, researchers have been looking hard for ways to make iPS cells without inserting any transgenes at all; for example, by using combinations of small molecules, integration-free DNA vectors, synthetic RNA encoding the transcription factors and even modified versions of the proteins themselves. But, as Yamanaka warns, merely eliminating Myc or integrating factors does not guarantee the safety of iPS cells. Some iPS cell–derived tumors could be caused by reactivating the gene encoding Myc, others by residual undifferentiated cells that resist cues for differentiation, a property that correlates with the tissue from which the iPS cells were originally derived3. And, according to Yamanaka, there may be other causes, so the safety of iPS -cells needs to be rigorously assessed before they can be used in the clinic.

nature biotechnology volume 29 number 3 march 2011

Meanwhile, Yamanaka says the Myc gene family may hold the key to other aspects of reprogramming. Last year, his group found that mutant versions of c-Myc, as well as the related protein L-Myc, could all promote reprogramming to iPS cells more efficiently than c-Myc, even though they have little transformation activity on their own4. Moreover, L-Myc does not promote tumor formation in mice. These results show that Myc family genes can promote reprogramming independent of their ability to make tumors. How this happens is still unclear. There are several possible mechanisms, says Yamanaka. Myc could affect chromatin architecture and function synergistically with other transcription factors; it could also regulate genes that in turn regulate cell differentiation. By delving further into the molecular mechanisms of reprogramming and identifying more factors, generating safer iPS cells may become faster and more predictable. And perhaps the next discovery will also come from negative controls with surprising results. Why some tumors don’t starve After a bit of evangelism and enough clinical data, the idea seemed obvious: drugs that deprive tumors of their blood supply should shrink them away to nothing. In practice, though, some tumors stayed put. Though resistance in tumors is common against toxic drugs that attack cancer cells directly, it was not expected to develop against angiogenesis inhibitors, which target presumably normal vascular epithelial cells rather than fast-dividing, quickly mutating cancerous ones. Vascular endothelial growth factor (VEGF), the signaling molecule that summons blood 221

© 2011 Nature America, Inc. All rights reserved.

f e at u r e vessels, was discovered in 1989 by Napoleone Ferrara, a cancer biologist at Genentech (South San Francisco, CA, USA). The discovery was actually a side Napoleone Ferrara was project; Genentech’s given the freedom to philosophy was to pursue his interest in give scientists freetumor angiogenesis. dom to pursue their own interests parttime. This paid off, as Avastin (bevacizumab), a humanized monoclonal antibody drug directed against VEGF that blocks this signaling, was approved in 2004 for metastatic colon cancer. When resistance to Avastin was observed, Ferrara decided to hunt down the mechanism responsible. Because Avastin stops cancer growth only in vivo, not in vitro, he reasoned biology outside the tumors had to be involved. Ferrara and his colleagues identified cell lines that, when grown in mice, produced tumors that were either sensitive or refractory to the anti-VEGF antibody. Thus began a long series of animal experiments. His team quickly ruled out that a specific immune attack was involved: anti-VEGF antibodies inhibited sensitive tumors in immunodeficient mice, and refractory tumors behaved the same in immunocompetent and immunodeficient mice. Ferrara suspected that studying inflammation might yield the answer to resistance; many reports at the time showed that inflammation played a role in cancer, but no one had looked at whether it might play a role in response to anti-VEGF treatments. In fact, regardless of treatment with the anti-VEGF antibody, refractory tumors contained many more bone marrow mononuclear cells (BMMNCs) and more vasculature than sensitive tumors, even though the BMMNCs were not contributing to blood vessels, at least not directly. To figure out what role the BMMNCs were playing, Ferrara’s team exposed the cells to extracts from sensitive and refractory tumors. After exposure to extracts from resistant tumors, BMMNCs promoted growth of sensitive tumors. Somehow, the refractory tumors ‘instructed’ BMMNCs to stimulate tumor growth. To unravel the mechanisms, Ferrara and his collaborators first had to figure out what type of BMMNCs were responsible. This work eventually fingered a subset of myeloid cells. The myeloid cells were promoting blood vessel growth by secreting factors in addition to VEGF. Suppressing myeloid cells made Avastin-refractory tumors more sensitive5. 222

Identifying the specific factors required additional work. In brief, either stroma or tumor cells themselves can produce colonystimulating factors (in this case, granulocytemacrophage colony-stimulating factor and granulocyte colony-stimulating factor) that mobilize myeloid cells, which in turn promote angiogenesis through a protein called BV8 (refs. 6,7). Although these experiments were done in mice, evidence is growing, says Ferrara, that a similar process also occurs in humans, though it is still unclear which subset of myeloid cells mediates this effect. (Right now, most scientists are betting on neutrophils.) In addition, several other reasons have been proposed for why tumors don’t respond to anti-angiogenesis drugs, such as poor delivery of the drug to the tumor or the presence of redundant signaling pathways promoting angiogenesis. Although the mechanism of resistance is complicated, Ferrara says the results of antiVEGF treatments has still come as a shock. In the early 1990s, those targeting tumor angiogenesis assumed that to see any efficacy, they would need to block many factors, until it became clear that knocking out VEGF alone could stop the blood supply at least in some cases. “What surprised me was that VEGF could be such an important molecule.” Next-generation sequencing captures methylation variation People are thinking less and less about ‘the’ human genome and ‘the’ human epigenome, says George Church of Harvard University (Cambridge, MA, USA). Instead, researchers are trying to home in on variation, comparing genomes and epigenomes across samples. In 2009, two papers in Nature Biotechnology described efficient ways to do so. One of the techniques described by Church and his colleagues relies on restriction enzymes that preferentially cut unmethylated sequences. Intense sequencing around the restriction sites can compare nearly 1.5 million restriction sites between samples with considerably less effort than whole-genome approaches require. A less comprehensive but more reproducible method uses 10,000 ‘padlock probes’, a clever technique for allowing many different sequences of DNA to be amplified in the same reaction vessel using a common set of primers8. To use padlock probes to detect methylation, DNA is first treated with bisulfite, which chemically converts unmethylated cytosine so

that it registers as thymine. It was a risky project, says Kun Zhang, who completed his postdoc with Church and is now at the University of California, San Diego. “Using padGeorge Church says that commonly held lock probes to detect but false beliefs methylation was among scientists really a bet. Once impede technology DNA is converted, development. the majority of the genome is made of just three members of the genetic code, so the genome complexity goes down dramatically.” In a separate study, Zhang designed some 30,000 probes to assess methylation at 66,000 sites in the genome and identified methylation differences between pluripotent and specialized cells9. The key to success was being able to make many padlock probes9. For this, Church and Zhang hit upon the idea of using microarrays as miniature DNA synthesizers: rather than using the array to detect DNA in a sample, the array is used to make probes, which are then clipped off. It wasn’t easy, recalls Church. “Murphy’s law was in full force. Almost every aspect would fail. Some chips had a 1% yield or a nonrandom yield, so you’d get some element overrepresented by a couple factors of ten.” Zhang began a collaboration with Agilent Technologies (Santa Clara, CA, USA), and the numbers of probes grew from 10,000 to 20,000 to over 55,000, recalls Zhang. “Once we had the capability of making tens of thousands of probes relatively easily, we could start to apply probes to a real study.” Zhang is currently working on ways to increase throughput still further, and Kun Zhang says that points to several using padlock probes improvements: the to detect methylation original protocol was really a bet. required ten purification steps, from introducing the probes to sequencing the new libraries, says Zhang. Now that is down to three steps. And the padlock probes are much improved as well. “Initially we didn’t understand all the parameters, so what we did was just brute force,” says Zhang. However, after measuring the performance of probes and feeding that data into a machine-learning algorithm, probes with much better performance

volume 29 number 3 march 2011 nature biotechnology

f e at u r e

© 2011 Nature America, Inc. All rights reserved.

are now being designed, he says. In the first papers, only ~50% of the reads were useful. Now, he says, ~80% are useful, a rate much closer to regular genome sequencing. The coverage of the genome is becoming both more complete and more uniform, he says. The idea of using padlock probes comprehensively was once considered unthinkable, if only for the difficulty of the DNA synthesis, says Church. It’s just one example where commonly held but false beliefs among scientists impede technology development, he says. “Sometimes they’ll say something is impossible, but if you drill down they say that it’s really expensive, and if you drill further, they say it’s really expensive now.” Nanoprobes, over and out, but not finished Getting an imaging probe out of the body can be just as important as putting it in. That’s especially true for quantum dots, which fluoresce brightly but are made from toxic elements. Thus, anyone interested in using dots for clinical applications needs to understand how they move through the body. Researchers led by John Frangioni at Beth Israel Deaconess Medical Center in Boston and Moungi Bawendi at the Massachusetts Institute of Technology in Cambridge described how they tracked dots with a wide range of physical parameters and showed that, at least in mice, quantum dots can be cleared through the kidney and into urine, provided that the dots are fewer than 6 nm across, about half the size of an antibody10. But quantum dots pose so many problems for human applications that Frangioni has decided to focus instead on alternative imaging agents. Even the smallest dots don’t clear as completely or quickly as he would like. The color of John Frangioni says, nanodots depends on “We hope our work both size and compowill make others sition, and to get the think about what’s most useful colors for potentially clinically imaging, very small viable and what’s a dead-end.” quantum dots must

be made of some of the most toxic elements. Furthermore, staying below 6 nm is difficult once dots are coated to prevent leaching and functionalized with molShuming Nie thinks ecules to target them that gold nanoparticles to particular tissues. will have multiple Despite what clinical uses. enthusiasts say, quantum dots are not particularly bright once they are in the body, says Frangioni. It is true that quantum dots outside the body can be stimulated with a broad bandwidth of light frequencies, but in scattering tissue, absorption is usually confined to a narrow, redder band. And on a per-volume basis, small molecules are brighter than quantum dots, says Frangioni. Thus, after spending almost a decade working on quantum dots for medical imaging11, he has decided to devote his efforts to small-molecule dyes instead. “The periodic table is just not in our favor,” he says. “We hope our work will make others think about what’s potentially clinically viable and what’s a deadend.” Other researchers are continuing work on nanotech imaging probes for clinical use, however, and one of these is Shuming Nie, who has appointments at both Emory University and the Georgia Institute of Technology in Atlanta. In 2008, he reported a gold-based nanoparticle that could be conjugated to an antibody and used to detect cancer12. Before this paper, the brightest tools used in cancer imaging were quantum dots. Not only is gold already used in humans, Nie was able to produce gold nanoparticles 200 times brighter than any previously reported. The brightness of the nanoparticles is a function of the gold itself as well as small-molecule dyes adsorbed to it, explains Nie. One potential application of these particles is demarcating the edges of tumors, allowing surgeons to be sure they remove all the cancerous tissues. Last year, his laboratory described a proof-of-principle experiment showing that a handheld device termed a SpectroPen could be used along with these nanoparticles to visualize the borders of tumors implanted in mice13. Nie thinks the particles could be used in other ways as well. The surface-enhancing Raman scattering dyes adsorbed to the particles come in a variety of colors, and so more than one kind of molecule could be imaged simultaneously within the same animal. It’s even possible, he says, that rare cells within an animal could be tagged with these nanoparticles and imaged over time. There are

nature biotechnology volume 29 number 3 march 2011

some problems that will have to be assessed or mitigated before clinical trials, says Nie. For instance, the nanoparticles accumulate in the liver and other organs. But Nie believes such issues can be overcome eventually. “I agree that for systemic applications quantum dots are not the way to go. These [gold] particles are really different.” Picking the right kinase Many commercially successful kinase inhibitors have offtarget effects, but until recently few researchers realized how extensive these could be. A few years ago, Ambit Biosciences (San Diego) introduced the idea of a ‘selectivity score’ based on the company’s technology for profiling hundreds of kinases against thousands of compounds14,15. Using data collected by the company’s KinomeScan profiling technology, these analyses produce kinase trees covered with dots, which show each compound’s selectivity and potency across the universe of kinases, or the kinome. These images are now iconic, but initially, the data were visualized quite differently, says author Patrick Zarrinkar. For example, data for multiple compounds would be shown on a single tree image. Then a collaborator requested an image showing data for just a single compound. “As soon as I saw that, it was clear that this would be the way we should display the profiling results,” he recalls. Pharmaceutical teams were surPatrick Zarrinkar says, prised to learn that “It was so many small molecules they had breakthroughs, we’d been working with just put our heads down and continue to work.” for years were less selective than they’d assumed. “They said ‘we thought we understood this compound, but it does all these other things’,” recalls Zarrinkar. “Before these papers, it was normal for people to take an inhibitor and test against five or six kinases and assume it was selective,” says Zarrinkar, now at Prognosys Biosciences (La Jolla, CA, USA). “You can’t do that today.” Ambit’s and others’ data showing that clinical kinase inhibitors hit multiple targets startled researchers into seeking broader profiling of potential drug candidates. This has 223

© 2011 Nature America, Inc. All rights reserved.

f e at u r e changed how people approach drug development. Now, says Zarrinkar, pharmaceutical companies are not just looking for molecules that hit a predefined target, but are asking which interesting targets can be hit by their molecules. “You can make decisions not just based on the biology of the target but also based on chemistry—which targets are hit by available compounds,” says Zarrinkar. This is exactly what Ambit did when selecting its own clinical candidate, now in phase 2 trials for acute myelogenous leukemia. After screening its compound library against a panel of kinases, the company picked a disease target, FMS-like tyrosine kinase 3 (FLT3), for which their library had the best hit. Some advisors discouraged them, citing evidence from FLT3 inhibitors already in the clinic that suggested it wasn’t a good target. After comparing the profiles between the compounds that were in the clinic and the hit Ambit had identified, the company decided its compound’s selectivity and potency profile was more promising and decided to go forward. Subsequently, it sold the KinomeScan technology to DiscoveRx (Fremont, CA, USA) and is focusing on clinical development of this and other drugs. Ironically, the decision to develop the kinase profiling assay in the first place may have benefited from having less information, says Zarrinkar. Ambit had been developing phagedisplay technology for a different purpose and realized it could be used to quickly build extensive panels of kinases for assays. Fortunately, company researchers were unaware how difficult some kinases could be to produce, says Zarrinkar. “Everyone told us we were crazy, but we didn’t know kinases, so we thought we could figure it out.” They did, though with considerably more effort than they had expected. Progress was slow and incremental, says Zarrinkar. “It was so many small breakthroughs, we’d just put our heads down and continue to work.” Eventually, team members including David Lockhart and Daniel Treiber realized they had to be very careful about which section of the kinase was displayed and how. In fact, many assays in the current version of KinomeScan rely on proteins produced in mammalian cells rather than on phage. Each profiling assay measures the binding equilibrium between a kinase, a test compound, and a known, small-molecule ligand. A biotinylated ligand is immobilized on streptavidincoated beads and introduced to a free test compound and a kinase. Kinases are tagged with a stretch of DNA, and so quantitative PCR can assess the number of bound kinases sensitively and over a broad dynamic range. The binding properties of kinases to the immobilized ligands are used to assess the test compounds’ affinity. 224

These assays can be easily multiplexed and allow tens of thousands of compounds to be assessed against hundreds of kinases quickly. Zarrinkar says that this profiling technology represents one way in which the human genome project has directly benefited drug discovery. Rather than enabling the discovery of novel targets, the genome is enabling the discovery of novel compounds. Systematic development of the kinase panel was made possible by the enumeration of the kinome by other scientists, which in turn required their having the human genome sequence16. The profiling technology then enabled a novel approach to kinase inhibitor discovery17, says Zarrinkar. “There’s a straight line from the human genome to the kinome to KinomeScan technology to novel drugs.” Quality control for microarrays In the early twentyfirst century, after a few years of unbridled excitement, doubts began to be raised about the validity of microarray results. Different laboratories, often using different platforms, were getting different results. Still, no one could deny that the technology offered powerful information: a read on what genes were being turned on, off, up and down inside cells. Even as the community was questioning the technology, the US Food and Drug Administration (FDA) was investing considerable effort embracing pharmacogenomics and thinking about how to incorporate genomic data submissions associated with drug applications. Leming Shi, a computational chemist at the FDA, believed something had to be done. Shi organized a large-scale, quality control effort to solve these problems. He convinced Applied Biosystems (Foster City, CA, USA) to supply quantitative PCR data—the gold standard for gene expression measurement— to compare multiparallel data from DNA microarrays. Several microarray manufacturers as well as academic scientists joined the effort. Eventually, Shi attracted 137 participants from 51 organizations; people with competing interests and different scientific views were able to work together under the leadership of a federal agency. In addition to supplying microarrays and reagents, manufacturers even provided information about proprietary probe sequences. Collectively, laboratories analyzed two common reference RNA samples on seven microarray platforms, plus three other technologies for independently assessing expression.

Ultimately, the project involved over 1,000 microarrays and represented a cumulative investment of around $2 million, says Shi. Whereas the project covered many experimental and analytical facets of quality control, how to select informative genes became the most controversial. From the beginning, Shi says, the problem was figuring out which subset of genes provides biological insights. Microarrays simultaneously monitor the activity of tens of thousands of genes in one experiment, and then individual genes must be compared between different sets of samples. Researchers had developed different ways to decide which genes to focus on. One leading approach picked genes based on a metric called the t-statistic, a measure derived from variation in gene expression measurements. Unfortunately this varied substantially from experiment to experiment, so Shi didn’t think this choice made sense. “People were dominated by pure statistical consideration without thinking of what the technology was trying to tell us,” he says. He asked participating scientists to analyze data in whatever way seemed best. “The only data that was reproducible was if you looked at the magnitude of the difference between the two conditions and then ranked all the genes on the chip based on the fold-change,” Shi says. Using those criteria, data were much more reproducible across different laboratories and across different platforms. The group, which named itself the MicroArray Quality Control Consortium (MAQC), published its results in a research article18 with five companion articles in 2006. But Shi hadn’t appreciated how invested many scientists were in the t-statistic. When presenting an initial analFDA computational ysis of his results, one chemist Leming Shi professor interrupted organized a largehis talk, demanding scale quality effort to address the problems that he recheck his with microarray results; Shi was told experiments. to restrict the MAQC studies to analyzing the amount of technological noise and to steer clear of evaluating statistical methods. Little by little, however, the statisticians within the MAQC were won over. A companion MAQC paper by Shi describing the rationale for ranking genes by fold-change was rejected in 2006 as unsound; two years later, another set of reviewers rejected the same paper as obvious. Meanwhile, the MAQC has moved on. A recent publication in Nature Biotechnology

volume 29 number 3 march 2011 nature biotechnology

f e at u r e

© 2011 Nature America, Inc. All rights reserved.

evaluated different approaches of using microarray data to create and validate predictive models19—so-called genomic classifiers— and is providing techniques and guidelines to avoid over-fitting data. The current project, says Shi, is evaluating data from next-generation sequencing. “We are seeing similar things where the performances of various platforms and data analysis approaches have not been adequately vetted before their widespread applications,” he says, but it’s “even more challenging for next-generation sequencing because the data are more complex and prodigious, and the applications are more diverse.” Modeling human metabolism Computer programs that model microbial metabolic networks have been around for nearly two decades, but human cells are trickier. The models for microbes constrain themselves to intuitive ‘metabolic objectives’: taking nutrients from the environment and using them to grow. Building such models requires detailed data about enzyme and metabolite concentrations over time, but such data are impractical to collect for largescale modeling of human tissues, whose metabolic objectives are also not so easily defined. A bacterial cell that grows as fast as possible is probably an evolutionary success; a human cell that does the same is probably cancerous. C o mp u t at i o n a l biologists Eytan Ruppin and Tomer Shlomi at Tel Aviv Tomer Shlomi says, University decided “[Modeling] lets you that, because they use something that couldn’t get the data is straightforward to they thought they measure and predict needed, they should something that is hard find a way to apply to measure on a large data they had. Starting scale.” from a model for a generic human cell developed by the Bernhard Palsson lab at the University of California, San Diego20, Ruppin and Shlomi used microarray and proteomic data to predict metabolic activity specific for ten tissue types21. “It lets you use something that is straightforward to measure and predict something that is hard to measure on a large scale,” explains Shlomi, now at the Technion Israel Institute of Technology in Haifa.

The model correctly predicts metabolic fluxes in yeast cells. In human tissues, it indicates post-transcriptional regulation for about a fifth of tissueEytan Ruppin says, specific genes. The “Building models model also accuis a matter of fine rately categorizes art, experience and tissue-specific scientific taste.” activity of dozens of known disease genes, and it is already being applied in other settings; researchers who want to look at the fluxes and post-transcriptional regulation predicted by their own data can use freely available software called iMAT (Integrative Metabolic Analysis Tool)22. Meanwhile, Shlomi and Ruppin have extended their work to make more precise models of liver and cancer metabolism. The latter model predicted certain proteins that are particularly important in a hereditary form of kidney cancer and compared the list against known targets for anticancer drugs. So far, one potential new target has been experimentally validated by collaborator Eyal Gottlieb, from the Beatson Institute for Cancer Research (Glasgow, UK), says Shlomi; back-to-back papers describing the model and the target should be published later this year. The researchers hope to make these metabolic models more sophisticated. Right now, the in silico networks do not include proteins involved in genetic regulation or cell signaling. And the model considers expression only qualitatively: genes can be expressed at high levels, low levels or not at all. Being able to include additional regulatory and signaling constraints and to use continuous, quantitative data for expression could improve the accuracy of predictions, says Shlomi. Numerous laboratories in addition to Ruppin’s and Shlomi’s have been able to capture sophisticated aspects of brain, kidney, liver and other organs. Building models, says Ruppin, “is a matter of fine art, experience and scientific taste,” combined with the drive to validate a model against relevant data. “In the kind of work that we do,” says Ruppin, “you don’t start from first principles and rigorously prove everything you do. You have to make some choices that may depend on your gut feeling of what might better reflect biological reality. Then you have the responsibility to check the robustness of the decisions.” No matter how sophisticated, however, models can only give an approxi-

nature biotechnology volume 29 number 3 march 2011

mate description of the underlying biology. “What we have,” Ruppin says, “are drafts that should be improved upon.” If those drafts point to new pathways and targets, they have already demonstrated their value. Wired sensors In 2005, researchers led by Charles Lieber at Harvard University reported that nanowires coated with antibodies could detect five cancer biomarkers in serum with femtomolar sensitivity23. Rather than requiring laborious DNA amplification or instrument-intensive fluorescent labels, the wires detected the biomarkers through changes in conductance caused when electrically charged proteins bound to the silicon wire. More than a biomarker’s presence could be detected. When functionalized with the appropriate DNA sequence, Charles Lieber says, nanowires could also “The big challenge be used to monitor is not to improve telomerase binding [technology], but to take what already has to and elongating the unique attributes and wire-bound primer. make a product that The nanowire system people can use.” could even be used to screen for telomerase inhibitors. And the nanowires were poised for extreme multiplexing. Lieber predicted at the time that a basic nanowire sensor chip could potentially contain 200 or so individually functionalized wires, each its own nano-assay. Since that paper, researchers have built other nanotech detection devices based on graphene, nanotubes and other materials. Proofof-principle experiments have been carried out for multiplexed and real-time detection for many chemical species, but they are not in mainstream use. Moving the technology from stunning proof of principle to useful tool requires the right combination of knowledge about semiconductors, surface chemistry and biology. An example within Lieber’s own laboratory shows how seemingly trivial oversights can mean big delays. After demonstrating that the nanowire sensors worked with biotin and streptavidin24, Lieber wanted a more biologically relevant example, such as detecting cancer biomarkers in serum. An engineering postdoc 225

© 2011 Nature America, Inc. All rights reserved.

f e at u r e worked on the problem for two years without making any progress, says Lieber. “Then this guy [Fernando] Patolsky came into the lab, and in three weeks everything was solved.” The commercial antibodies that Lieber’s laboratory had purchased were stabilized in bovine serum albumin, and the postdocs had unwittingly been attaching the albumin rather than the antibody when fabricating the nanowires. In fact, Lieber is hopeful that co-author Patolsky might be able to commercialize the nanosensor technology at a new startup, now that another company, Nanosys (Palo Alto, CA, USA), which owns the relevant intellectual property, has indicated some willingness to sublicense. Using the nanowire technology to make a single chip that detects a variety of biomarkers in real time is doable, but it’s not something that people used to working with conventional techniques readily consider. “People are used to their ingrained technologies,” Lieber says. “Maybe the big challenge for the field is not to improve it, but to take what already has unique attributes and make a product that people more generally can use.” Lieber is taking that idea with him into his current project, kinked nanowires capable of monitoring chemistry inside a living cell25. “We need to make biologists something that they know how to use,” he says. Though the kinked nanowires have been put onto chips, neurobiologists are more familiar with patch clamps, so Lieber is looking into putting the nanowires on something resembling these tools. At this stage, the key to making nanotechnology detection devices widespread is not always in pushing the science forward, says Lieber; it’s often in improving usability, a task that is outside the scope of an academic laboratory. “The limitation and also frustration as a scientist is the commercialization,” he says. Eying islets On his first day at work at a fledgling biotech company in San Diego, Ed Baetge found rather less than he’d expected. As a newcomer to Cythera (which later became Novocell and is now ViaCyte), his task was to turn embryonic stem cells into insulinsecreting beta cells, but he didn’t even have the starting material. About a year later, in 2002, colleague Alan Agulnick produced the first line. “It was just a tiny little colony growing up on a bed of fibroblast feeders, and we had to do needle passage, where you would dissect away the undifferenti226

ated material, all done morphologically on a dissecting scope.” Other lines followed, and Baetge and his colleagues began puzzling out how to move cells from a pluripotent state to a Ed Baetge says, “[Islet cells are] probably pancreatic one. the most difficult The first step was biological product you to make definitive could ever consider.” endoderm, one of the three germ layers of nonreproductive tissue, and the one that gives rise to pancreatic tissue. The goal was to mimic normal development, says Baetge. They began lowering the concentration of serum in the culture media and also adding the growth factor activin. The key to success was constant monitoring of gene expression in response to daily changes of media. “The culture had to be examined not on a daily basis but on an hourly basis,” Baetge recalls. The next steps involved applying various morphogens and growth factors, which finally resulted in the production of endocrine cells and the 2006 Nature Biotechnology paper26. The step that mattered, though, was the previous one27. “If we hadn’t gotten good, clean, definitive endoderm, we never would have gotten there.” Still, endocrine tissue produced was a long way from functional pancreatic cells. For a long time, says Baetge, the cells produced little insulin in response to glucose and didn’t express pancreatic markers stably. “We used to call them schmendocrine cells, a schmucky kind of endocrine cell,” recalls Baetge. Extrapolating from published research, Baetge speculated that progenitors needed some signals from mesoderm tissues to form functional pancreatic cells, but the signals didn’t seem to be present in the in vitro culture system, so he and colleagues came to the surprising idea of transplanting the cells into mice. The thinking, he recalls, was “why not take the cells and put them in vivo and see if some of the signals they were missing would be delivered?” After four weeks and no sign of insulin, one principal investigator was ready to stop the experiment, but Baetge urged them to wait. He had to urge again at eight weeks and at ten weeks, and then—finally— in the twelfth week, they found insulin being secreted in response to glucose28. The length of time is not so surprising, Baetge says, considering that it takes at least 100 days after conception for a developing embryo to develop bona fide islet cells. Now, researchers at ViaCyte are working on ways to

make larger volumes of cells and shield them from the immune system. The progress has been impressive, says Baetge, now head of the Nestlé Institute of Health Sciences in Lausanne, Switzerland, but there is still a long way to go. “It’s probably one of the most challenging biological products you could ever consider developing.” Researchers working on their own differentiation projects have to be willing to perform the same experiments repeatedly. “The most important thing is to have very good control of the culturing conditions,” Baetge says. “Don’t keep the ES [embryonic stem]cells around forever. Make a large bank of cells in vials. Use a vial to do the experiment and when it’s done, don’t keep passaging the cells. Thaw a new vial and begin again.” Taking account of proteins Five years ago in our anniversary issue, proteomics experts told us that they dreamed of the day when entire proteomes could be displayed and meaningful cross-platform and cross-laboratory comparisons could be made. The field is now closer to realizing that goal, according to Christine Vogel, who in 2007, with her then post-doctoral advisor, Edward Marcotte, and colleagues, reported a method for large-scale quantification of proteins. Previously, quantifying proteins could only be done on a relative scale, and relied on laborious, impractical and sometimes expensive labeling methods. Marcotte and his colleagues at the University of Texas, Austin, moved the field toward a labelfree approach with a technique they call APEX (absolute protein expression profiling)29. Edward Marcotte says, APEX works by “It’s clear that ability taking a count of to monitor proteomes peptides (so-called allows you to look spectral counting) directly at all sorts of things.” and then adjusting that value by the likelihood that a particular peptide will be present, or by what Vogel calls its “flyability” (called by others “frequent flyers”), which is a function of various things, such as ionization efficiency and solvent conditions. Using this method in conjunction with expression profiling, they

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

f e at u r e showed for several thousand yeast and Escherichia coli proteins that a protein’s abundance correlates most of the time with the amount of message present—70% of the time with yeast, Christine Vogel says, “APEX has helped slightly less with establish quantitative E. coli. This work protein concentration dovetailed nicely with as a novel data type.” a paper from Ruedi Aebersold’s group in the same issue, in which a computational approach was taken to identify some 16,000 proteotypic peptides (those with the greatest flyability, most likely to be detected by mass spectrometry) for over 4,000 yeast proteins30. Since then, APEX has become a workhorse to address a range of research questions in the Marcotte laboratory (and will soon in the Vogel laboratory, which is in the processing of being set up at New York University). “It was fun for us. It opened up a lot of studies,” says Marcotte. “It’s clear that ability to monitor proteomes allows you to look directly at all sorts of things, organization of proteins into complexes, the function of proteins. We still don’t know the function of a large number of proteins. It will allow us to nail down function. Think of metaproteomics, [APEX can reveal] which gene products survive in different niches.” More recent work from the Marcotte laboratory and elsewhere has shown that the link between mRNA levels and protein levels is not as simple as it appeared in yeast and E. coli. Applying APEX to a human tumor cell line, for example, Marcotte found that you can explain only 30% of the variation in protein concentrations with the knowledge of mRNA concentrations in the same cell, which means that over two-thirds is a function of post-transcriptional

regulation of some kind31. Not only that, in the hands of a group of Swiss researchers32 as well as Marcotte33, APEX has been used to show that protein abundances across species from mice to man show greater conservation than RNA levels. Marcotte likes to think about it in this way: “RNA abundances are free to diverge, while post-translational regulation brings protein levels back in line.” Since the paper came out, APEX has been accepted as a fast and easy method for getting absolute quantity. Label-free techniques open the door for experiments where labeling would be impossible or when it might change an organism’s metabolism or a protein’s properties. “Along with work by others, [the Nature Biotechnology paper] helped establish the prediction of flyability, which, although not perfect, is usable for everyday life, and has helped establish quantitative protein concentration as a novel data type,” says Vogel. APEX has been taken up by two groups of researchers—method developers (“the MS [mass spectrometry] crowd”) and biologists— according to Vogel, who puts herself in the latter group. “It points the researchers to the usefulness of absolute concentrations and our ability to now measure them at large scale. Absolute concentrations are needed to start thinking about ‘rates’ of protein production and degradation. Several papers have appeared which start thinking about translation rates, and such models of rates.” And as for the dreams of five years ago, several have come true. Researchers can now routinely identify proteins at large scale and quantify them using ICAT (isotope-coded affinity tags), SILAC (stable isotope labeling with amino acids in cell culture) or other labeling methods, as well as label-free strategies. Moreover, the Aebersold laboratory has created a proteome-wide map of proteotypic peptides for yeast and (to a large extent) for human34.

nature biotechnology volume 29 number 3 march 2011

1. Takahashi, K. & Yamanaka, S. Cell 126, 663–676 (2006). 2. Nakagawa, M. et al. Nat. Biotechnol. 26, 101–106 (2008). 3. Miura, K. et al. Nat. Biotechnol. 27, 743–745 (2009). 4. Nakagawa, M. et al. Proc. Natl. Acad. Sci. USA 107, 14152–14157 (2010). 5. Shojaei, F. et al. Nat. Biotechnol. 25, 911–920 (2007). 6. Shojaei, F. et al. Nature 450, 825–831 (2007). 7. Shojaei, F. et al. Proc. Natl. Acad. Sci. USA 106, 6742– 6747 (2009). 8. Ball, M.P. et al. Nat. Biotechnol. 27, 361–368 (2009). 9. Deng, J. et al. Nat. Biotechnol. 27, 353–360 (2009). 10. Choi, H.S. et al. Nat. Biotechnol. 25, 1165–1170 (2007). 11. Choi, H.S. & Frangioni, J.V. Mol. Imaging 9, 291–310 (2010). 12. Qian, X. et al. Nat. Biotechnol. 26, 83–90 (2008). 13. Mohs, A.M. et al. Anal. Chem. 82, 9058–9065 (2010). 14. Fabian, M.A. et al. Nat. Biotechnol. 23, 329–336 (2005). 15. Karaman, M.W. et al. Nat. Biotechnol. 26, 127–132 (2008). 16. Manning, G. et al. Science 298, 1912–1934 (2002). 17. Goldstein, D.M., Gray, N.S. & Zarrinkar, P.P. Nat. Rev. Drug Discov. 7, 391–397 (2008). 18. Shi, L. et al. Nat. Biotechnol. 24, 1151–1161 (2006). 19. Shi, L. et al. Nat. Biotechnol. 28, 827–838 (2010). 20. Duarte, N.C. et al. Proc. Natl. Acad. Sci. USA 104, 1777–1782 (2007). 21. Shlomi, T., Cabilil, M.N., Herrgård, M.J., Palsson, B.Ø. & Ruppin, E. Nat. Biotechnol. 26, 1003–1010 (2008). 22. Zur, H., Ruppin, E. & Shlomi, T. Bioinformatics 26, 3140–3142 (2010). 23. Zheng, G., Patolsky, F. & Cui, Y. Wang, W. U., & Lieber, C. M. Nat. Biotechnol. 23, 1294–1301 (2005). 24. Cui, Y., Wei, Q., Park, H. & Lieber, C.M. Science 293, 1289–1292 (2001). 25. Tian, B. et al. Science 329, 830–834 (2010). 26. D’Amour, K.A. et al. Nat. Biotechnol. 24, 1392–1401 (2006). 27. D’Amour, K.A. et al. Nat. Biotechnol. 23, 1534–1541 (2005). 28. Kroon, E. et al. Nat. Biotechnol. 26, 443–452 (2008). 29. Lu, P., Vogel, C., Wang, R., Yao, X. & Marcotte, E. Nat. Biotechnol. 25, 117–124 (2007). 30. Mallick, P. et al. Nat. Biotechnol. 25, 125–131 (2007). 31. Vogel, C. et al. Mol. Syst. Biol. 6, 400 (2010). 32. Schrimpf, S.P. et al. PLoS Biol. 7, e48 (2009). 33. Laurent, J.M. et al. Proteomics 10, 4209–4212 (2010). 34. Picotti, P. et al. Cell 138, 795–806 (2009).

227

p at e n t s

Unsettled expectations: how recent patent decisions affect biotech Brenda M Simon & Christopher T Scott

© 2011 Nature America, Inc. All rights reserved.

A look back shows that broad patents are a thing of the past and biotech inventors face heightened requirements for patentability.

T

here is perhaps no industry more dependent on the value of intellectual property than biotech. Yet obtaining and enforcing patents has become more difficult in this technological space over the last decade. Not only must biotech inventions be new, useful and nonobvious, but they also must meet heightened requirements for patentability. Here we review several recent patent decisions that can profoundly affect the biotech industry. In these cases, the courts ruled on essential questions to patentability, including whether inventions are even eligible for protection. We discuss how changes in the law can affect decisions of whether and when to seek patent protection, the expectations of investors, and most importantly, the validity and enforceability of entire classes of patents. Bilski: a challenge for diagnostics Under US patent law, the first step for patent protection is showing eligibility. To determine whether the subject matter of an invention is eligible, the courts must ask whether the claims described in the patent preempt a fundamental principle, such as an abstract idea, natural phenomenon or law of nature (not eligible), or whether they apply a law or formula to a known structure or process (likely eligible). The ‘machine-or-transformation’ test is one way to determine eligibility. Under this test, the invention must be tied to a specific machine or transform an article. The machine or transformation must impose ‘meaningful limits’ on the Brenda M. Simon is at the Thomas Jefferson School of Law, San Diego, California, USA, and is a non-resident fellow at the Stanford University Center for Law and the Biosciences, Stanford, California, USA; Christopher T. Scott is at the Stanford University Center for Biomedical Ethics, Stanford, California, USA. e-mail: [email protected] or [email protected]

claims—not just adding a field-of-use restriction, or requiring insignificant extra-solution activity, such as data gathering. The United States Supreme Court recently addressed the issue of eligibility in Bilski v. Kappos1 in 2010. The question was whether a method of commodities hedging constitutes patentable subject matter—in the words of the Patent Act, is the method a “new and useful process, machine, manufacture, or composition of matter”2? The worry was that if method patents always had to be subjected to the machineor-transformation test, it could significantly reshape the biotech landscape. Many patents that undergird corporate portfolios are based on method claims, such as advanced diagnostic medicine techniques. But the court held that the machine-or-transformation test is not the sole test for patent eligibility, but instead is an “important clue.” In light of Bilski, are diagnostic methods patentable? The US Court of Appeals for the Federal Circuit tackled this question in Prometheus v. Mayo3 in 2010. The claims discussed a method of measuring a patient’s metabolism of a drug, and in response adjusting a drug dosage to treat gastrointestinal disorders. Because the claims covered the application of a natural phenomenon in a particular method of treatment, they did not preempt a fundamental principle. When a claim is directed to a method of treatment involving administration of a particular class of drugs, the court found it is transformative. Measuring drug metabolism in clinical tests was similarly transformative, as it requires extraction from the substances to be measured. This satisfied the machine-or-transformation test, and diagnostic claims tied to methods of treatment will be considered patent eligible. The Federal Circuit will shortly reconsider a similar case, Classen v. Biogen4, which concerns a method of optimizing vaccination schedules. The question

nature biotechnology volume 29 number 3 march 2011

is whether the Federal Circuit will develop some other tool besides the machine-or-transformation test to assess preemption in these cases. The Federal Circuit is also reviewing BRCA sequence patents in the Myriad5 litigation. Several of the method claims compare a given DNA sequence with normal and mutated BRCA sequences to determine susceptibility to cancer. Another claim describes a method of drug screening using recombinant cells that express a BRCA protein. Isolated DNA sequences and cDNA molecules comprise the compositionof-matter claims. Whether Myriad’s patents are eligible will likely depend on how broadly they are interpreted and whether the court applies the machine-or-transformation test. If practicing the method requires manipulation of the measured substances—as opposed to comparisons that can be performed mentally—the method would seem to satisfy the machine-or-transformation test articulated by the Federal Circuit. The analysis of the composition claims will be more difficult. In general, products of nature are ineligible for patent protection. However, a product of nature that is altered, isolated or purified by an inventor has long been considered eligible for patent at the US Patent & Trademark Office (USPTO). Now, the determination of eligibility may depend on how broadly the term ‘isolated’ is defined. The Myriad decision has the potential to affect not only patents on isolated and purified genes, which are becoming less relevant as the discovery of single gene disorders slows down, but also those related to antibodies, proteins, cell lines and stem cells. The case for usefulness Even if an invention is eligible for a patent, the inventor must still show it is useful. Biotech inventions must show a higher degree of utility than many other types of inventions. Claims must show both specific and substantial utility. 229

© 2011 Nature America, Inc. All rights reserved.

pat e nts For specific utility, the inventor must indicate that the invention provides a particular public benefit. For substantial utility, the inventor must show the invention has “a significant and presently available benefit to the public”6. In 2005, the Federal Circuit addressed the utility required for DNA fragments known as “expressed sequence tags” (ESTs) in In re Fisher6. An EST corresponds to a portion of a gene being expressed, which can be useful in identifying an unknown gene and its location. But, an EST does not explain the purpose and use of the gene. In Fisher, the court rejected an attempt to patent ESTs where the function of the genes represented by the ESTs was not yet identified. Awarding a patent in this case “would amount to a hunting license” for performing research that might not result in anything useful. In this way, the utility standard precludes patenting sequence information for genes of unknown function. This results in a heightened standard for biotech patents, which can delay filing until the required utility can be shown. Are sequences and me-too drugs novel? An invention must also be novel. Novelty is an increasingly difficult requirement for patents that claim genetic sequences and metabolites of previously patented drugs. Now that sequenced human genomes are accessible, the novelty of claims to genetic sequences is being called into question. In 2009, the Federal Circuit held in In re Gleave7 that a laundry list of nucleotide sequences renders claims to particular oligonucleotides not novel, even though there was no known use for the listed sequences. Although methods of use for such sequences are likely novel, new composition patents on them are not, and may soon become obsolete. Until recently, drug manufacturers could obtain composition patents on metabolites formed after ingestion of their previously patented medications. This strategy would hinder market entry by generic manufacturers, even if the key patent on the drug had expired. But in 2003, the court in Schering v. Geneva8 invalidated a patent covering a metabolite of loratadine, the active ingredient in Claritin. It ruled that the parent patent on loratadine undercut the novelty of the later metabolite patent. Even though the patent holder did not recognize that the metabolite was formed until after filing the primary patent, the court held that the inventors did not have to know that the metabolite existed for it to invalidate the later patent. In view of this decision, patents claiming metabolites of previously patented medicines are called into question. Beware the obvious In addition to the utility, eligibility and novelty requirements, an invention must not be obvious. 230

To determine nonobviousness, the court asks whether one of ordinary skill in the art would have found the invention obvious in view of what is published in the literature, or is otherwise prior art. In 2007, the US Supreme Court provided for more flexibility in determining obviousness in KSR v. Teleflex9. The court held that the prior art does not have to provide some teaching, suggestion or motivation to make the claimed invention. Instead, courts can take into account predictability and what a person of ordinary skill, exercising creativity, would find obvious to try. The Federal Circuit applied this reasoning to In re Kubin10 in 2009, which significantly changed the obviousness rule for biotech. This classic biotech invention at issue required isolating and sequencing a gene that encoded a known protein. Both the sequence and the protein were obtainable by known methods. The prior art both identified the protein and suggested its function. The court ruled the gene was obvious, as it was reasonably expected in view of the prior art, and obvious to sequence in view of the limited number of predictable solutions. Many gene patents still may be nonobvious despite Kubin, as the prior art in that case identified the protein, as well as suggested its purpose and use. Obviousness was at the center of the controversial inter partes reexamination of three foundational embryonic stem cell patents held by the Wisconsin Alumni Research Foundation (WARF) and inventor James Thomson. The USPTO examiner first invalidated the claims for anticipation and obviousness, but later withdrew the rejections. The Foundation for Taxpayer and Consumer Rights appealed. In its 2010 decision11, the Board of Patent Appeals and Interferences invoked rulings in KSR and Kubin and cited 12 instances of prior art in its rejection of the broadest of the three patents, which claimed rights to the lines themselves. Even if Bilski limits the patentability of genes, a higher bar for overcoming obviousness could have a greater impact on biotech patents. Show possession To obtain a patent, applicants must adequately describe and enable their inventions. Enablement requires the applicant to teach one of ordinary skill in the art how to make and use the invention without undue experimentation. The written description requirement generally calls for the applicant to demonstrate possession of the invention at the time of filing. Compared to other fields, the courts ask for a rather stringent written description for biotech inventions. For genetic inventions, an applicant generally needs to provide the sequence of the claimed gene to satisfy the written description

requirement. Merely discussing how to obtain the sequence will not do the job. The Federal Circuit recently clarified the scope of the written description requirement in Ariad v. Lilly12, with all of the active judges rehearing the case in 2010. In Ariad, the court held that written description is a separate and distinct requirement from enablement. Ariad’s claims recited methods of reducing the activity of a transcription regulator. For support, Ariad had discussed three classes of molecules that could potentially carry out this inhibition. The court invalidated the claims, holding Ariad’s hypothetical description of the molecules was inadequate in view of the nascent state of the technology and the breadth of the claims. In effect, the written description requirement hinders claims to a genus, as it has become difficult to describe a sufficient number of species for support. Inventors need to show they possess more than a couple of members of a large group and may need to show additional knowledge or functional relationship between them. This makes obtaining patents with genus claims more difficult, and easier to invalidate once they have been granted. A high threshold for written descriptions results in patents with narrower claims that can more easily be designed around. Broad patents: a thing of the past? The past decade has been a challenging one for biotech. Recent patent cases have made it difficult to obtain the broader protection to which the industry had become accustomed, while making challenges more likely. Considerable uncertainty exists in the areas of eligibility, obviousness and disclosure. How these emerging tests are applied has the potential to profoundly affect the biotech industry. In recognition of emerging developments in areas like personalized medicine, stem cells and synthetic biology, the courts seem reluctant to award broad patents that may tie up downstream innovation. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 1. Bilski v. Kappos, 561 US ___ (2010). 2. 23 USC §101. 3. Prometheus Laboratories, Inc. v. Mayo Collaborative Services, No. 2008-1403 (Fed. Cir. Dec. 17, 2010). 4. Classen Immunotherapies, Inc. v. Biogen IDEC, Nos. 2006-1634 and 2006-1649 (Fed. Cir. 2008). 5. Association for Molecular Pathology v. USPTO and Myriad Genetics, Inc., No. 2010-1406 (Fed. Cir. 2010). 6. In re Fisher, 421 F.3d 1365 (Fed. Cir. 2005). 7. In re Gleave, 560 F.3d 1331 (Fed. Cir. 2009). 8. Schering Corp. v. Geneva Pharm., Inc., 339 F.3d 1373 (Fed. Cir. 2003). 9. KSR Int’l Co. v. Teleflex, Inc., 550 U.S. 398 (2007). 10. In re Kubin, 561 F.3d 1351 (Fed. Cir. 2009). 11. USPTO BPAI, Foundation of Taxpayer & Consumer Rights v. Patent of WARF, Appeal 2010-001854, Patent 7,029,913 (April 28, 2010). 12. Ariad Pharmaceuticals, Inc. v. Eli Lilly & Co., 598 F.3d 1336 (Fed. Cir. 2010) (en banc).

volume 29 number 3 march 2011 nature biotechnology

patents

© 2011 Nature America, Inc. All rights reserved.

Recent patent applications in antibody fragments Priority application date

Publication date

Dang W, Desjarlais JR, Hayes R, Karki SB, Lazar GA, Vafa O, Vielmetter J

3/3/2003

1/27/2011

–

5/22/2009

1/13/2011

6/23/2009

1/13/2011

Adekar S, Dessain S, O’Nuallain B

7/1/2009

1/6/2011

Ni J, Rosen CA, Yu G, Zhang J

11/7/1994

1/6/2011

Brewer LA, Duan DR, Ebner R, Endress GA, Florence KA, Komatsoulis GA, Lafleur DW, Moore PA, Mucenski M, Olsen HS, Rosen CA, Ruben SM, Shi Y, Soppet DR, Young PE

7/15/1998

1/6/2011

Katrukha AG, Kharitonov AV, Postnikov AB, Solovyeva TI

6/29/2009

1/6/2011

O’Connor TP

2/22/2005

12/30/2010

DRG International Geacintov CE, (Mountainside, NJ, USA) Janetzko A, Kulaksiz H, Stremmel W

11/19/2002

12/30/2010

Akamatsu Y, DuBridge RB, Akamatsu Y, Harding FA, Powers DB DuBridge RB, Facet Biotech (Redwood City, CA, USA), Harding FA, Powers DB

6/17/2009

12/23/2010

Patent number

Description

Assignee

Inventor

US 20110021755

An antibody or immunoadhesin comprising an Fc variant of parent fragment crystallizable (Fc) polypeptide; useful in pharmaceutical compositions for preventing and/or treating autoimmune or inflammatory disease.

Xencor (Monrovia, CA, USA)

JP 2011007782

Kyowa Medex A method of measuring soluble interleukin-2 receptor (sIL-2R) in a sample, comprising react- (Tokyo) ing a first antibody or its fragment that binds sIL-2R with a magnetic particle in an aqueous medium; reacting a second antibody coupled with sIL-2R or its fragment with the complex in an aqueous medium; isolating a magnetic particle in the reaction liquid; and measuring the second antibody on the magnetic particle.

WO 2011005481

An antibody or its antigen-binding fragment comprising the Fc region and method of its production, and a method of detecting or treating cancer, autoimmune, inflammatory or infectious diseases or disorders, comprising administering the antibody or its antigen binding fragment.

MedImmune DiMasi N, Gao C (Gaithersburg, MD, USA)

WO 2011002494, US 20110002945

A method of binding aggregated amyloidogenic proteins such as amyloid fibril or oligomer, involving exposing aggregated amyloidogenic proteins to an antibody chain, comprising heavy chain, light chain, or a portion of heavy or light chain of immunoglobulin.

Thomas Jefferson University (Philadelphia), Adekar S, Dessain S, O’Nuallain B, University of Tennessee Research Foundation (Knoxville, TN, USA)

US 20110003399

An isolated antibody or its fragment capable of binding specifically to protein or TNF-g protein; useful in a pharmaceutical composition for preventing and/or treating, e.g., cancer, cachexia, septic shock and malaria.

Human Genome Sciences (Rockville, MD, USA)

US 20110003383

Human Genome An isolated antibody or its fragment capable of Sciences specifically binding to isolated HT5GJ57 poly(Rockville, MD, USA) peptide; used in a pharmaceutical composition for diagnosing and treating disorders, e.g., hyperproliferative disorders.

WO 2011001029

An antibody that recognizes novel epitopes originated by enzyme-dependent cleavage of insulin-like growth factor binding protein-4, comprising specific amino acid sequence or an amino acid sequence having at least 80% homology to specific amino acid sequence; useful for diagnosing, e.g., cancer.

US 20100330661

A device comprising a solid support that is useful IDEXX Laboratories for the detection and quantification of Ehrlichia (Westbrook, ME, USA) ewingii, Ehrlichia ewingii antibodies, antibody fragments and polypeptides.

US 20100330595

A method of diagnosing hereditary hemochromatosis or chronic renal insufficiency, involving obtaining a tissue or fluid sample from the subject, contacting the sample with an antibody or its fragment and quantifying hepcidin level in the sample.

US 20100322931, WO 2010148223

An anti-vascular endothelial growth factor (VEGF) antibody or an anti-VEGF binding fragment of antibody, comprising complementarity determining regions having amino acid sequences corresponding to fully defined amino acids given in the specification; useful for treating cancer, age-related macular degeneration and immune disorders.

HyTest (Turku, Finland)

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).

nature biotechnology volume 29 number 3 march 2011

231

news and views

IPSCs put to the test Hyesoo Kim & Lorenz Studer

© 2011 Nature America, Inc. All rights reserved.

Analysis of a test set of cell lines shows that induced pluripotent stem cells perform as well as embryonic stem cells in differentiating to motor neurons. The ability to reprogram adult skin fibroblasts into induced pluripotent stem cells (iPSCs) represents one of the most remarkable recent feats in the biological sciences. A particularly exciting development is the use of patientspecific iPSCs to capture aspects of human disease in a petri dish, as illustrated recently for several neurodevelopmental disorders1–3. However, for this approach to work, it is essential that disease phenotypes not be masked by technical variability associated with the derivation or differentiation of iPSCs. In this issue, Boulting et al.4 address this question head on by systematically comparing differentiation behavior across a well-designed set of human iPSC and embryonic stem cell (ESC) lines. Overall, the paper provides welcome news to the community, suggesting that the mean differentiation performance of human iPSC and ESC lines is indistinguishable and that poorly performing iPSC lines can be ‘rescued’ by simple changes in culture conditions. The advent of iPSC technology raises tantalizing questions about the molecular control and reversibility of differentiated cell fates. Of particular importance is the question of whether iPSCs represent a fully reprogrammed state that is functionally equivalent to ESCs. Recent studies in the mouse suggest that iPSCs retain a partial epigenetic memory characteristic of the somatic cell of origin, and that such memory can influence differentiation behavior5,6. Previous work on neural differentiation suggested that the performance of human iPSC lines may be more variable and on average inferior to the performance of wellcharacterized human ESC lines such as H9 (ref. 7). However, these results have been difficult to interpret as H9 may not be an average Hyesoo Kim and Lorenz Studer are at the Center for Stem Cell Biology and the Developmental Biology Program, Sloan-Kettering Institute, New York, New York, USA. e-mail: [email protected]

ESC line but rather one that is particularly efficient at neural differentiation. In fact, H9 has been used to optimize many neural differentiation protocols. Human ESCs are known to differ substantially in their propensity to differentiate to particular lineages8. For iPSCs, it is possible that the specific reprogramming technology used could affect differentiation results. For example, a recent study found that the retention or excision of a reprogramming cassette in human iPSCs led to reproducible changes in gene expression9. The paper by Boulting et al.4 represents the largest and most systematic study of these issues to date. The authors compare the efficiency with which 16 human iPSC lines and 6 ESC lines generate spinal motor neurons, a well-defined differentiated cell type (Fig. 1). The iPSC lines are selected to span differences in donor age, sex, genotype and health status (amyotrophic lateral sclerosis (ALS) patients versus healthy controls) as well as two reprogramming methods (with and without the c-MYC transgene). The authors also study the effects on differentiation of karyotypic instability and persistent transgene expression. Finally, all differentiation assays are performed in parallel in two different laboratories to control for variability in cell handling. The data demonstrate a remarkable reproducibility in motor neuron yield for any given line, suggesting that differentiation behavior is a stable trait for each human pluripotent stem cell line. Notably, a comparison of the motor neuron yield between human ESC and human iPSC lines shows no significant differences. And of all the possible sources of variability examined, none correlate with differentiation efficiency except possibly donor identity and donor sex. However, as the authors observe only one case for which donor identity significantly affects motor neuron yield, further studies will be required to address the important question of variability between lines from different individuals. Similarly, differences related

nature biotechnology volume 29 number 3 march 2011

to donor sex (higher motor neuron yield from female lines) will require independent confirmation, although it is interesting to speculate about potential Y-chromosome-linked or hormone-related factors that could affect differentiation. Three out of the 16 iPSC clones tested do not yield motor neurons under standard differentiation conditions despite expressing a normal pluripotency marker profile. However, these poorly performing clones can be rescued simply by using a stronger neural induction stimulus (a modified version of the dual SMAD–inhibition protocol developed in our laboratory10). What does this study teach us for future disease-modeling or drug-discovery efforts? The finding that iPSC lines do not perform statistically worse than a set of well-characterized ESC lines implies that any potential epigenetic memory related to the fibroblast origin of iPSCs does not negatively affect neural differentiation propensity. Furthermore, these data suggest that the higher variability in differentiation outcomes among iPSCs reported previously7 may be due to differences in protocols and cell-line choice rather than reflecting a fundamental limitation of iPSC technology. For example, the current study is limited to iPSC lines established using retroviral vector technology, whereas other studies included lines generated using lentiviral or episomal vector systems6. These issues stress the need for a reliable set of pluripotent stem cell lines that allows for meaningful crosslaboratory comparisons. The high reproducibility of motor neuron yield for any given cell line shown by Boulting et al.4 speaks well for using this test set of pluripotent stem cell lines as a robust research tool across the research community. The finding that neither karyotype status nor residual transgene expression affects motor neuron yield is surprising—hence, the rather bizarre result of ISL1+ putative motor neurons expressing OCT4 233

ne w s and v ie w s Boulting et al.

Neural differentiation (SHH/RA)

Motor neuron yield

PASS 13 lines 16 iPSC lines

FAIL 3 lines

RESCUED 3 lines Dual SMAD inhibition

PASS 6 lines

No correlation ESC vs. iPSC Donor age Transgene silencing Karyotype change Variable ALS status +/- c-MYC Different lab Possible correlation Donor identity Donor sex

6 ESC lines

© 2011 Nature America, Inc. All rights reserved.

Bock et al.

Predict lines that fail

Predict motor neuron yield

Deviation scorecard

Lineage scorecard

Methylation and transcriptional profiling at pluripotent stage

Transcriptional profiling at embryoid body stage

Figure 1 Boulting et al.4 establish a test set of 16 human iPSC lines from seven genotypes and 6 human ESC lines and investigate factors that may affect the efficiency with which the lines differentiate to motor neurons. Motor neuron yield from iPSC and ESC lines is similar and is not affected by donor age, transgene silencing, karyotypic changes, ALS status, presence of c-MYC or handling of cells between two different laboratories. In contrast, the study shows preliminary evidence that donor sex and donor identity can affect motor neuron yield. Three lines recalcitrant to differentiation could be rescued by simple adaptation of culture conditions using a modified dual-SMAD inhibition protocol. A related paper by Bock et al.12 makes use of the test set of lines from Boulting et al.4 to validate a molecular profiling approach that predicts the differentiation behavior of pluripotent stem cells. A deviation scorecard, based on global DNA methylation and transcriptome profiling, can spot lines with highly aberrant differentiation. High-throughput quantitative gene expression profiling in cells differentiated to embryoid bodies is used to establish a lineage scorecard, which reliably predicts motor neuron yields described by Boulting et al.4.

protein in at least one line. One might argue that the study lacks sufficient statistical power to rule out small effects of transgene silencing or karyotypic integrity on motor neuron yield, but clearly there were no major differences. On the other hand, motor neuron derivation is not the end point but rather the beginning of a disease-modeling or drug-discovery project. For such applications, it would be prudent to avoid the use of cell lines that retain transgene expression or exhibit karyotypic abnormalities as these issues may affect cell function even if they do not affect motor neuron numbers. It is notable that overall motor neuron yield was rather low in this study compared with previous studies in mouse and human ESCs. Mouse ESCs generally yield percentages of HB9+ motor neurons that are ~10 times higher than those reported here for human iPSC or ESC lines11. Previous work in human pluripotent stem cells also reported higher numbers of HB9+ motor neurons7, although the precise numbers of HB9+ cells can be difficult to compare across studies in the absence of a reliable 234

reporter system. Therefore, it will be important to develop improved differentiation protocols and to reassess whether cell line–related differences in motor neuron yield are further diminished or exacerbated under such optimized conditions. The test set of human pluripotent stem cell lines validated here for motor neuron differentiation offers a valuable resource for the field to benchmark other differentiation protocols across laboratories. A strong argument for using this resource is a related paper published concomitantly in Cell by Bock et al.12, which makes use of the test set to validate two independent algorithms that predict differentiation behavior. Based on a collection of 20 human ESC lines, Bock et al.12 first established a reference signature for global transcriptome and DNA methylome analysis. Interestingly, most of the human iPSC lines (all from the test set of Boulting et al.4) fell within the range of normal human ESC variability for those parameters, although the authors were able to define a bioinformatics algorithm that was fairly reliable

at distinguishing iPSC and ESC identity. With those data in hand, the study went on to develop a ‘deviation scorecard’ that identifies unusual gene expression or DNA methylation patterns in human pluripotent stem cell lines. Using the deviation scorecard in the human pluripotent stem cell probe set, the authors could reliably predict poorly behaving lines prone to aberrant differentiation. In a next step, Bock et al.12 tried to develop an algorithm that can predict lineage differentiation propensity. To this end, 18 of the 20 reference human ESC lines were differentiated into embryoid bodies, followed by highthroughput expression profiling for a set of 500 well-selected, lineage-specific transcripts. Based on these data, the authors devised a ‘lineage scorecard’ that showed impressive performance at predicting motor neuron yield in the test set established by Boulting et al.4. How can the biotech community make best use of the results from these two exciting studies? Most importantly, the findings present a new strategy of addressing cell lines’ variability, currently a limiting factor in many applications of human pluripotent stem cells. Traditional characterization of these cells by means of morphology, marker expression and teratoma formation is crude, and the deviation scorecard of Bock et al.12 offers a superior strategy for weeding out ‘bad’ clones. The ability to predict lineage differentiation propensity could have broader implications, such as the establishment of libraries of human pluripotent stem cell lines optimized for the generation of tissue-specific progenitors, improved diseasemodeling strategies by minimizing ‘noise’ due to cell line variability or the identification of novel molecular regulators that define lineagespecific behavior. However, it should be noted that performing the scorecard assays with embryoid bodies for a large number of cell lines would be rather laborious. Also, it will be important to determine whether the lineage scorecard performs as well for predicting the results of directed differentiation to cell types other than motor neurons. Furthermore, it may be interesting to incorporate additional parameters into the scorecards, such as microRNA expression or global histone modification. The ultimate dream may be the identification of markers that reliably predict lineage propensity already at the undifferentiated state without the need for embryoid body predifferentiation studies. However, it is obviously not known whether such markers exist. Finally, the paper by Boulting et al.4 makes an important statement about the fact that at least some of the cell line–specific behaviors can be overcome by adjusting the differentiation protocols. The rescue data of the three recalcitrant lines using

volume 29 number 3 march 2011 nature biotechnology

ne w s and v ie w s a stronger neural induction stimulus indicate that similar strategies may exist to correct other types of cell line–specific phenomena. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 1. Ebert, A.D. et al. Nature 457, 277–280 (2009). 2. Lee, G. et al. Nature 461, 402–406 (2009). 3. Marchetto, M.C. et al. Cell 143, 527–539 (2010).

Boulting, G.L. et al. Nat. Biotechnol. 29, 279–286 (2011). Polo, J.M. et al. Nat. Biotechnol. 28, 848–855 (2010). Kim, K. et al. Nature 467, 285–290 (2010). Hu, B.Y. et al. Proc. Natl. Acad. Sci. USA 107, 4335–4340 (2010). 8. Osafune, K. et al. Nat. Biotechnol. 26, 313–315 (2008). 9. Soldner, F. et al. Cell 136, 964–977 (2009). 10. Chambers, S.M. et al. Nat. Biotechnol. 27, 275–280 (2009). 11. Wichterle, H., Lieberam, I., Porter, J.A. & Jessell, T.M. Cell 110, 385–397 (2002). 12. Bock, C. et al. Cell 144, 439–452 (2011). 4. 5. 6. 7.

Chemoproteomics quantifies complexity © 2011 Nature America, Inc. All rights reserved.

Edward B Holson & Stuart L Schreiber

i nhibitors to increase the therapeutic window and to expand into new therapeutic indications. These studies usually rely on enzyme inhibition assays using purified recombinant enzyme preparations, synthetic substrates and fluorescent visualization techniques. Inside cells, however, HDACs appear to usually reside within multiprotein complexes and to function not only as deacetylating enzymes but also as scaffolding proteins in transcriptional complexes. The organizational complexity introduced by this dual functionality is represented in Figure 1 for HDAC1 and 2 and their known complexes7. Bantscheff et al.1 address this problem by adapting an approach they used previously with kinase inhibitors8 to interrogate smallmolecule binding events in intact megadalton

The affinities of small molecules for proteins in megadalton complexes can be measured by mass spectrometry. Biomedical researchers often characterize the affinities and selectivities of small molecules for target proteins using purified proteins or protein domains, even if the target proteins exist in large complexes in vivo. Such assays may not reflect the interactions of the small molecules inside cells. In this issue, Bantscheff et al.1 use mass spectrometry to measure the binding affinities of small molecules to native histone deacetylase (HDAC) megadalton complexes in cell lysates for the first time. They measure IC50s of small molecules to multiple native HDAC protein complexes, analyzing a diverse range of HDAC inhibitors, cell states, cell types and protein complexes and producing small-molecule binding profiles that are likely to be more physiologically relevant than the results of conventional assays. Moreover, the authors discover new targets of HDAC inhibitors and new protein components of native HDAC complexes. The limitations of using interactions with recombinant proteins or protein domains as a surrogate for interactions with proteins within large complexes in vivo are self-evident. For instance, rapamycin (sirolimus), the first smallmolecule protein kinase inhibitor approved for clinical use, binds both TOR1 and TOR2 in isolation. However, in yeast cells it binds only TOR1 (ref. 2), and in mammalian cells it binds mTOR only within the MTORC1, but not the mTORC2, complex3. This subtlety of small molecule–protein affinities contributes Edward B. Holson and Stuart L. Schreiber are at the Broad Institute of Harvard and MIT, Cambridge, Massachusetts, USA. e-mail: [email protected] or [email protected]

to the frequent disconnect between manipulations involving nucleic acids (e.g., using gene deletions, RNA interference) and those involving the proteins encoded by the nucleic acids (e.g., using small molecules). It can lead to unwanted surprises during drug development when ‘target validation’ relies only on data generated by modulating nucleic acids and ‘target specificity’ inferences rely only on test-tube measurements. HDACs are an important class of epigenetic drug target4. The 18 HDAC family members regulate gene expression by modifying the acetylation state of histones and have been implicated in a range of human diseases, from cancer to neurodegenerative disorders5,6. Two small-molecule HDAC inhibitors from two chemical classes have received US Food and Drug Administration approval for the treatment of cutaneous T-cell lymphoma: vorinostat (Zolinza), a hydroxamic acid, and romidepsin (Istodax), a disulfide ‘prodrug’. Current drug discovery and development efforts aim to identify novel, more potent and, in some cases, more selective HDAC Figure 1 The affinities of small molecules for HDAC1 and HDAC2 can be studied at multiple levels of complexity. These range from the least complex assays, involving purified or truncated HDACs (top), to the more physiologically relevant situation where transcriptional complexes associated with chromatinized DNA are studied (bottom). Bantscheff et al.1 show that different complexes with a catalytic core comprising HDAC1 and HDAC2 can be differentially inhibited by a given HDAC inhibitor. The representations of the complexes are based on ref. 7. The placement of the constituent subunits is not intended to reflect direct physical interactions.

nature biotechnology volume 29 number 3 march 2011

Ligand-binding domain

Small molecule

HDAC1

Truncated or full-length recombinant enzyme

HDAC1

HDAC2

HDAC1/HDAC2 heterodimer RbAp46 RbAp48 MBD P66 MTA Mi2

NuRD SAP30 SAP18 SDS

MeCP2

CoREST REST

Sin3

Sin3

CoREST

HDAC1-HDAC2-containing multiprotein complexes

LSD1

G9a DNMT1 SUVAR39H1 HP1

HDAC1-HDAC2-containing transcriptional complexes

235

ne w s and v ie w s a stronger neural induction stimulus indicate that similar strategies may exist to correct other types of cell line–specific phenomena. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 1. Ebert, A.D. et al. Nature 457, 277–280 (2009). 2. Lee, G. et al. Nature 461, 402–406 (2009). 3. Marchetto, M.C. et al. Cell 143, 527–539 (2010).

Boulting, G.L. et al. Nat. Biotechnol. 29, 279–286 (2011). Polo, J.M. et al. Nat. Biotechnol. 28, 848–855 (2010). Kim, K. et al. Nature 467, 285–290 (2010). Hu, B.Y. et al. Proc. Natl. Acad. Sci. USA 107, 4335–4340 (2010). 8. Osafune, K. et al. Nat. Biotechnol. 26, 313–315 (2008). 9. Soldner, F. et al. Cell 136, 964–977 (2009). 10. Chambers, S.M. et al. Nat. Biotechnol. 27, 275–280 (2009). 11. Wichterle, H., Lieberam, I., Porter, J.A. & Jessell, T.M. Cell 110, 385–397 (2002). 12. Bock, C. et al. Cell 144, 439–452 (2011). 4. 5. 6. 7.

Chemoproteomics quantifies complexity © 2011 Nature America, Inc. All rights reserved.

Edward B Holson & Stuart L Schreiber

i nhibitors to increase the therapeutic window and to expand into new therapeutic indications. These studies usually rely on enzyme inhibition assays using purified recombinant enzyme preparations, synthetic substrates and fluorescent visualization techniques. Inside cells, however, HDACs appear to usually reside within multiprotein complexes and to function not only as deacetylating enzymes but also as scaffolding proteins in transcriptional complexes. The organizational complexity introduced by this dual functionality is represented in Figure 1 for HDAC1 and 2 and their known complexes7. Bantscheff et al.1 address this problem by adapting an approach they used previously with kinase inhibitors8 to interrogate smallmolecule binding events in intact megadalton

The affinities of small molecules for proteins in megadalton complexes can be measured by mass spectrometry. Biomedical researchers often characterize the affinities and selectivities of small molecules for target proteins using purified proteins or protein domains, even if the target proteins exist in large complexes in vivo. Such assays may not reflect the interactions of the small molecules inside cells. In this issue, Bantscheff et al.1 use mass spectrometry to measure the binding affinities of small molecules to native histone deacetylase (HDAC) megadalton complexes in cell lysates for the first time. They measure IC50s of small molecules to multiple native HDAC protein complexes, analyzing a diverse range of HDAC inhibitors, cell states, cell types and protein complexes and producing small-molecule binding profiles that are likely to be more physiologically relevant than the results of conventional assays. Moreover, the authors discover new targets of HDAC inhibitors and new protein components of native HDAC complexes. The limitations of using interactions with recombinant proteins or protein domains as a surrogate for interactions with proteins within large complexes in vivo are self-evident. For instance, rapamycin (sirolimus), the first smallmolecule protein kinase inhibitor approved for clinical use, binds both TOR1 and TOR2 in isolation. However, in yeast cells it binds only TOR1 (ref. 2), and in mammalian cells it binds mTOR only within the MTORC1, but not the mTORC2, complex3. This subtlety of small molecule–protein affinities contributes Edward B. Holson and Stuart L. Schreiber are at the Broad Institute of Harvard and MIT, Cambridge, Massachusetts, USA. e-mail: [email protected] or [email protected]

to the frequent disconnect between manipulations involving nucleic acids (e.g., using gene deletions, RNA interference) and those involving the proteins encoded by the nucleic acids (e.g., using small molecules). It can lead to unwanted surprises during drug development when ‘target validation’ relies only on data generated by modulating nucleic acids and ‘target specificity’ inferences rely only on test-tube measurements. HDACs are an important class of epigenetic drug target4. The 18 HDAC family members regulate gene expression by modifying the acetylation state of histones and have been implicated in a range of human diseases, from cancer to neurodegenerative disorders5,6. Two small-molecule HDAC inhibitors from two chemical classes have received US Food and Drug Administration approval for the treatment of cutaneous T-cell lymphoma: vorinostat (Zolinza), a hydroxamic acid, and romidepsin (Istodax), a disulfide ‘prodrug’. Current drug discovery and development efforts aim to identify novel, more potent and, in some cases, more selective HDAC Figure 1 The affinities of small molecules for HDAC1 and HDAC2 can be studied at multiple levels of complexity. These range from the least complex assays, involving purified or truncated HDACs (top), to the more physiologically relevant situation where transcriptional complexes associated with chromatinized DNA are studied (bottom). Bantscheff et al.1 show that different complexes with a catalytic core comprising HDAC1 and HDAC2 can be differentially inhibited by a given HDAC inhibitor. The representations of the complexes are based on ref. 7. The placement of the constituent subunits is not intended to reflect direct physical interactions.

nature biotechnology volume 29 number 3 march 2011

Ligand-binding domain

Small molecule

HDAC1

Truncated or full-length recombinant enzyme

HDAC1

HDAC2

HDAC1/HDAC2 heterodimer RbAp46 RbAp48 MBD P66 MTA Mi2

NuRD SAP30 SAP18 SDS

MeCP2

CoREST REST

Sin3

Sin3

CoREST

HDAC1-HDAC2-containing multiprotein complexes

LSD1

G9a DNMT1 SUVAR39H1 HP1

HDAC1-HDAC2-containing transcriptional complexes

235

© 2011 Nature America, Inc. All rights reserved.

ne w s and v ie w s HDAC complexes. In their method, the HDAC inhibitor of interest is immobilized on sepharose beads, which are then added to cell lysates spiked with increasing concentrations of the free inhibitor. As higher concentrations of free inhibitor reduce binding of the immobilized inhibitor to its cellular targets, quantitative mass spectrometry of the bound HDAC complexes enables the determination of IC50 values. The authors study 16 HDAC inhibitors (including vorinostat and romidepsin) spanning a diverse set of ‘chemotypes’ (hydroxamic acids, ketones, carboxylic acids and amino benzamides). In total, 267 proteins from a human myelogenous leukemia cell line, including 6 HDACs and 29 proteins known to form complexes with HDACs, bind the chemical probes. The authors demonstrate the ubiquitous expression of these proteins across a panel of six human cell lines and a range of mouse tissue types, including brain, heart, kidney, liver, lung and spleen. Importantly, the authors show striking differences in potency and selectivity profiles for individual inhibitors across different protein complexes in cell lysates compared with the results of assays with recombinant proteins. The study reports several other notable findings. The aminobenzamide class of HDAC inhibitor does not bind the Sin3 multiprotein complex but shows particular selectivity for the HDAC3-NCoR complex. Valproate, considered a weak class I HDAC inhibitor, has higher affinity for the CoREST and NuRD complexes than for the Sin3 complex. The marketed orphan anti-inflammatory drug bufexamac binds class IIb (HDAC6 and 10) HDAC complexes. Accordingly, the drug induces tubulin hyperacetylation at concentrations within the range of its anti-inflammatory effects. Finally, the authors also identify a novel HDAC complex, named MiDAC, which forms during mitosis; division-arrested cells show increased deacetylation activity of MiDAC but not of other complexes. Although the method of Bantscheff et al.1 offers exciting possibilities for interrogating biological systems, the barrier to entry for most laboratories is not low. A prerequisite for its application is a probe matrix that efficiently captures the subproteome of interest. Researchers should therefore be prepared to undertake a synthetic chemistry effort combined with the determination of structureactivity relationships. In addition, thorough validation of the targets identified using complementary immunoprecipitation techniques is an important step in the process. These concerns aside, the new method shows promise as a means of probing the actual 236

interactions of small molecules with macro molecular complexes in different cellular contexts. Such studies may lead to drugs with beneficial tissue specificity. For example, a recent report9 highlights the importance of context-specific binding by showing that a small-molecule inhibitor of protein kinase C (PKC) binds its target only in vitro and not in cells due to a more stable PKC-AKAP79 cellular complex. In this cellular context, AKAP79 blocks binding of the inhibitor to its target9. It is tempting to speculate that this inhibitor may have tissue-specific effects that could not be anticipated from biochemical measurements alone. Thus, the ability to quantify binding interactions of small molecules with native protein complexes in different cellular

c ontexts as described by Bantscheff et al.3 should expand the utility of a wide range of small molecules as probes and drugs. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 1. Bantscheff, M. et. al. Nat. Biotechnol. 29, 255–265 (2011). 2. Zheng, X.-F. et al. Cell 82, 121–130 (1995). 3. Sarbassov, D.D. Curr. Biol. 14, 1296–1302 (2004). 4. Taunton, J., Hassig, C.A. & Schreiber, S.L. Science 272, 408–411 (1996). 5. Dey, P. Curr. Med. Chem. 13, 2909–2919 (2006). 6. Kazantsev, A.G. & Thompson, L.M. Nature Drug Dis. 7, 854–868 (2008). 7. Brunmeir, R. et al. Int. J. Dev. Biol. 53, 275–289 (2009). 8. Bantscheff, M. et al. Nat. Biotechnol. 25, 1035–1044 (2007). 9. Hoshi, N. et al. Mol. Cell 37, 541–550 (2010).

Biomarkers in aggregate Fred S Apple Implantable devices that measure the cumulative release of biomarkers promise new diagnostic options. For all of the excitement surrounding disease biomarkers, progress in applying them for most indications has been slow. One challenge in monitoring biomarker levels is that they are often measured only at discrete time points, ignoring fluctuations in abundance that occur between measurements. To address this limitation, a report in this issue1 proposes to measure the total abundance of a biomarker over time using an implanted sensor that is read out by nuclear magnetic resonance spectroscopy. Ling et al.1 apply the sensor to detect the accumulation of three heart-disease biomarkers in mice over 3 days after induced myocardial infarction. Notably, they show that the measurements correlate with the size of the infarct. Although still very preliminary with respect to clinical application, this approach might one day circumvent the need for multiple blood draws to monitor cardiac disease. Biomarkers—molecules associated with the presence or severity of disease—are touted as having greater potential to improve therapy and lower healthcare costs than any other recent innovation in medicine. In principle and in practice, biomarkers enable earlier detection Fred S. Apple is in the Department of Laboratory Medicine and Pathology, University of Minnesota School of Medicine and Hennepin County Medical Center, Minneapolis, Minnesota, USA. e-mail: [email protected]

of disease, better risk stratification, more effective therapeutic intervention and more reliable profiling of the response to treatment. In the field of cardiac biomarkers, the past 15 years have witnessed such remarkable progress in the clinical utility of biomarkers for myocardial infarction that the greatest remaining obstacle lies in improving analytical technologies to provide precise measurements at low concentrations. Yet the interpretation of cardiac biomarker measurements can still be confounded by fluctuations in abundance, depending on the patients’ clinical presentation. Ling et al.1 attempt to tackle this problem by focusing on a different parameter—cumulative biomarker levels, an approach that avoids the need to collect samples at multiple time points and could in principle provide valuable information not accessible through serial sampling. The authors create sensor discs (~8 mm in diameter and <2 mm thick) containing superparamagnetic iron oxide nanoparticles functionalized with antibodies specific for either cardiac troponin I (cTnI), creatine kinase-MB (CK-MB) or myoglobin. In the presence of the antibody target, the nanoparticles aggregate, decreasing the spin-spin relaxation time (T2) of water (Fig. 1). This irreversible clustering can be measured by nuclear magnetic resonance relaxometry and, if allowed to proceed for an extended period of time, provides a readout of total biomarker abundance analogous to that of a radiation dosimeter.

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

ne w s and v ie w s HDAC complexes. In their method, the HDAC inhibitor of interest is immobilized on sepharose beads, which are then added to cell lysates spiked with increasing concentrations of the free inhibitor. As higher concentrations of free inhibitor reduce binding of the immobilized inhibitor to its cellular targets, quantitative mass spectrometry of the bound HDAC complexes enables the determination of IC50 values. The authors study 16 HDAC inhibitors (including vorinostat and romidepsin) spanning a diverse set of ‘chemotypes’ (hydroxamic acids, ketones, carboxylic acids and amino benzamides). In total, 267 proteins from a human myelogenous leukemia cell line, including 6 HDACs and 29 proteins known to form complexes with HDACs, bind the chemical probes. The authors demonstrate the ubiquitous expression of these proteins across a panel of six human cell lines and a range of mouse tissue types, including brain, heart, kidney, liver, lung and spleen. Importantly, the authors show striking differences in potency and selectivity profiles for individual inhibitors across different protein complexes in cell lysates compared with the results of assays with recombinant proteins. The study reports several other notable findings. The aminobenzamide class of HDAC inhibitor does not bind the Sin3 multiprotein complex but shows particular selectivity for the HDAC3-NCoR complex. Valproate, considered a weak class I HDAC inhibitor, has higher affinity for the CoREST and NuRD complexes than for the Sin3 complex. The marketed orphan anti-inflammatory drug bufexamac binds class IIb (HDAC6 and 10) HDAC complexes. Accordingly, the drug induces tubulin hyperacetylation at concentrations within the range of its anti-inflammatory effects. Finally, the authors also identify a novel HDAC complex, named MiDAC, which forms during mitosis; division-arrested cells show increased deacetylation activity of MiDAC but not of other complexes. Although the method of Bantscheff et al.1 offers exciting possibilities for interrogating biological systems, the barrier to entry for most laboratories is not low. A prerequisite for its application is a probe matrix that efficiently captures the subproteome of interest. Researchers should therefore be prepared to undertake a synthetic chemistry effort combined with the determination of structureactivity relationships. In addition, thorough validation of the targets identified using complementary immunoprecipitation techniques is an important step in the process. These concerns aside, the new method shows promise as a means of probing the actual 236

interactions of small molecules with macro molecular complexes in different cellular contexts. Such studies may lead to drugs with beneficial tissue specificity. For example, a recent report9 highlights the importance of context-specific binding by showing that a small-molecule inhibitor of protein kinase C (PKC) binds its target only in vitro and not in cells due to a more stable PKC-AKAP79 cellular complex. In this cellular context, AKAP79 blocks binding of the inhibitor to its target9. It is tempting to speculate that this inhibitor may have tissue-specific effects that could not be anticipated from biochemical measurements alone. Thus, the ability to quantify binding interactions of small molecules with native protein complexes in different cellular

c ontexts as described by Bantscheff et al.3 should expand the utility of a wide range of small molecules as probes and drugs. COMPETING FINANCIAL INTERESTS The authors declare no competing financial interests. 1. Bantscheff, M. et. al. Nat. Biotechnol. 29, 255–265 (2011). 2. Zheng, X.-F. et al. Cell 82, 121–130 (1995). 3. Sarbassov, D.D. Curr. Biol. 14, 1296–1302 (2004). 4. Taunton, J., Hassig, C.A. & Schreiber, S.L. Science 272, 408–411 (1996). 5. Dey, P. Curr. Med. Chem. 13, 2909–2919 (2006). 6. Kazantsev, A.G. & Thompson, L.M. Nature Drug Dis. 7, 854–868 (2008). 7. Brunmeir, R. et al. Int. J. Dev. Biol. 53, 275–289 (2009). 8. Bantscheff, M. et al. Nat. Biotechnol. 25, 1035–1044 (2007). 9. Hoshi, N. et al. Mol. Cell 37, 541–550 (2010).

Biomarkers in aggregate Fred S Apple Implantable devices that measure the cumulative release of biomarkers promise new diagnostic options. For all of the excitement surrounding disease biomarkers, progress in applying them for most indications has been slow. One challenge in monitoring biomarker levels is that they are often measured only at discrete time points, ignoring fluctuations in abundance that occur between measurements. To address this limitation, a report in this issue1 proposes to measure the total abundance of a biomarker over time using an implanted sensor that is read out by nuclear magnetic resonance spectroscopy. Ling et al.1 apply the sensor to detect the accumulation of three heart-disease biomarkers in mice over 3 days after induced myocardial infarction. Notably, they show that the measurements correlate with the size of the infarct. Although still very preliminary with respect to clinical application, this approach might one day circumvent the need for multiple blood draws to monitor cardiac disease. Biomarkers—molecules associated with the presence or severity of disease—are touted as having greater potential to improve therapy and lower healthcare costs than any other recent innovation in medicine. In principle and in practice, biomarkers enable earlier detection Fred S. Apple is in the Department of Laboratory Medicine and Pathology, University of Minnesota School of Medicine and Hennepin County Medical Center, Minneapolis, Minnesota, USA. e-mail: [email protected]

of disease, better risk stratification, more effective therapeutic intervention and more reliable profiling of the response to treatment. In the field of cardiac biomarkers, the past 15 years have witnessed such remarkable progress in the clinical utility of biomarkers for myocardial infarction that the greatest remaining obstacle lies in improving analytical technologies to provide precise measurements at low concentrations. Yet the interpretation of cardiac biomarker measurements can still be confounded by fluctuations in abundance, depending on the patients’ clinical presentation. Ling et al.1 attempt to tackle this problem by focusing on a different parameter—cumulative biomarker levels, an approach that avoids the need to collect samples at multiple time points and could in principle provide valuable information not accessible through serial sampling. The authors create sensor discs (~8 mm in diameter and <2 mm thick) containing superparamagnetic iron oxide nanoparticles functionalized with antibodies specific for either cardiac troponin I (cTnI), creatine kinase-MB (CK-MB) or myoglobin. In the presence of the antibody target, the nanoparticles aggregate, decreasing the spin-spin relaxation time (T2) of water (Fig. 1). This irreversible clustering can be measured by nuclear magnetic resonance relaxometry and, if allowed to proceed for an extended period of time, provides a readout of total biomarker abundance analogous to that of a radiation dosimeter.

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

ne w s and v ie w s In vivo use of nanoparticle magnetic relaxation switches (MRSw) is not new, but Ling et al.1 are the first to show that biomarkers detected in this way correlate with disease severity. An earlier study2, which tried a similar approach to measure the beta subunit of human chorionic gonadotropin, could not correlate levels of this cancer biomarker with tumor progression owing to the xenograft model used. Ling et al.1 implant six MRSw discs subcutaneously per animal and measure changes in the levels of cTnI, CK-MB and myoglobin at 72 h after induced myocardial infarction. Their results show that these three biomarkers extravasate from serum into the subcutaneous space and that the signal intensity for each of them correlates with the extent of tissue death caused by the infarction. They also show that implanted MRSw sensors can detect the cardiotoxic effect of the chemotherapeutic drug doxorubicin, suggesting that the approach might be of value in screening drugs for cardiotoxicity1. Several limitations of the prototype device of Ling et al.1 would have to be addressed before it could become a serious player in the world of cardiac biomarkers. First, it should be redesigned to focus on sensitive detection of cTnI. Cardiac troponin T (cTnT) and cTnI are currently the preferred biomarkers for diagnosing myocardial infarction, with myoglobin and CK-MB assays increasingly falling from favor. Assays for cTnI and cTnT eliminate the need to measure CK-MB and myoglobin levels3 provided that they have sensitivities of ≤1 pg/ml4,5. To achieve this with the MRSw sensor, it would be important to choose optimal anti-cTnI antibodies, which would detect the multiple forms of cTnI (including free cTnI, binary and ternary complexed forms, and oxidized, reduced and phosphorylated isoforms of both free and complexed forms), ideally with equimolar cross-reactivity. Second, MRSw sensors would need a CV of <10% for clinical use, as required for cardiac troponin assays at their reference-value cutoff. These assays currently have a sensitivity of ~10 pg/ml4. Third, the detection window of the MRSw sensor should be optimized for use in a very early time frame. For patients presenting with symptoms of ischemia, the primary clinical goal is to detect myocardial injury, including

Dispersed MRSw (high T2)

Aggregated MRSw (low T2)

Figure 1 The aggregation of functionalized superparamagnetic iron oxide nanoparticles, also called magnetic relaxation switches (MRSw), in the presence of biomarkers decreases the transverse relaxivity (T2) of surrounding water protons, generating a signal detectable using magnetic resonance relaxometry. Ling et al.1 quantify a biomarker of interest (blue) by functionalizing nanoparticles (black) with polyclonal antibodies, represented as two classes of antibody (red and yellow).

infarction, as early as possible—within hours of the index ischemic event. The standard biomarker for this purpose is cTnI, as it can clearly differentiate myocardial necrosis from nonmyocardial pathologies because of its unique 100% cardiac-tissue specificity3. The four high-sensitivity cTn (hs-cTn) assays that have been described in the literature (three hs-cTnI assays and one hs-cTnT assay)4 detect ischemic acute coronary syndrome injury within 2 h6,7. In contrast, Ling et al.1 take measurements 24–72 h after myocardial infarction, a time frame that is not relevant to current clinical practice. Once these limitations have been addressed, the MRSw should be applied to determine both the short- and long-term biological variation of cardiac troponin in healthy, normal populations. This would entail studying a wide range of ages (by decade) within gender, and among different ethnicities and races. The current hs-cTnI and hs-cTnT assays detect biological variations in the 40–80% range; but the studies were very limited in the subject demographics. These findings would potentially provide baseline cTnI concentrations to be used in primary prevention studies8. In addition, incorporation of other novel biomarkers that would allow risk stratification of patients before the onset of a mycocardial infarction, identifying inflammation, plaque rupture

nature biotechnology volume 29 number 3 march 2011

or instability, ischemia and myocardial dysfunction, would be extremely valuable in assisting clinicians in overall patient management and therapy. Despite the challenges that remain before the device of Ling et al.1 could be approved for use in patients with acute and chronic heart disease, sensors that measure cumulative biomarker levels might find application in a range of other therapeutic contexts. For example, determining whether cancer survivors remain in remission would benefit from the ability to capture vanishingly small amounts of tumor biomarkers. In the short term, such approaches are likely to be of greater utility in basic research than in clinical practice. 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/. 1. Ling, Y. et al. Nat. Biotechnol. 29, 273–277 (2011). 2. Daniel, K.D. et al. Biosens. Bioelectron. 24, 3252–3257 (2009). 3. Thygesen, K. et al. Eur. Heart J. 31, 2197–2204 (2010). 4. Apple, F.S. Clin. Chem. 55, 1303–1306 (2009). 5. Morrow, D.A. & Antman, E.M. Clin. Chem. 55, 5–8 (2009). 6. Wilson, S.R. et al. Am. Heart J. 158, 386–391 (2009). 7. Reichlin, T. et al. N. Engl. J. Med. 361, 858–867 (2009). 8. de Lemos, J.A. et al. J. Am. Med. Assoc. 304, 2503–2512 (2010).

237

news and v iews

© 2011 Nature America, Inc. All rights reserved.

A modENCODE snapshot Since 2007 a consortium of research groups has been studying the genomes of two model organisms, the fruitfly Drosophila melanogaster and the nematode worm Caenorhabditis elegans, in a project called model organism encyclopedia of DNA elements (modENCODE)1. The latest results from this project were described recently in two papers in Science2,3 and a suite of companion papers in Nature and Genome Research (http://blog. modencode.org/papers). The studies report both massive genome-scale data sets and analytic strategies for data integration. They substantially increase the annotated fractions of the fly and worm genomes and provide a wealth of data for understanding these model organisms and for developing new bioinformatic methods. Here we provide an overview of the data and some perspective from scientists on challenges for the field. The goal of modENCODE is to catalog sequence-based functional DNA elements in the fly and worm genomes. Such a catalog may be used to study regulatory networks and other emergent properties of the genomes, and, perhaps, to better understand the human genome. The project also seeks to generate experimental reagents for use by the research community. A summary of the new data sets is presented in Tables 1 and 2. To increase the number of functional genomic regions discovered, the studies analyzed organisms at different developmental stages. For the fly, ~700 data sets were generated from whole embryos, larvae and adult female and male insects as well as from a few cell lines and tissues. For the worm, ~240 data sets covered all major developmental stages along with some mutants, isolated tissues and animals exposed to pathogens. For both organisms, microarrays and sequencing were used to characterize gene expression, the binding sites of transcription factors and other proteins associated with DNA, origins of DNA

238

replication, nucleosome turnover rates, salt-fractionated chromatin, the genomic locations of nucleosomes and the sites of different histone modifications. Looking forward, projects similar to modENCODE now seem feasible for studying other organisms and a broad range of biological problems. What will be the major challenges of such projects? Not sequencing, says Jun Wang, executive director of BGI in Shenzhen, China. He estimates that generating the equivalent of the fly modENCODE data set using today’s technology could take less than 2 months, including less than a month for library construction and a month for sequencing (although in practice more time may be required if several replicates are necessary). The main technical barrier, he says, will be preparing large numbers of samples from different tissues, developmental stages and conditions. A major challenge will be data integration, says Tom Gingeras of Cold Spring Harbor. For instance, robust integrative approaches are needed that combine genomic, transcriptional, regulatory and epigenomic signals, according to Olga Troyanskaya of Princeton University. Roded Sharan, a computational biologist at Tel Aviv University, agrees, adding that integrative analysis is required to identify an organism’s signaling and regulatory pathways and to elucidate how they vary over time and across cell types. He notes that current algorithmic work is focused on analyzing at most a few networks at a time and will have to be significantly scaled up to understand the complex developmental programs of fly or worm. Overall, says Troyanskaya, existing bioinformatic methods are not adequate, suggesting that novel ways of conceptualizing problems may be needed. New methods are needed to deal with the heterogeneity of the data types, to correct for technical and experimental biases and to detect biological signals hidden in experimental noise. Moreover,

the sheer volume of data will require new approximation algorithms, computational infrastructure and strategies for disseminating the results. As a result of modENCODE, the catalog of DNA elements has grown larger, but many questions remain unanswered. The translation of annotated genomes into systems-level descriptions of the fly and the worm is a long-term goal. In worm, the number of candidate noncoding RNAs has increased severalfold, up from 1,061 at the start of the project, but the biological roles of these RNAs are not yet clear. Moreover, “pervasive post-transcriptional regulation of gene expression emerges as a theme from the modENCODE data,” says Thomas Sandmann, a fly geneticist at the German Cancer Research Center. In worm, ~22,000 genes were found to generate ~65,000 different transcripts; in fly, 74% of the ~17,000 genes showed at least one transcript isoform that differed from previous annotations. “Frankly,” says Sandmann, “we don’t have a good idea what this complexity is good for or how it works.” Other open questions involve the evolutionary conservation of DNA functional elements. Manolis Kellis, a member of the modENCODE consortium, explains that a “next step is tackling the comparative analysis of fly and worm to each other and to human, to understand the conservation of the regulatory principles learned, and the relevance of our results to the study of human biology and disease.” Finally, also in progress is ENCODE—a sister project analyzing functional elements in humans—which published results from a pilot project several years ago and is now progressing into its next stage of analysis across the entire human genome.

Markus Elsner & H. Craig Mak, Associate Editors, Nature Biotechnology 1. Celniker, S.E. et al. Nature 459, 927–930 (2009). 2. Gerstein, M.B. et al. Science 330, 1775–1787 (2010). 3. The modENCODE Consortium. Science 330, 1787– 1797 (2010).

volume 29 number 3 march 2011 nature biotechnology

news and v iews Table 1 Fruitfly modENCODE Genome-scale measurements

Transcriptome Whole animals from 30 developmental stages, four Biological samples cell lines, analyzed limited coverage of some tissues and additional cell lines

© 2011 Nature America, Inc. All rights reserved.

Technologiesa

Insightsc Fig.d New annotations of genes, noncoding RNAs, alternative isoforms, transcriptional start sites and promoters

2

Chromatin signatures of functional elements

3,4

Transcript assembly, mapping to genome, RNA secondary structure prediction, evolutionary conservation

✓

Positional correlation, unsupervised hidden Markov models

✓

3

Logistic regression

Highly occupied target regions

5

Complexity score based on Gaussian kernel density estimation

Functional network of regulatory proteins matched to target genes, with interactions supported by expression and chromatin marks

7

Prediction of gene function for genes lacking a functional annotation

8

Prediction of stage-specific gene expression regulators

9a

Prediction of gene expression levels in embryo stages and cell lines

9b

9c

Chromosomal protein–DNA binding

Histone marks

DNA replication

Nucleosome occupancy

Genome sequences Alignments of 12 previously sequenced Drosophila species

Whole embryos; Whole three cell lines embryos; one to four cell lines

18 histone modification marks and histone variants in two cell lines; 5 marks across developmental time-course

One to three cell lines

Cell lines; some tissues and embryos

ChIP-chip (54), ChIP-chip ChIP-seq (6) (111), ChIPseq (32)

ChIP-chip (153), ChIPseq (69)

ChIP-chip (1), ChIP-seq (6), aCGH (5)

DNA tiling array (34)

Computational methods

Predictions of active promoters

Prediction of cell type–specific chromatin regulators

RNA tiling array (80), RNA-seq (141)b

Transcription factor–DNA binding

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Unsupervised machine learning methods

Guilt-by-association methods that combine co-expression and network Dynamic Regulatory Events Miner (DREM)

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

✓

Correlation in activity patterns Linear regression model trained using cross-validation

✓

✓

✓

✓

✓

✓

✓

✓

✓

aCGH, array comparative genome hybridization. aTechnologies used, followed by number of experiments. bAnd cDNA sequencing (3,204), CAGE (1) and RACE (8,727). cShaded boxes indicate data analyzed to derive insights. dFigure in ref. 3 showing results.

nature biotechnology volume 29 number 3 march 2011

239

news and v iews Table 2 Nematode worm modENCODE Genome-scale measurements

Transcriptome

Biological samples Whole animals of analyzed all major developmental stages, isolated tissues and pathogeninfected animals Technologiesa RNA tiling array (44), RNA-seq (58), 3'-RACEseq (14)

© 2011 Nature America, Inc. All rights reserved.

Insightsb Fig.c

Chromosomal protein–DNA binding

23 GFP-tagged transcription factors and RNA Pol II, primarily at developmental stage of highest expression

16 proteins in early- and/ or mixed-stage embryos

19 histone modi- Early- and mixedfication marks stage embryos, and 2 variants in adults early embryos and L3-stage larvae

ChIP-seq (34)

ChIP-chip (27)

ChIP-chip (34)

Histone marks

Nucleosome occupancy

Genome sequences

Alignments of seven previously sequenced nematodes

MNase-seq (5)

Computational methods

New annotations of genes, noncoding RNAs, alternative splicing, 3' and 5' UTRs and pseudogenes

1

Network of regulatory proteins matched to target genes and microRNAs

3

Highly occupied target regions

4

Statistical enrichment of binding sites

Megabase-scale domains of histone marks

5

Heat map

Models relating histone marks to transcription factor binding and gene expression

6,7

Evolutionary conservation of annotated regions including genes, transcription factor binding sites and noncoding RNAs

Transcription factor–DNA binding

Mapping to genome, iterative gene model building, RNA secondary structure prediction, supervised machine learning for identifying noncoding RNAs, evolutionary conservation

✓

Peak finding, interactions and activator-repressor status supported by expression

✓

✓

✓

✓

✓

8

Machine learning models, support vector regression

✓

✓

✓

✓

✓

✓

✓

✓

Multiple sequence alignment ✓

MNase-seq, micrococcal nuclease digestions followed by sequencing. aTechnologies

240

used, followed by number of experiments. bShaded boxes indicate data analyzed to derive insights. cFigure in ref. 2 showing results.

volume 29 number 3 march 2011 nature biotechnology

r e s e a r c h h i gh l i ght s

Rumen metagenomics

© 2011 Nature America, Inc. All rights reserved.

Jonas Lovaas Gjerstad

The cost-effectiveness of biofuel may depend on the availability of better lignocellulolytic enzymes. An ideal source of these catalysts would be microbes in cow rumen, which digest complex plant polysaccharides with remarkable efficiency. Yet most members of the rumen microbiome defy culture. Hess et al. incubated nylon bags containing switchgrass, a promising energy crop, in the stomachs of fistulated cows for 72 h. Organisms adhering to the partially digested plant material were sequenced, yielding 268 Gbp of data and >2.5 million coding sequences. The authors identified >27,000 genes encoding putative carbohydrate-active enzymes, 43% of which are novel (having <50% identity to known proteins). Of 90 candidate gene products tested for activity, 51 were active on at least one of ten different substrates, including two potential biofuel feedstocks. Further analysis of this catalog of biomassdegrading enzymes may help to overcome a major bottleneck in biofuel production. (Science 331, 463–467, 2011) PH

Single-molecule imaging Pulses of extremely bright X-rays have been proposed as tools for working out the structures of large biological macromolecules that cannot be determined by other methods. Seibert et al. report on progress toward this goal. They show diffraction patterns that result from pulses of 1,700,000 photons being scattered by single capsids of the 0.45 µm mimivirus, which is too large for cryoelectron microscopy and cannot be crystallized owing to a dense outer layer of fibrils attached to the capsid. The diffraction patterns can be reconstructed into an image having an estimated resolution of 32 nm. In principle, this approach can achieve a resolution of 1 nm or less, but the linear particle accelerator used in these experiments was not powerful enough to attain this limit. Notably, the method does not require modifications to the sample, such as staining, freezing, sectioning, radiolabelling or crystallization. (Nature 470, 78–81, 2011) CM

Chromatin determines transcription factor binding For most transcription factors, only a fraction of potential binding sites is occupied by the protein at any given time. What accounts for this selectivity is largely unknown. John et al. demonstrate in genome-wide experiments that the binding behavior of a glucocorticoid receptor is determined largely by chromatin structure. They investigate the relationship between the binding sites of the receptor after hormone stimulation with the pre-existing patterns of chroWritten by Kathy Aschheim, Laura DeFrancesco, Markus Elsner, Peter Hare & Craig Mak nature biotechnology volume 29 number 3 march 2011

matin accessibility in two cell types. Their comparison of ChIP-seq data with analysis of DNA accessibility by digital DNase I profiling reveals that up to 95% of the binding occurred at sites with an open chromatin formation. The dominant effect of chromatin structure can be modulated by the specific glucocorticoid receptor binding element present, as some motifs are more likely to be bound even if the DNA is inaccessible to DNase I, and by nearby binding sites of additional co-regulatory factors, such as AP-1. As glucocorticoids are widely used to combat inflammatory, autoimmune and allergic conditions, the study raises the intriguing possibility that patient responses to nuclear receptor ligand drugs might be stratified on the basis of epigenetic status (Nat. Genet. published online, doi:10.1038/ ng.759, 23 January 2011) ME

Profiling cancer drugs Chemogenetic profiling screens a drug against a panel of genetic mutants to predict the drug’s mechanism of action. This strategy has been demonstrated using yeast deletion mutants (deleting a drug target creates drug resistance), and something analogous has been attempted with panels of cancer cell lines. However, not all mammalian pathways are represented in yeast, and cancer cell lines have drawbacks that limit their broad application. Jiang et al. describe a chemogenetic profiling approach feasible in mammalian cells that uses RNA interference to knock down genes in pathways putatively targeted by drugs. In a proof-of-principle experiment, they tested 15 chemotherapy drugs in a Burkitt’s lymphoma cell line, chosen for its sensitivity to a broad range of drugs. Lines transfected with individual short hairpin RNAs (shRNAs) directed against 29 genes central to cell-fate decisions were treated with the drugs, creating a pattern of resistance for each drug across the set of cultures. Drugs with similar modes of action are expected to have similar patterns of resistance, and in fact the 15 drugs tested fell into six clusters, according to their modes of action. With this system, the authors tested three approved drugs with unknown mechanisms, which led to a seventh cluster, and characterized a derivative of a nitrogen mustard, finding that its mode of action was unchanged, even though the toxicity was increased. Finally, they showed that profiling an additional 16 known drugs with a subset of 8 of the initial 29 shRNAs was sufficient to correctly predict the drugs’ mechanisms of action. (Nat. Chem. Biol. 7, 92–100, 2011) LD

Off-the-shelf blood vessels Tissue-engineered blood vessels would be useful for certain cases of arterial disease that are not amenable to autologous or synthetic grafts. Blood vessels have been generated in vitro for autologous transplants by generating sheets of tissue from fibrolasts, wrapping the sheets around rods to form tubes, and seeding the lumens with endothelial cells. But this patient-specific method is costly and requires 6–9 months of culture time. Dahl et al. propose a more practical solution based on allogeneic cells. They seed smooth muscle cells on a biodegradable polymer scaffold and allow the cells to secrete extracellular matrix proteins, such as collagen. Then the construct is decellularized to remove immunogenic material, leaving a collagenous tube suitable for transplantation. The authors test the approach in animal models of arteriovenous access for hemodialysis and of peripheral and coronary artery bypass. The grafts remain patent for up to 6 months and are resistant to dilatation, calcification and intimal hyperplasia. (Science Transl. Med. published online, doi: 10.1126/scitranslmed.3001426, 2 February 2011) KA 241

PROFILE

David Haussler Human genome pioneer David Haussler talks about the evolving role of annotated data repositories.

© 2011 Nature America, Inc. All rights reserved.

W

hen the Human Genome Project was assembling DNA sequences to play catch-up with Celera (Rockville, MD, USA), the public initiative turned to David Haussler’s group at the University of California, Santa Cruz. Since then, Haussler’s team has built and maintained the UCSC Genome Browser (http://genome.ucsc.edu/) (Genome Res. 12, 996–1006, 2002), a repository for storing genome sequences and annotations, such as genes and transcripts, as well as a tool for data analysis and visualization. Remarkably, Haussler’s roots in computational biology— not that it was a discipline at the time—can be traced back to the intellectual hotbed of a weekly meeting for graduate students run by Andrzej Ehrenfeucht at the University of Colorado at Boulder, the same meeting that shaped the careers of Gary Stormo and Gene Myers, two other pillars in the field of bioinformatics. How did that weekly meeting at UC Boulder influence you? David Haussler: Gene and I and Gary all went to that meeting. Gene and I were in the computer science department, and Gary was with Larry Gold in molecular biology. At the time, the total amount of DNA that was available was the complete sequence of phi X, and a couple of other short viruses and snippets of Escherichia coli. You could save all of it on a little floppy disk. Right there in that seminar we started thinking, “Well, how do we apply computers to analyze DNA sequences?” I actually went to Boulder to work with Andrzej Ehrenfeucht on logic and theoretical computer science. Andrzej is a polymath. He is off-the-charts brilliant, and knows so much about so many different fields. But I tell you, I had not intended to study DNA, but in that group we covered so many areas

that we were completely bombarded by new things, and that’s certainly where it started. How has the creation of annotated data repositories changed biology over the past decade? DH: Biology is increasingly an informationdriven field. But molecular biology has historically been an information-limited field, where the means of communication have been the traditional means: presenting a talk at a conference and publishing a paper in a journal. That is an incredibly limited way to get information. To transcend this, all of the data need to be unified in one format that’s easily accessible—Google for the genome—where you can just make queries and find what you want. We tried to build that with the UCSC Genome Browser, and we think that that has had an

“All of the data need to be unified in one format that’s easily accessible—Google for the genome—where you can just make queries and find what you want.” enormous impact because it has given people orders of magnitude more information at their fingertips than they had before. Can you provide an example? DH: Let’s take comparative genomics. On the [UCSC Genome] Browser you can see the human genome compared with almost 50 other vertebrate genomes. For any gene or non-coding regulatory element, you can look at the corresponding element in these other vertebrates, to the extent that it exists. And so it’s very common to look and say, “here’s an enhancer for this gene and look, all the mammals have it and the nonmammals don’t!” Increasingly, now we want to supplement that evolutionary information with experimental evidence that says, hey, this DNA hypersensitivity track says that this [chromatin] is open and active during neural development. And, wow! This neural developmental transcription factor actually has a binding site in this enhancer.

nature biotechnology volume 29 number 3 MARCH 2011

You can see how it snowballs. It’s incredibly exciting to layer on these other functional data and to combine them with the evolutionary data, and you start to get a picture now that says, wow, this was really an innovation in mammals and it has something to do with brain development and it’s particularly controlled by these transcription factors, and it turns on these genes at certain times in development. And now biology [not computer science] is happening. We’re talking about storing, literally, millions upon millions of these types of data in an accessible format. Where do you see analytic tools going from here? DH: Our feeling is that, if you can answer a question interactively on the web, you should be able to do that, because seldom do people ask exactly the right question they wanted to the first time out. If it’s a more complicated question, you should be able to easily download the data you need to determine the answer offline. We also interact with a tool called Galaxy (Genome Biol. 11, R86, 2010), which is a workflow management system. It provides a visual interface that allows you to set up a pipeline of processing activities that you can run, edit, store and exchange with your friends. We have found that for more professional programmers, or people who want to go deeper, the combination of the UCSC Browser with the Galaxy workflow management system is a sweet deal. What about erroneous annotation in biological databases—how can this be tackled? DH: What we would like to stimulate is as much community feedback and community correction as possible. One of the great models for this is Wikipedia. Are there errors in Wikipedia? You bet. Do they get fixed? Well, there are a lot of eyes on it. And the more eyes on it, the more they do get fixed. Getting rid of errors is mainly a function of how many eyes you can put on it. There needs to be a mechanism, an easy feedback mechanism, so that when somebody spots an error then something can actually be done about it. Definitely much more can be done in this vein, but I think the more we go toward easy and direct community feedback, the better.

243

perspective

Beyond natural antibodies: the power of in vitro display technologies

© 2011 Nature America, Inc. All rights reserved.

Andrew R M Bradbury1, Sachdev Sidhu2, Stefan Dübel3 & John McCafferty4 In vitro display technologies, best exemplified by phage and yeast display, were first described for the selection of antibodies some 20 years ago. Since then, many antibodies have been selected and improved upon using these methods. Although it is not widely recognized, many of the antibodies derived using in vitro display methods have properties that would be extremely difficult, if not impossible, to obtain by immunizing animals. The first antibodies derived using in vitro display methods are now in the clinic, with many more waiting in the wings. Unlike immunization, in vitro display permits the use of defined selection conditions and provides immediate availability of the sequence encoding the antibody. The amenability of in vitro display to high-throughput applications broadens the prospects for their wider use in basic and applied research. For the past 35 years, hybridoma technology has enhanced our capacity for research and diagnostics by providing monoclonal antibody reagents to track, detect and quantify target molecules in cells and serum. Recently, several more advanced methods to harness the immune response1–3 have substantially increased the number of antibody-producing cells that can be screened. In addition to these improvements in more traditional methods of making monoclonal antibodies, researchers can also benefit from refinements in in vitro display technologies. Although their merits do not seem to be fully appreciated across the broader research community, display technologies permit more control over the nature of the derived antibodies than immunization. Since the conception4 and first implementation 5 of phage display, perhaps the most notable landmarks in the evolution of display technologies have been the expression of antibody fragments in bacteria6 and PCR-mediated amplification of antibody genes and libraries7–11. The most popular technologies, phage8,12,13 and yeast display14,15, which are complementary in their properties, can be used with naive, immunized or synthetic repertoires. The advent of high-throughput biology has dramatically increased the demand for renewable, high-quality affinity reagents for use in proteome-scale experiments. Whereas further advances in animal immunization technologies are expected to be slim, in vitro methods have the potential to substantially improve the parallelization, 1Los Alamos National Laboratory, Los Alamos, New Mexico, USA. 2University of Toronto, Toronto, Ontario, Canada. 3Technical University Braunschweig, Braunschweig, Germany. 4University of Cambridge, Cambridge, UK. Correspondence should be addressed to A.R.M.B. ([email protected]).

Published online 9 March 2011; doi:10.1038/nbt.1791

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

a utomation and miniaturization of antibody screens. Furthermore, a raft of recent papers16–22 point to an alarmingly high proportion of commercial antibodies demonstrating poor specificity, or even failing to recognize their targets at all. Given that much of modern biological research relies on the fidelity of commercially supplied antibodies, this underscores the importance of robust approaches to improve antibody quality. The high-throughput potential of in vitro techno logies make them ideal platforms for large-scale projects to derive antibodies for all human proteins. Once completed, these initiatives are likely to have impacts that potentially rival the completion of the human genome. Our primary goal here is not to reiterate the ways in which libraries used in display technologies are made or used. Instead, we discuss the scope of in vitro display technologies and what they have enabled, in particular in the context of distinguishing subtle differences in protein sequences and conformations or even minor changes in the chemical structures of small molecules. We hope to illustrate how in vitro display methods have yielded antibodies with remarkable properties, some of which have never been obtained by immunization. Most of the examples we discuss relate to antibody fragments. However, display technologies have allowed the development of nonantibody scaffolds. These also provide affinity reagents with similar, or in some applications, superior properties to those described here. Selection platforms23–25 and different scaffold proteins26–28, including antibody fragments29, have been reviewed elsewhere. Unique features of display technologies By permitting control over selection and screening conditions, display technologies allow the generation of antibodies against defined antigen conformations or epitopes (Fig. 1). For instance, the inclusion of competitors can direct selection toward specific targets. The use of variable regions from immunized sources with display technologies can also enable selection of specificities not detectable by traditional immunological techniques30. Because the gene encoding the antibody is cloned at the same time as the antibody is selected, simple subcloning steps after in vitro display permit the creation of constructs with added functionalities (Fig. 1). Libraries of mutagenized variants can be created and the same selection process repeated to yield variants with superior specificity and affinity. The improvement of antibody affinity to picomolar levels31–35 has become relatively routine, with one study describing an antibody in the femtomolar range36. These affinities are far higher than those of antibodies obtained by immunization, which are limited to ~100 pM by the physiological mechanism of B-cell activation37,38. In addition, antibody specificities can be broadened or narrowed by appropriate selection and screening.

245

Perspective

© 2011 Nature America, Inc. All rights reserved.

Figure 1 The unique capabilities of in vitro selection offer advantages over the immunization of animals for antibody generation. The direct coupling of the antibody and its encoding gene is characteristic of all display methodologies, including phage, yeast and ribosome display. Defined panning conditions with the desired buffer conditions, cofactors and competitors ensure that libraries can be screened using antigens with the desired conformation and biochemical properties to select for the requisite level of binder specificity and affinity. Binders can be selected sequentially, using different antigens to identify shared epitopes. The immediate availability of the antibody gene provides much additional value relative to antibodies obtained by immunizing animals.

As these in vitro methods are based on microbial systems, selection and screening are more amenable to automation than earlier hybridoma-based approaches. This provides the potential for highthroughput generation of binders39,40. In vitro methods also overcome immunological tolerance, allowing the selection of affinity reagents that recognize highly conserved targets such as ubiquitin41, histones42, hemoglobins43 and post-translational modifications44–46. Several approaches are available for overcoming tolerance during immunization47,48. However, none is required to select antibodies against conserved proteins using in vitro display methods. Remarkably, the selection of hundreds of different antibodies from naive human antibody repertoires against many different individual human targets has not been problematic39,49,50. Recognition of chemical modifications and small molecules A single methyl or hydroxyl group can have a considerable effect on the biological properties of a steroid hormone. Similarly, protein phosphorylation, acetylation and sulfation, all of which are relatively simple post-translational modifications in chemical terms, can dramatically affect signal transduction. Binders capable of discerning such relatively simple chemical modifications are of great value in studying these effects. Monoclonal and polyclonal antibodies with specificities for small molecules have been obtained by traditional immunization51–55. Even so, the ability of display methods to tailor both affinity and specificity has generated antibodies capable of discerning minor differences between related small molecules far better than those obtained by immunization (Table 1 and Fig. 2). Although space constraints do not permit discussion of all of these studies, the use of phage display to study tyrosine sulfation warrants particular mention. The sulfation of tyrosine residues is a post-translational modification predicted to occur in 30% of all secretory and membrane Table 1 In vitro selected antibodies recognizing small molecules and modifications Targets

Notes

Reference

6-monoacetylmorphine and morphine

Competition with morphine during panning to avoid cross-reactivity

154

Fluorescein

Affinity matured to 48 fM by yeast display

36

Testosterone, progesterone and 17β-estradiol

Structurally similar steroid hormones 155–159 with very different physiological effects

Sulfotyrosine as a posttranslational modification

Antibodies recognize all sulfotyrosinated proteins and peptides

Sulfur mustard–modified keratin

Antibodies recognize skin affected by 160 sulfur mustard

Fluorogenic dyes

Antibody binding increases dye 161 fluorescence up to 15,000 times by limiting conformational movement

Metallic gold

Two-step selection strategy

246

45,46

162

• pH • Salt concentration • Modulators • Competitors

In vitro antibody selection under controlled conditions defining requisite antigen conformation(s)

Sequence posting, web-based distribution, gene synthesis

Directed evolution to improve affinity, specificity, expression or stability Isolated antibody gene

Expression in other hosts

Fusion to Fc or enzymes, dimerization or multimerization, in vitro biotinylation

Protein knockouts with intrabodies for functional studies

proteins56. Despite decades of efforts involving immunization, it has proved impossible to generate antibodies recognizing sulfotyrosine using traditional means. This probably results from the innate tolerance of immune systems for such ubiquitous protein modifications, as well as the presence of the recognized target in the secretory pathway, resulting in retention and an inability to secrete the antibody. Using phage display, two groups recently selected antibodies recognizing proteins containing sulfotyrosine (but not tyrosine phosphate), independently of protein context or sequence45,46. These antibodies recognized sulfotyrosinated proteins in western blot analysis, immunofluorescence, enzyme-linked immunosorbent assays (ELISA) and immunoprecipitation, and recognition could be abolished by sulfatase treatment or preincubation with free tyrosine sulfate. This represents an enormous advance in the analysis of this modification, which has traditionally required thin-layer chromatography of radiolabeled protein hydrolysates57 or mass spectrometry58, with the presence of sulfate groups often inferred, rather than proven. The ability to select antibodies with such subtle recognition properties is likely to be a boon in the development of novel materials, in which antibodies could be used to facilitate and detect chemical patterning. Recognition of subtle differences in proteins In vitro display technologies allow the generation of antibodies against almost any target, including toxins, pathogens and antigens that are neither immunogenic nor highly conserved. With respect to protein targets, the exquisite specificity of the antibodies selected is exemplified by binders capable of differentiating, for example, as shown (see A.B. and collaborators59), between chicken and quail lysozyme (which differ by a single surface amino acid) and the SH2 domains of ABL1 and ABL2 (refs. 60,61 and Table 2 (which have 89% sequence similarity)) Phage antibody libraries have been widely used to select antibodies against infectious agents. These include antibodies that discriminate between different strains of hantavirus62, dengue virus63, influenza64,65, Ebola66 and Venezuelan equine encephalitis virus67. Given that many of these viruses are classified serologically, the ability to select phage antibodies with similar specificities is not surprising. Nonetheless, unlike antibodies generated by immunization, these have the potential to be used therapeutically. Human antibodies, some of which are protective in animal models68–70, have also been selected against a number of bacterial biothreat targets, including Brucella melitensis71, Burkholderia mallei and Burkholderia pseudomallei72, anthrax toxins 68,73,74 and spores 75, and botulinum toxin 31,76. VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

perspective Figure 2 In vitro selected antibodies can recognize minute differences in small molecules. (a) Antibodies against 6-monoacetylmorphine, the major heroin metabolite, do not recognize the closely related morphine 154. (b) Many different antibodies have been selected and subsequently had both affinity and specificity matured to specifically recognize 17β-estradiol, testosterone and progesterone without cross-reacting with closely related steroids (Table 1). (c) Antibodies against proteins bearing sulfated tyrosine residues do not recognize proteins containing either tyrosine or tyrosine phosphate45,46.

© 2011 Nature America, Inc. All rights reserved.

One library70 was generated from military donors vaccinated against a

plethora of different biothreat agents, reflecting the additional ability of display technologies to exploit antibodies generated during traditional immunization. The in vitro nature of phage display technology has been exploited to target particular features of blood cells. In one study43, antibodies recognizing fetal hemoglobin, but not adult hemoglobin, were selected by depleting high-affinity, cross-reactive antibodies followed by a selection against the fetal protein. Notably, the selected discriminatory antibody was of much lower affinity than cross-reactive antibodies, demonstrating the power of negative selections to favor clones with desirable binding specificities, even if their affinity is lower. Similar methods applied to cells have been used to select antibodies specifically recognizing fetal nucleated red blood cells77. Protein allostery is a common means for the regulation of protein function, and many signaling proteins exist in alternative conformational states that mediate different cellular responses. Antibodies that recognize specific protein conformers are powerful tools for probing the details of cell signaling (Table 2). However, the generation of such antibodies by immunization is complicated by the difficulty of maintaining a particular protein conformation in an immunized animal. Two ways in which in vitro selection technologies can address these limitations involve the use of negative selections to deplete nonspecific binders as well as those binders recognizing the nondesired protein conformation, and then following these with affinity maturation strategies to fine-tune specificity. In one study, single-chain Fvs (scFvs) specific to the GTP-bound form of the small GTPase Rab6 were generated by performing selections against a GTP-locked mutant78. In another study, one of our groups (S.S. and collaborators 79) used small molecules to covalently lock caspase-1 in either the active or inactive form and the locked antigens were used to select Fabs that

a

HO

HO

O O H

O H

H N

H N

H N

O N H

HO

O Monoacetylmorphine

b

c

Morphine

O

OH H H

O 17β-Estradiol

H

H N

S

OH

Tyrosine sulfate

O

O N H

HO

Tyrosine

OH OH H H

Testosterone H

H N

O

O

O

N H O

Progesterone

H H

O

P

OH

Tyrosine phosphate

OH

H

O

were highly selective for either the ‘on’ or ‘off ’ form of the protease. The concept of using in vitro selections to generate conformation- specific antibodies has also been combined with selections on whole cells in a powerful strategy that enables the probing of cell surfaces for conformational changes in response to various stimuli80. Phage display has also generated antibodies able to recognize structured RNA molecules81, which are essentially nonimmunogenic, and not amenable to detection using simple nucleotide probes (Table 2). By ensuring a nuclease-free in vitro environment and selecting under conditions optimized for the structural stability of the RNA, high-affinity Fabs were isolated against a structured domain from the Tetrahymena group I intron. These results establish general methods applicable to the generation of antibodies against other structured RNAs and may be useful to decipher the biological roles of the vast numbers of noncoding RNAs found in metazoan transcriptomes. Recognition of cell surface receptors Communication between cells is largely controlled by interactions between cell surface receptors and the ligands they recognize.

Table 2 In vitro selected antibodies recognizing protein sequences and conformations Target

Notes

Reference

Peptide MHC complexes (similar to T-cell receptor recognition)

Similar antibodies obtained by immunization have lower affinities

163–168

Fibronectin splice variants, EDA and EDB

Selection directed toward recognition of both human and mouse variants, allowing same antibody to be used in both models and clinical studies

169–171

Fetal hemoglobin Fetal nucleated red cells

Allow identification and potential purification of fetal red blood cells in maternal 43 circulation 77

GTP-bound Rab6

Antibodies were used to track activated Rab6 in the cell as GFP fusions

78

Caspase 1

Antibodies recognize either the ‘on’ or ‘off’ forms

79

Integral membrane proteins

CitS from Klebsiella pneumoniae and KcsA from Streptomyces lividans. KcsA antibodies used as crystallization chaperones

172,173

RNA

Structured domain from Tetrahymena group I intron. Antibody used as crystallization chaperone

80

ABL1 versus ABL2

Differ by only 11%

60,61

Chicken versus quail lysozyme

Differ by only four amino acids, of which only one is surface exposed

59

MHC, major histocompatibility complex.

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

247

Perspective

© 2011 Nature America, Inc. All rights reserved.

Figure 3 Mechanisms for blocking or activating receptor signaling using antibodies. The EGF receptor, a single transmembrane domain with multiple extracellular domains (ovals) with different functional domains, is used to exemplify mechanisms by which antibodies can block signaling by different classes of receptor. In this example, binding of ligand (green circle) occurs at domain 3, receptor dimerization occurs through domain 2 and interactions between domains 2 and 4 stabilize the ‘closed’ conformation of the receptor. (a–d) Antibodies can block signaling by binding to the ligand and preventing interaction with receptor (a), binding the ligand-binding site of the receptor and preventing interaction with ligand (b), preventing dimerization by binding the dimerization domain or sterically blocking the interaction (c) or stabilizing the closed conformation of the receptor (d). (e) Activation can occur by binding the ligand-binding site typically with bivalent antibodies.

Antibodies can modify such interactions and many therapeutic antibodies exert their effects by interfering in communications at the cell surface using different mechanisms (Fig. 3 and Table 3). In vitro display technologies provide a powerful route to generating functional antibodies that interfere with normal or pathological extracellular signaling. Although it is usually difficult to select for function directly, display technologies have the ability to generate thousands of independent binders, each of which can then be screened for functional activity. For example, >1,200 different antibodies directed toward B-lymphocyte stimulator (BLyS) were generated by phage display82. This large panel was subsequently screened using biochemical and cellular assays to identify antibodies that bound to BLyS, preventing its interaction with the receptor (Fig. 3a) and thereby blocking B-cell activation. In some cases, blocking antibodies with subnanomolar affinities were isolated directly from the naive antibody–phage display library82. One of these antibodies, specific only for the secreted form of BLyS (Benlysta), was affinity matured83 and is close to approval for treatment of systemic lupus erythematosus. Similar results have been reported for the selection of phage antibodies against a panel of 28 different, potentially therapeutic targets, with an average of 120 functionally active (that is, antagonistic or agonistic) antibodies selected per target50.

a

b

c

d

e

1 2 Ligand 3 4 Cell membrane Signaling

An alternative strategy to block receptor signaling is to target the ligand-binding sites on the receptors, thereby blocking access to the natural ligand (Fig. 3b). This was used to select antibodies that prevent the interaction of insulin-like growth factor type 1 (IGF-1) with the IGF-1 receptor84. Several groups of receptor binders were generated that competed with ligand binding and blocked cell growth in vitro and in vivo. These antibodies were also found to reduce receptor expression by internalization and catabolism. Studies on a panel of therapeutic antibodies targeting the epidermal growth factor (EGF) receptor (ErbB1) have also shown competition with ligand binding. However, antibodies can also block receptor signaling by alternative mechanisms 85. The four extracellular domains of ErbB1 adopt a mainly closed conformation in the absence of ligand, and a more extended conformation (allowing dimerization and subsequent phosphorylation of the intracellular domain) in the presence of ligand. Structural studies have shown that whereas antibodies such as zalutumumab keep intracellular domains apart, preventing phosphorylation (Fig. 3c), cetuximab (Erbitux) stabilizes the receptor in the closed conformation (Fig. 3d). Among the anti-ErbB2 antibodies, pertuzumab (Omnitarg) appears to work by preventing dimerization (Fig 3c), whereas trastuzumab (Herceptin) prevents receptor shedding and

Table 3 In vitro selected antibodies recognizing cell surface receptors Target

Notes

Reference

BLyS

Systemic lupus erythematosus

82

Tumor necrosis factor-α Phage display was used to convert a murine monoclonal antibody into a human antibody by guided selection. Rheumatoid arthritis, ankylosing spondylitis, chronic plaque psoriasis and Crohn’s disease, antibody developed by guided selection phage display

174,175

IGF-1 receptor

Blocking of ligand-binding site of receptor and receptor downregulation by endocytosis. Potential application in cancer

84

Notch

Prevent proteolysis of juxtamembrane NRR domain

86,87

Met

Dimeric antibodies are agonistic, monomeric ones are antagonistic and prospected for non-small cell lung cancer

88

MuSK

Agonistic antibodies demonstrate that MuSK activation is capable of triggering a key event in neuromuscular junction formation

176

CD40

Agonistic antibodies that activate normal human B cells suppress HIV-1 infection in vitro

177

Hemagglutinin

Antibodies recognize a previously unknown conserved conformational epitope. Isolated from both naive and immunized libraries

30,90,91

EphA2 and CD44

Selected from phage antibody library on yeast-displayed antigen, followed by selection for internalization on cells

94

CD166 ErbB2 Transferrin receptor EGFR

Internalizing antibodies selected directly for internalization on cancer cells (CD166 on prostate, ErbB2 and transferrin receptor on breast, EGFR on A431). Antigen identified after selection. Potential utility for internalization of chemotherapeutics

178,179 92 92

TRAIL-R

Over 500 different scFvs and Fabs isolated by phage display

93 89

NRR, negative regulatory region.

248

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

perspective forms inactive tetramers 85. Although the Table 4 Affinity and specificity maturation of antibodies by in vitro selection methods original blocking antibodies in these examTarget Notes Reference ples were generated from mice, they demonAffinity maturation strate the therapeutic approaches that could b enefit from human antibodies isolated HIV CDRs targeted for mutation, 15pM affinity 35 directly from display technologies. c-erbB-2 CDRs targeted for mutation, 13pM affinity 34 Antibodies that block Notch signaling Insulin Ribosome display, random errors, 82pM affinity 33 reveal yet another mechanism of action. Following ligand binding, a conformaFluorescein Affinity matured to 48 fM by yeast display 36 tional change occurs at the juxtamembrane Modification of recognition specificities negative regulatory region. This exposes CXCL10 and CXCL9 Antibody selected against CXCL10 and evolved to also 102 a protease cleavage site, resulting in the recognize CXCL9 release and translocation to the nucleus of the intracellular domain. In addition to VEGF and ErbB2 Antigens are completely unrelated, and antibody binds with 103 3 nM and 0.2 nM affinity to VEGF and ErbB2, respectively generating antibodies that block the inter Botulinum toxins A, B, E and F One antibody able to recognize all botulinum types afflicting 107 action with ligand, antibodies recognizing man was selected by yeast display the negative regulatory region domains stabilized the ‘closed’ confirmation of the Notch receptor (Fig. 3d), preventing the proteolytic cleavage Improving antibody affinity and specificity Although initial leads can be used directly as affinity reagents, a and translocation of the intracellular domain86,87. Dimeric antibodies targeting ligand-binding domains sometimes major advantage of in vitro methods is that it is possible to further mimic the natural ligand, causing receptor activation rather than improve function by constructing secondary libraries that introduce inhibition. This is the case for antibodies recognizing c-Met88, with additional mutations. Secondary libraries are most commonly used monomeric antibodies being antagonistic. Even so, in the case of to improve affinity, and all three major display formats (phage, yeast tumor necrosis factor–related apoptosis-inducing ligand receptor 1 and ribosome) have been applied to develop affinities that exceed (TRAIL-R1) and receptor 2 (TRAIL-R2) (ref. 89), an analysis of those possible with natural antibodies (Table 4). Both stepwise97 and >500 distinct selected antibodies revealed some that were agonistic computational98 methods generate similarly high-affinity antibodies, even as monomeric scFvs or Fabs. This is difficult to reconcile with but they have not been used as widely as in vitro display methods. the mode of action of TRAIL, which is a homotrimeric ligand that There are many examples of in vitro affinity maturation, and here causes multimerization of TRAIL receptors, leading to apoptosis par- we highlight some key studies. In ribosome display, each selection cycle involves a PCR-amplification step, which is ideal for introducticularly in tumor cells overexpressing the receptor. Antibodies also have great potential in blocking protein interactions ing additional mutations by error-prone PCR. This strategy has been associated with viral entry into target cells, illustrated by antibodies used to simultaneously select and affinity-mature anti-insulin antiselected from naive antibody libraries against recombinant hem bodies with affinities in the subnanomolar range33. Yeast-displayed agglutinin 5 (H5) influenza ectodomain90,91. Structural analysis of one libraries are smaller than phage and ribosome libraries, yet they allow of the antibodies showed that it bound to hemagglutinin at a highly quantitative and exhaustive screening by fluorescence-activated cell conserved, previously unrecognized pocket found in many different sorting. Coupled with sequential rounds of error-prone PCR, modest influenza viruses. Binding prevents the structural reorganization libraries of 105–107 unique clones were sufficient to affinity-mature required for membrane fusion, rendering the antibody neutralizing. an anti-fluorescein scFv to affinities <100 fM36. Although antibodies have not been generated against this epitope by Antibody engineering efforts usually strive to obtain specificity traditional immunization or infection, antibodies with similar VH for an antigen of interest. However, in certain applications, defined gene usage and neutralizing activity have been selected from phage cross-reactivity is extremely useful. Species cross-reactivity allows antibody libraries created from recently infected individuals30, show- better assessment of therapeutic efficacy and toxicity in animal ing that phage display can access the diversity of immune responses models. Unfortunately, cross-reactive antibodies are often difficult in ways not possible by traditional immunological means. to obtain by hybridoma methods owing to tolerance. In contrast, In vitro selection schemes have also been devised that allow the in vitro libraries are unaffected by immune tolerance and antibodies direct selection of antibodies mediating internalization92. This was targeting conserved sites across species have proven to be more the carried out by incubating phage libraries with target cells and iso- rule than the exception. For highly conserved proteins, such as vaslating those phage antibodies found within the cell after removing cular endothelial growth factor (VEGF), human/mouse cross-reactive phage antibodies bound to the cell surface. The recognized antigen antibodies have been obtained directly from naive libraries99,100. In is usually identified after selection. However, the use of mammalian the case of less conserved orthologs, such as BAFF/BLyS receptor 3 cells transfected with the target of interest93, or yeast displaying (BR3), initial anti-human antibodies with weak cross-reactivity to the targets of interest on their surface94, provides a means of carry- mouse protein have been obtained from naive libraries and evolved ing this out on predetermined targets. This approach is particularly to be highly cross-reactive101. Similar approaches have been used suitable for the selection of antibodies used for specific targeting to generate antibodies recognizing two closely related chemokines of chemotherapeutics95,96. (C-X-C motif ligand 10 and C-X-C motif 9) 102, thereby permitting In summary, antibodies and other binding molecules provide a means neutralization of two human chemokines with a single antibody. of modulating biological function by specifically interfering in protein Most attempts to engineer specificity involve improving on pre interactions. In vitro display systems provide a means of presenting existing weak recognition, due to homology between the recognized targets in appropriate conformations, including on cell surfaces. This targets. In perhaps the most extreme case of engineered crossfacilitates rapid screening for potentially rare functional binders. reactivity, trastuzumab has been evolved to recognize a very different nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

249

Perspective

© 2011 Nature America, Inc. All rights reserved.

Figure 4 An engineered dual specificity synthetic Fab. The bH1 Fab binds to both Her2 (orange, Protein Data Bank (PDB) ID: 3BDY) and VEGF (red, PDB ID: 3BE1). The heavy and light chains of the Fab are colored cyan/gray or blue/black respectively, with the different colors derived from structures of bH1 binding to either Her2 or BEGF.

protein, vascular endothelial growth factor (VEGF), as well as its original target, ErbB2 (ref. 103). After considerable evolution, the affinities for both targets were comparable to those of therapeutic antibodies (Kd = 3/0.2 nM for VEGF/ErbB2). The antibody inhibited both VEGF and ErbB2-mediated cell proliferation in vitro and tumor progression in mouse models. The structures of the bispecific Fab in complex with ErbB2 or VEGF revealed a common paratope, with the ErbB2 functional paratope located predominantly on VH, and that for VEGF on VL (Fig. 4). The ability to design antigen-binding sites with dual specificity against structurally unrelated antigens may be important in therapeutic strategies targeting two distinct signaling pathways with a single antibody. The ability to improve affinity and broaden specificity also has major implications for the development of antibodies against pathogens. For the effective inhibition of viral infection and bacterial toxins, antibodies would ideally recognize a variety of antigen subtypes with high affinity, to afford broad protection against pathogen variants. Furthermore, several studies have shown that multiple antibodies targeting distinct epitopes provide synergistic effects necessary for effective neutralization of pathogens104,105. In vitro antibody technologies provide an effective means for achieving these demanding criteria, as exemplified by a long-term study of neutralizing antibodies against the botulinum neurotoxin. Phage antibody libraries from immunized mice and humans resulted in the isolation of three antibodies recognizing nonoverlapping epitopes on botulinum neurotoxin106. The use of these three antibodies together as an oligoclonal IgG provided strong synergy that dramatically increased toxin neutralization. A long series of affinity and specificity maturation cycles using yeast display resulted in the final development of a remarkable antibody able to recognize botulinum toxins A, B, E and F, the four serotypes afflicting humans107,108. Exploiting the recombinant nature of antibodies selected in vitro All in vitro selection systems immediately provide the genes and corresponding sequences of antibodies selected against a particular target. This provides ready access to additional antibody formats by simple subcloning. Functions adopted using this ‘gene-based’ approach include dimerization109, multimerization110,111 and fusions to enzymes112, tags113 or fluorescent proteins114 (Fig. 1). Fusion to alkaline phosphatase is a particularly useful example of improved functionality. As this is a dimeric enzyme, fusing antibodies, either individually or as libraries, to alkaline phosphatase simultaneously provides dimerization and alkaline phosphatase activity, greatly facilitating screening by increasing the effective affinity and avoiding the need for secondary reagents39,112. Short peptides acting as in vivo biotinylation tags113, placed at the C terminus of antibody fragments, allow stoichiometrically defined, site-specific antibody biotinylation, as well as straightforward multimerization115. Antibody fragments can also be transformed into full-length antibodies116, or scFv-Fc fusions, which are very similar in many aspects117. The use of engineered Fc regions can result in improved pharmacokinetics and effector functions (for reviews, see refs. 118,119), including bispecific IgG, in which engineering of two different Fc regions allows only heterologous pairing120,121. Other approaches to generate bispecific antibodies build upon the observation that some scFv fragments form bivalent dimers 250

(diabodies)122, trimers123,124 and even tetramers125 when the VH/ VL linker is shortened. Additional bispecific antibody designs are reviewed elsewhere126. Even more radically, completely novel biochemical entities have been added to antigen-binding fragments. Fusions of scFv and Fab fragments to heterologous proteins, such as interleukins and cytokines127,128, apoptotic ligands, enzymes, toxins or RNases (see refs. 129,130 for reviews) have allowed novel therapeutic paradigms. Many of the above candidate therapeutic antibody constructs arose from antibody genes initially isolated from mouse hybridomas, but this is expected to change as more human antibodies are made available from engineered repertoires. Microinjected antibodies have long been used to knock out intracellular functions131. Antibody fragments can be expressed within target cells and targeted to various subcellular compartments116,132 by adding suitable signal sequences, allowing visualization or functional modification of proteins in different compartments. Removing the standard leader sequence results in cytoplasmic expression, whereas the addition of a nuclear localization signal results in antibodies being translocated to the nucleus. The combination of a leader sequence and the endoplasmic reticulum (ER) retention sequence retains expressed antibodies in the ER and has been used to prevent the expression of membrane proteins by sequestration in the ER. These include human interleukin 2 receptor, the ErbB2 receptor, β-amyloid precursor protein, vascular adhesion molecule 1 and many others133–136. The advantage of this strategy is that it requires antibodies that bind to any accessible epitope to provide the functional knockout, as opposed to the functional activity required of cytoplasmically expressed antibodies. Functional studies of membrane receptors or secreted proteins can thus be attempted by a single standardized subcloning step immediately after in vitro antibody selection, providing equivalence to RNA interference knockdowns at the protein level. Although expression in the secretory pathway is straightforward, folding of antibody fragments in the cytoplasm is far more challenging, owing to the reducing environment and the absence of specific chaperones, which prevents disulfide bond formation 137. Despite these problems, there are examples where cytoplasmic proteins have been targeted with intracellular scFvs78,138. The success of this approach has been improved by the creation of libraries of particularly stable scFvs139–141, preselecting antibodies for functional cytoplasmic expression142,143, or by using binder libraries based on VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

perspective

© 2011 Nature America, Inc. All rights reserved.

molecular scaffolds that do not rely on disulfide bond formation, such as engineered ankyrin repeat proteins144,145. An important advantage of using such protein-based allosteric blockers is their ability to generate very specific binders capable of distinguishing between closely related family members. Although the need to genetically modify the target cell is a major disadvantage, this has been partly alleviated by fusion to internalizing sequences that allow antibodies to enter the cell from the outside146. High-throughput selection by in vitro display methods The ease with which antibodies can be selected, screened and produced by in vitro display technologies enables simpler and faster generation and screening of antibodies than with hybridomas. Typically, a panel of ELISA-positive monoclonal antibody fragments can be generated within 2 weeks. Early experiments demonstrated the feasibility of semi-automated selection and/or screening of phage antibody libraries147–149 on small numbers of targets. More recently, selections on >400 different antigens were successful, with 54% of bacterially produced and 88% of mammalian-produced antigens39 yielding antibodies. The differences between the two protein classes probably arise from differences in the levels of correct folding. In a recent international comparative study, antibodies were raised to 20 different human Src-homology 2 (SH2) domains using hybridoma or phage display. Results from two of the participating phage display laboratories60,61 show that antibodies (some with subnanomolar affinities) were generated against all antigens, with 55% of positive antibodies specific for target SH2 domains when assessed against the entire SH2 panel. These antibodies were validated in a broad range of assays, including microarray analysis, immunoblotting, immunofluorescence and immunoprecipitation. Prospects If antibodies selected by in vitro methods are so powerful, why are they not more widely perceived as valuable research reagents? Part of the answer lies in the difficult patent situation, which resulted in restriction of this technology to the high-margin therapeutic markets for commercial use. It is worth noting in this regard that hybridoma technology was never patented, and achieved relatively wide acceptance within a short period. The situation for some of the core phagedisplay patents is now changing rapidly, as most platform patents have either expired or will do so over the next few years150. The technology may become more widely disseminated as a result. Although largely unrecognized by the research community, it is worth emphasizing that some commercial ‘monoclonal antibodies’ are actually recombinant antibodies selected by phage display and then reformatted to look like traditional murine antibodies by the fusion of Fc regions to human variable regions (e.g., the sulfotyrosine antibody described above45). Indeed, unmodified recombinant Fab fragments selected by phage display are commercially available. It therefore seems that the most important impediments to widespread adoption are a lack of knowledge of the capabilities of this technology, coupled with limited expertise and library availability. Furthermore, the number of companies willing to carry out in vitro selection for a fee is vanishingly small compared with the 180 companies willing to generate antibodies by immunization16. Another explanation of why the strength of display technologies appears to be underappreciated relates to the difficulties encountered in expressing many of these antibodies. Although some of the specificities described above are remarkable, the expression and stability of antibody fragments varies enormously—from exceptionally stable scFv fragments used in clinical trials151 to other fragments with extremely low expression nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

levels. A typical selection almost always generates several different binders to any well-folded antigen. Among these, usually at least one is sufficiently stable and well produced for research use. Furthermore, it is expected that stability and expression levels will improve as libraries are based on more stable scaffolds152. The studies described above indicate that this goal can now be met in highly parallelized screening setups with low effort per antigen60,61, provided that libraries of sufficient diversity and optimized protocols are used. Furthermore, stability and expression screening can be easily included as part of the high-throughput screening process. An additional issue with antibodies derived in vitro is that they are either not glycosylated if expressed in bacteria, or incorrectly glycosylated if expressed in standard yeast strains. If correct glycosylation is necessary, this can be overcome by expression in human cells or yeast modified to give human glycosylation patterns153. Once an antibody is generated, it can be defined precisely by sequence and even ‘distributed’ in this way. Gene synthesis is progressing at a remarkable pace, with the cost-per-base of synthesized genes falling dramatically. In fact, genes corresponding to the sequences of specific antibody fragments can now be synthesized for less than the cost of purchase of some antibodies from traditional vendors. The present state of this field can be compared to the situation for sequencing technologies at the start of the human genome project. Just as enormous technical advances occurred in the human genome project once it was started and rigorous industrial processes were applied, so we anticipate dramatic improvement in all aspects of selection, screening, downstream use and distribution of affinity reagents derived in vitro once a proteome-scale project is initiated and financed. In summary, in vitro display technologies permit the facile generation of antibodies by providing access to billions of potential binders in large ‘universal’ or immune display libraries. The technologies facilitate production, screening and maturation of selected binders, allowing selection on target conformations and formats not possible by more traditional routes based on immunization. Furthermore, the easy availability of the gene sequence not only provides a definitive description of the product but also allows electronic sharing and re-creation of the binding molecule through gene synthesis. The last 20 years have witnessed the successful application of display technologies to the development of therapeutic antibody candidates. In the coming decade we expect to also see increased realization of the benefits of this technology within the research and diagnostic markets. Acknowledgments A.R.M.B. is grateful to the US National Institutes of Health (P50GM085273 and R01-HG004852-01A1), US Department of Energy (GTL program) and the US Department of Defense, Defense Threat Reduction Agency for funding. S.D. gratefully acknowledges funding by the EU 7th framework programme (Projects: Affinomics and AffinityProteome). J.M. is pleased to acknowledge funding by the Wellcome Trust. 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.

2.

3. 4.

Love, J.C., Ronan, J.L., Grotenbreg, G.M., van der Veen, A.G. & Ploegh, H.L. A microengraving method for rapid selection of single cells producing antigenspecific antibodies. Nat. Biotechnol. 24, 703–707 (2006). Jin, A. et al. A rapid and efficient single-cell manipulation method for screening antigen-specific antibody-secreting cells from human peripheral blood. Nat. Med. 15, 1088–1092 (2009). Reddy, S.T. et al. Monoclonal antibodies isolated without screening by analyzing the variable-gene repertoire of plasma cells. Nat. Biotechnol. 28, 965–969 (2010). Smith, G.P. Filamentous fusion phage: novel expression vectors that display cloned antigens on the virion surface. Science 228, 1315–1317 (1985).

251

Perspective 5. 6. 7.

8. 9.

10. 11.

12.

13. 14.

© 2011 Nature America, Inc. All rights reserved.

15.

16. 17.

18.

19.

20.

21. 22. 23. 24.

25. 26. 27. 28. 29. 30.

31. 32.

33.

34.

35.

36.

37. 38.

252

Scott, J.K. & Smith, G.P. Searching for peptide ligands with an epitope library. Science 249, 386–390 (1990). Skerra, A. & Pluckthun, A. Assembly of a functional immunoglobulin Fv fragment in Escherichia coli. Science 240, 1038–1041 (1988). Larrick, J.W. et al. Rapid cloning of rearranged immunoglobulin genes from human hybridoma cells using mixed primers and the polymerase chain reaction. Biochem. Biophys. Res. Commun. 160, 1250–1256 (1989). Marks, J.D. et al. By-passing immunization. Human antibodies from V-gene libraries displayed on phage. J. Mol. Biol. 222, 581–597 (1991). Orlandi, R., Gussow, D.H., Jones, P.T. & Winter, G. Cloning immunoglobulin variable domains for expression by the polymerase chain reaction. Proc. Natl. Acad. Sci. USA 86, 3833–3837 (1989). Huse, W.D. et al. Generation of a large combinatorial library of the immunoglobulin repertoire in phage lambda. Science 246, 1275–1281 (1989). Sastry, L. et al. Cloning of the immunological repertoire in Escherichia coli for generation of monoclonal catalytic antibodies: construction of a heavy chain variable region-specific cDNA library. Proc. Natl. Acad. Sci. USA 86, 5728–5732 (1989). McCafferty, J., Griffiths, A.D., Winter, G. & Chiswell, D.J. Phage antibodies: filamentous phage displaying antibody variable domains. Nature 348, 552–554 (1990). Breitling, F., Dübel, S., Seehaus, T., Klewinghaus, I. & Little, M. A surface expression vector for antibody screening. Gene 104, 147–153 (1991). Boder, E.T. & Wittrup, K.D. Yeast surface display for screening combinatorial polypeptide libraries. Nat. Biotechnol. 15, 553–557 (1997). Feldhaus, M.J. et al. Flow-cytometric isolation of human antibodies from a nonimmune Saccharomyces cerevisiae surface display library. Nat. Biotechnol. 21, 163–170 (2003). Bordeaux, J. et al. Antibody validation. Biotechniques 48, 197–209 (2010). Jositsch, G. et al. Suitability of muscarinic acetylcholine receptor antibodies for immunohistochemistry evaluated on tissue sections of receptor gene-deficient mice. Naunyn Schmiedebergs Arch. Pharmacol. 379, 389–395 (2009). Jensen, B.C., Swigart, P.M. & Simpson, P.C. Ten commercial antibodies for alpha1-adrenergic receptor subtypes are nonspecific. Naunyn Schmiedebergs Arch. Pharmacol. 379, 409–412 (2009). Spicer, S.S., Spivey, M.A., Ito, M. & Schulte, B.A. Some ascites monoclonal antibody preparations contain contaminants that bind to selected Golgi zones or mast cells. J. Histochem. Cytochem. 42, 213–221 (1994). Pozner-Moulis, S., Cregger, M., Camp, R.L. & Rimm, D.L. Antibody validation by quantitative analysis of protein expression using expression of Met in breast cancer as a model. Lab. Invest. 87, 251–260 (2007). Grimsey, N.L. et al. Specific detection of CB1 receptors; cannabinoid CB1 receptor antibodies are not all created equal! J. Neurosci. Methods 171, 78–86 (2008). Saper, C.B. An open letter to our readers on the use of antibodies. J. Comp. Neurol. 493, 477–478 (2005). Paschke, M. Phage display systems and their applications. Appl. Microbiol. Biotechnol. 70, 2–11 (2006). Zahnd, C., Amstutz, P. & Pluckthun, A. Ribosome display: selecting and evolving proteins in vitro that specifically bind to a target. Nat. Methods 4, 269–279 (2007). Chao, G. et al. Isolating and engineering human antibodies using yeast surface display. Nat. Protoc. 1, 755–768 (2006). Binz, H.K., Amstutz, P. & Pluckthun, A. Engineering novel binding proteins from nonimmunoglobulin domains. Nat. Biotechnol. 23, 1257–1268 (2005). Binz, H.K. & Pluckthun, A. Engineered proteins as specific binding reagents. Curr. Opin. Biotechnol. 16, 459–469 (2005). Skerra, A. Alternative non-antibody scaffolds for molecular recognition. Curr. Opin. Biotechnol. 18, 295–304 (2007). Bradbury, A.R. & Marks, J.D. Antibodies from phage antibody libraries. J. Immunol. Methods 290, 29–49 (2004). Throsby, M. et al. Heterosubtypic neutralizing monoclonal antibodies crossprotective against H5N1 and H1N1 recovered from human IgM+ memory B cells. PLoS ONE 3, e3942 (2008). Razai, A. et al. Molecular evolution of antibody affinity for sensitive detection of botulinum neurotoxin type A. J. Mol. Biol. 351, 158–169 (2005). Lee, C.V. et al. High-affinity human antibodies from phage-displayed synthetic Fab libraries with a single framework scaffold. J. Mol. Biol. 340, 1073–1093 (2004). Hanes, J., Schaffitzel, C., Knappik, A. & Pluckthun, A. Picomolar affinity antibodies from a fully synthetic naive library selected and evolved by ribosome display. Nat. Biotechnol. 18, 1287–1292 (2000). Schier, R. et al. Isolation of picomolar affinity anti-c-erbB-2 single-chain Fv by molecular evolution of the complementarity determining regions in the center of the antibody binding site. J. Mol. Biol. 263, 551–567 (1996). Yang, W.P. et al. CDR walking mutagenesis for the affinity maturation of a potent human anti-HIV-1 antibody into the picomolar range. J. Mol. Biol. 254, 392–403 (1995). Boder, E.T., Midelfort, K.S. & Wittrup, K.D. Directed evolution of antibody fragments with monovalent femtomolar antigen-binding affinity. Proc. Natl. Acad. Sci. USA 97, 10701–10705 (2000). Foote, J. & Eisen, H.N. Breaking the affinity ceiling for antibodies and T cell receptors. Proc. Natl. Acad. Sci. USA 97, 10679–10681 (2000). Batista, F.D. & Neuberger, M.S. Affinity dependence of the B cell response to antigen: a threshold, a ceiling, and the importance of off-rate. Immunity 8, 751–759 (1998).

39. Schofield, D.J. et al. Application of phage display to high throughput antibody generation and characterization. Genome Biol. 8, R254 (2007). 40. Dübel, S., Stoevesandt, O., Taussig, M.J. & Hust, M. Generating recombinant antibodies to the complete human proteome. Trends Biotechnol. 28, 333–339 (2010). 41. Koide, A., Bailey, C.W., Huang, X. & Koide, S. The fibronectin type III domain as a scaffold for novel binding proteins. J. Mol. Biol. 284, 1141–1151 (1998). 42. Philibert, P. et al. A focused antibody library for selecting scFvs expressed at high levels in the cytoplasm. BMC Biotechnol. 7, 81 (2007). 43. Parsons, H.L. et al. Directing phage selections towards specific epitopes. Protein Eng. 9, 1043–1049 (1996). 44. Lassen, K.S., Bradbury, A.R., Rehfeld, J.F. & Heegaard, N.H. Microscale characterization of the binding specificity and affinity of a monoclonal antisulfotyrosyl IgG antibody. Electrophoresis 29, 2557–2564 (2008). 45. Kehoe, J.W. et al. Using phage display to select antibodies recognizing posttranslational modifications independently of sequence context. Mol. Cell. Proteomics 5, 2350–2363 (2006). 46. Hoffhines, A.J., Damoc, E., Bridges, K.G., Leary, J.A. & Moore, K.L. Detection and purification of tyrosine-sulfated proteins using a novel anti-sulfotyrosine monoclonal antibody. J. Biol. Chem. 281, 37877–37887 (2006). 47. Grunewald, J. et al. Mechanistic studies of the immunochemical termination of self-tolerance with unnatural amino acids. Proc. Natl. Acad. Sci. USA 106, 4337–4342 (2009). 48. Dalum, I. et al. Therapeutic antibodies elicited by immunization against TNFalpha. Nat. Biotechnol. 17, 666–669 (1999). 49. Hust, M. et al. A human scFv antibody generation pipeline for proteome research. J. Biotechnol. published online, doi:10.1016/j.jbiotec.2010.09.945 (29 September 2010). 50. Lloyd, C. et al. Modelling the human immune response: performance of a 1011 human antibody repertoire against a broad panel of therapeutically relevant antigens. Protein Eng. Des. Sel. 22, 159–168 (2009). 51. Wright, K., Collins, D.C. & Preedy, J.R. Comparative specificity of antisera raised against estrone, estradiol-17 and estriol using 6–0-carboxy-methyloxime bovine serum albumin derivatives. Steroids 21, 755–769 (1973). 52. Haning, R. et al. The evolution of titer and specificity of aldosterone binding antibodies in hyperimmunized sheep. Steroids 20, 73–88 (1972). 53. Exley, D., Johnson, M.W. & Dean, P.D. Antisera highly specific for 17-oestradiol. Steroids 18, 605–620 (1971). 54. Tateishi, K., Hamaoka, T., Takatsu, K. & Hayashi, C. A novel immunization procedure for production of anti-testosterone and anti-5 alpha-dihydrotestosterone antisera of low cross-reactivity. J. Steroid Biochem. 13, 951–959 (1980). 55. Smith, T.W. & Skubitz, K.M. Kinetics in interactions between antibodies and haptens. Biochemistry 14, 1496–1502 (1975). 56. Monigatti, F., Gasteiger, E., Bairoch, A. & Jung, E. The Sulfinator: predicting tyrosine sulfation sites in protein sequences. Bioinformatics 18, 769–770 (2002). 57. Sako, D. et al. A sulfated peptide segment at the amino terminus of PSGL-1 is critical for P-selectin binding. Cell 83, 323–331 (1995). 58. Rigby, P.W., Gething, M.J. & Hartley, B.S. Construction of intergeneric hybrids using bacteriophage P1CM: transfer of the Klebsiella aerogenes ribitol dehydrogenase gene to Escherichia coli. J. Bacteriol. 125, 728–738 (1976). 59. Ayriss, J., Woods, T., Bradbury, A. & Pavlik, P. High-throughput screening of single-chain antibodies using multiplexed flow cytometry. J. Proteome Res. 6, 1072–1082 (2007). 60. Pershad, K. et al. Generating a panel of highly specific antibodies to 20 human SH2 domains by phage display. Protein Eng. Des. Sel. 23, 279–288 (2010). 61. Mersmann, M. et al. Towards proteome scale antibody selections using phage display. New Biotechnol. 27, 118–128 (2009). 62. Velappan, N. et al. Selection and characterization of scFv antibodies against the Sin Nombre hantavirus nucleocapsid protein. J. Immunol. Methods 321, 60–69 (2007). 63. Cabezas, S. et al. Phage-displayed antibody fragments recognizing dengue 3 and dengue 4 viruses as tools for viral serotyping in sera from infected individuals. Arch. Virol. 154, 1035–1045 (2009). 64. Lim, A.P. et al. Neutralizing human monoclonal antibody against H5N1 influenza HA selected from a Fab-phage display library. Virol. J. 5, 130 (2008). 65. Okada, J. et al. Monoclonal antibodies in man that neutralized H3N2 influenza viruses were classified into three groups with distinct strain specificity: 1968– 1973, 1977–1993 and 1997–2003. Virology 397, 322–330 (2010). 66. Meissner, F. et al. Detection of antibodies against the four subtypes of Ebola virus in sera from any species using a novel antibody-phage indicator assay. Virology 300, 236–243 (2002). 67. Kirsch, M.I. et al. Development of human antibody fragments using antibody phage display for the detection and diagnosis of Venezuelan equine encephalitis virus (VEEV). BMC Biotechnol. 8, 66 (2008). 68. Maynard, J.A. et al. Protection against anthrax toxin by recombinant antibody fragments correlates with antigen affinity. Nat. Biotechnol. 20, 597–601 (2002). 69. Mabry, R. et al. Passive protection against anthrax by using a high-affinity antitoxin antibody fragment lacking an Fc region. Infect. Immun. 73, 8362–8368 (2005). 70. Wild, M.A. et al. Human antibodies from immunized donors are protective against anthrax toxin in vivo. Nat. Biotechnol. 21, 1305–1306 (2003).

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

perspective 71. Hayhurst, A. et al. Isolation and expression of recombinant antibody fragments to the biological warfare pathogen Brucella melitensis. J. Immunol. Methods 276, 185–196 (2003). 72. Zou, N., Newsome, T., Li, B., Tsai, S. & Lo, S.C. Human single-chain Fv antibodies against Burkholderia mallei and Burkholderia pseudomallei. Exp. Biol. Med. 232, 550–556 (2007). 73. Steiniger, S.C., Altobell, L.J. III, Zhou, B. & Janda, K.D. Selection of human antibodies against cell surface-associated oligomeric anthrax protective antigen. Mol. Immunol. 44, 2749–2755 (2007). 74. Cirino, N.M. et al. Disruption of anthrax toxin binding with the use of human antibodies and competitive inhibitors. Infect. Immun. 67, 2957–2963 (1999). 75. Zhou, B., Wirsching, P. & Janda, K.D. Human antibodies against spores of the genus Bacillus: a model study for detection of and protection against anthrax and the bioterrorist threat. Proc. Natl. Acad. Sci. USA 99, 5241–5246 (2002). 76. Garcia-Rodriguez, C. et al. Molecular evolution of antibody cross-reactivity for two subtypes of type A botulinum neurotoxin. Nat. Biotechnol. 25, 107–116 (2007). 77. Huie, M.A. et al. Antibodies to human fetal erythroid cells from a nonimmune phage antibody library. Proc. Natl. Acad. Sci. USA 98, 2682–2687 (2001). 78. Nizak, C. et al. Recombinant antibodies to the small GTPase Rab6 as conformation sensors. Science 300, 984–987 (2003). 79. Gao, J., Sidhu, S.S. & Wells, J.A. Two-state selection of conformation-specific antibodies. Proc. Natl. Acad. Sci. USA 106, 3071–3076 (2009). 80. Eisenhardt, S.U., Schwarz, M., Bassler, N. & Peter, K. Subtractive single-chain antibody (scFv) phage-display: tailoring phage-display for high specificity against function-specific conformations of cell membrane molecules. Nat. Protoc. 2, 3063–3073 (2007). 81. Ye, J.D. et al. Synthetic antibodies for specific recognition and crystallization of structured RNA. Proc. Natl. Acad. Sci. USA 105, 82–87 (2008). 82. Edwards, B.M. et al. The remarkable flexibility of the human antibody repertoire; isolation of over one thousand different antibodies to a single protein, BLyS. J. Mol. Biol. 334, 103–118 (2003). 83. Baker, K.P. et al. Generation and characterization of LymphoStat-B, a human monoclonal antibody that antagonizes the bioactivities of B lymphocyte stimulator. Arthritis Rheum. 48, 3253–3265 (2003). 84. Runnels, H.A. et al. Human monoclonal antibodies to the insulin-like growth factor 1 receptor inhibit receptor activation and tumor growth in preclinical studies. Adv. Ther. 27, 458–475 (2010). 85. Peipp, M., Dechant, M. & Valerius, T. Effector mechanisms of therapeutic antibodies against ErbB receptors. Curr. Opin. Immunol. 20, 436–443 (2008). 86. Li, K. et al. Modulation of Notch signaling by antibodies specific for the extracellular negative regulatory region of NOTCH3. J. Biol. Chem. 283, 8046–8054 (2008). 87. Wu, Y. et al. Therapeutic antibody targeting of individual Notch receptors. Nature 464, 1052–1057 (2010). 88. Martens, T. et al. A novel one-armed anti-c-Met antibody inhibits glioblastoma growth in vivo. Clin. Cancer Res. 12, 6144–6152 (2006). 89. Dobson, C.L. et al. Human monomeric antibody fragments to TRAIL-R1 and TRAIL-R2 that display potent in vitro agonism. MAbs 1, 552–562 (2009). 90. Sui, J. et al. Structural and functional bases for broad-spectrum neutralization of avian and human influenza A viruses. Nat. Struct. Mol. Biol. 16, 265–273 (2009). 91. Sun, L. et al. Generation, characterization and epitope mapping of two neutralizing and protective human recombinant antibodies against influenza A H5N1 viruses. PLoS ONE 4, e5476 (2009). 92. Poul, M.A., Becerril, B., Nielsen, U.B., Morisson, P. & Marks, J.D. Selection of tumor-specific internalizing human antibodies from phage libraries. J. Mol. Biol. 301, 1149–1161 (2000). 93. Heitner, T. et al. Selection of cell binding and internalizing epidermal growth factor receptor antibodies from a phage display library. J. Immunol. Methods 248, 17–30 (2001). 94. Zhou, Y., Zou, H., Zhang, S. & Marks, J.D. Internalizing cancer antibodies from phage libraries selected on tumor cells and yeast-displayed tumor antigens. J. Mol. Biol. 404, 88–99 (2010). 95. Park, J.W. et al. Tumor targeting using anti-her2 immunoliposomes. J. Control. Release 74, 95–113 (2001). 96. Nielsen, U.B. et al. Therapeutic efficacy of anti-ErbB2 immunoliposomes targeted by a phage antibody selected for cellular endocytosis. Biochim. Biophys. Acta 1591, 109–118 (2002). 97. Wu, H. et al. Stepwise in vitro affinity maturation of Vitaxin, an alphav beta3specific humanized mAb. Proc. Natl. Acad. Sci. USA 95, 6037–6042 (1998). 98. Lippow, S.M., Wittrup, K.D. & Tidor, B. Computational design of antibody-affinity improvement beyond in vivo maturation. Nat. Biotechnol. 25, 1171–1176 (2007). 99. Fellouse, F.A., Wiesmann, C. & Sidhu, S.S. Synthetic antibodies from a fouramino-acid code: a dominant role for tyrosine in antigen recognition. Proc. Natl. Acad. Sci. USA 101, 12467–12472 (2004). 100. 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). 101. Lee, C.V. et al. Synthetic anti-BR3 antibodies that mimic BAFF binding and target both human and murine B cells. Blood 108, 3103–3111 (2006).

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

102. Fagete, S. et al. Specificity tuning of antibody fragments to neutralize two human chemokines with a single agent. MAbs 1, 288–296 (2009). 103. Bostrom, J. et al. Variants of the antibody Herceptin that interact with HER2 and VEGF at the antigen binding site. Science 323, 1610–1614 (2009). 104. Volk, W.A., Bizzini, B., Snyder, R.M., Bernhard, E. & Wagner, R.R. Neutralization of tetanus toxin by distinct monoclonal antibodies binding to multiple epitopes on the toxin molecule. Infect. Immun. 45, 604–609 (1984). 105. Marks, J.D. Deciphering antibody properties that lead to potent botulinum neurotoxin neutralization. Mov. Disord. 19 Suppl 8, S101–S108 (2004). 106. Nowakowski, A. et al. Potent neutralization of botulinum neurotoxin by recombinant oligoclonal antibody. Proc. Natl. Acad. Sci. USA 99, 11346–11350 (2002). 107. Kalb, S.R. et al. Extraction of BoNT/A, /B, /E, and /F with a single, high affinity monoclonal antibody for detection of botulinum neurotoxin by Endopep-MS. PLoS ONE 5, e12237 (2010). 108. Garcia-Rodriguez, C. et al. Neutralizing human monoclonal antibodies binding multiple serotypes of botulinum neurotoxin. Protein Eng. Des. Sel. published online, doi:10.1093/protein/gzq111 (12 December 2010). 109. de Kruif, J. & Logtenberg, T. Leucine zipper dimerized bivalent and bispecific scFv antibodies from a semi-synthetic antibody phage display library. J. Biol. Chem. 271, 7630–7634 (1996). 110. Hudson, P.J. & Kortt, A.A. High avidity scFv multimers; diabodies and triabodies. J. Immunol. Methods 231, 177–189 (1999). 111. Dübel, S. et al. Bifunctional and multimeric complexes of streptavidin fused to single chain antibodies (scFv). J. Immunol. Methods 178, 201–209 (1995). 112. Griep, R.A. et al. pSKAP/S: An expression vector for the production of single-chain Fv alkaline phosphatase fusion proteins. Protein Expr. Purif. 16, 63–69 (1999). 113. Cloutier, S.M. et al. Streptabody, a high avidity molecule made by tetramerization of in vivo biotinylated, phage display-selected scFv fragments on streptavidin. Mol. Immunol. 37, 1067–1077 (2000). 114. Casey, J.L., Coley, A.M., Tilley, L.M. & Foley, M. Green fluorescent antibodies: novel in vitro tools. Protein Eng. 13, 445–452 (2000). 115. Thie, H., Binius, S., Schirrmann, T., Hust, M. & Dübel, S. Multimerization domains for antibody phage display and antibody production. New Biotechnol. 26, 314–321 (2009). 116. Persic, L. et al. An integrated vector system for the eukaryotic expression of antibodies or their fragments after selection from phage display libraries. Gene 187, 9–18 (1997). 117. Hu, S. et al. Minibody: A novel engineered anti-carcinoembryonic antigen antibody fragment (single-chain Fv-CH3) which exhibits rapid, high-level targeting of xenografts. Cancer Res. 56, 3055–3061 (1996). 118. Beck, A. et al. Trends in glycosylation, glycoanalysis and glycoengineering of therapeutic antibodies and Fc-fusion proteins. Curr. Pharm. Biotechnol. 9, 482–501 (2008). 119. Presta, L.G. Molecular engineering and design of therapeutic antibodies. Curr. Opin. Immunol. 20, 460–470 (2008). 120. Merchant, A.M. et al. An efficient route to human bispecific IgG. Nat. Biotechnol. 16, 677–681 (1998). 121. Ridgway, J.B., Presta, L.G. & Carter, P. ‘Knobs-into-holes’ engineering of antibody CH3 domains for heavy chain heterodimerization. Protein Eng. 9, 617–621 (1996). 122. Perisic, O., Webb, P.A., Holliger, P., Winter, G. & Williams, R.L. Crystal structure of a diabody, a bivalent antibody fragment. Structure 2, 1217–1226 (1994). 123. Atwell, J.L. et al. scFv multimers of the anti-neuraminidase antibody NC10: length of the linker between VH and VL domains dictates precisely the transition between diabodies and triabodies. Protein Eng. 12, 597–604 (1999). 124. Pei, X.Y., Holliger, P., Murzin, A.G. & Williams, R.L. The 2.0-A resolution crystal structure of a trimeric antibody fragment with noncognate VH-VL domain pairs shows a rearrangement of VH CDR3. Proc. Natl. Acad. Sci. USA 94, 9637–9642 (1997). 125. Le Gall, F., Kipriyanov, S.M., Moldenhauer, G. & Little, M. Di-, tri- and tetrameric single chain Fv antibody fragments against human CD19: effect of valency on cell binding. FEBS Lett. 453, 164–168 (1999). 126. Muller, D. & Kontermann, R.E. Bispecific antibodies for cancer immunotherapy: Current perspectives. BioDrugs 24, 89–98 (2010). 127. Xiang, J. Targeting cytokines to tumors to induce active antitumor immune responses by recombinant fusion proteins. Hum. Antibodies 9, 23–36 (1999). 128. Schliemann, C. & Neri, D. Antibody-based targeting of the tumor vasculature. Biochim. Biophys. Acta 1776, 175–192 (2007). 129. Deckert, P.M. Current constructs and targets in clinical development for antibodybased cancer therapy. Curr. Drug Targets 10, 158–175 (2009). 130. Fuchs, H. & Bachran, C. Targeted tumor therapies at a glance. Curr. Drug Targets 10, 89–93 (2009). 131. Gawlitta, W., Osborn, M. & Weber, K. Coiling of intermediate filaments induced by microinjection of a vimentin-specific antibody does not interfere with locomotion and mitosis. Eur. J. Cell Biol. 26, 83–90 (1981). 132. Kontermann, R.E. Intrabodies as therapeutic agents. Methods 34, 163–170 (2004). 133. Beerli, R.R., Wels, W. & Hynes, N.E. Intracellular expression of single chain antibodies reverts ErbB-2 transformation. J. Biol. Chem. 269, 23931–23936 (1994). 134. Richardson, J.H., Sodroski, J.G., Waldmann, T.A. & Marasco, W.A. Phenotypic knockout of the high-affinity human interleukin 2 receptor by intracellular singlechain antibodies against the alpha subunit of the receptor. Proc. Natl. Acad. Sci. USA 92, 3137–3141 (1995).

253

© 2011 Nature America, Inc. All rights reserved.

Perspective 135. Paganetti, P., Calanca, V., Galli, C., Stefani, M. & Molinari, M. beta-site specific intrabodies to decrease and prevent generation of Alzheimer’s Abeta peptide. J. Cell Biol. 168, 863–868 (2005). 136. Strebe, N. et al. Functional knockdown of VCAM-1 at the posttranslational level with ER retained antibodies. J. Immunol. Methods 341, 30–40 (2009). 137. Biocca, S. & Cattaneo, A. Intracellular immunization: antibody targeting to subcellular compartments. Trends Cell Biol. 5, 248–252 (1995). 138. Biocca, S., Pierandrei-Amaldi, P., Campioni, N. & Cattaneo, A. Intracellular immunization with cytosolic recombinant antibodies. Nat. Biotechnol. 12, 396–399 (1994). 139. Desiderio, A. et al. A semi-synthetic repertoire of intrinsically stable antibody fragments derived from a single-framework scaffold. J. Mol. Biol. 310, 603–615 (2001). 140. der Maur, A.A. et al. Direct in vivo screening of intrabody libraries constructed on a highly stable single-chain framework. J. Biol. Chem. 277, 45075–45085 (2002). 141. Tanaka, T., Chung, G.T., Forster, A., Lobato, M.N. & Rabbitts, T.H. De novo production of diverse intracellular antibody libraries. Nucleic Acids Res. 31, e23 (2003). 142. Auf der Maur, A., Escher, D. & Barberis, A. Antigen-independent selection of stable intracellular single-chain antibodies. FEBS Lett. 508, 407–412 (2001). 143. Visintin, M., Tse, E., Axelson, H., Rabbitts, T.H. & Cattaneo, A. Selection of antibodies for intracellular function using a two-hybrid in vivo system. Proc. Natl. Acad. Sci. USA 96, 11723–11728 (1999). 144. Amstutz, P. et al. Intracellular kinase inhibitors selected from combinatorial libraries of designed ankyrin repeat proteins. J. Biol. Chem. 280, 24715–24722 (2005). 145. Kohl, A. et al. Allosteric inhibition of aminoglycoside phosphotransferase by a designed ankyrin repeat protein. Structure 13, 1131–1141 (2005). 146. Rizk, S.S. et al. An engineered substance P variant for receptor-mediated delivery of synthetic antibodies into tumor cells. Proc. Natl. Acad. Sci. USA 106, 11011–11015 (2009). 147. Hallborn, J. & Carlsson, R. Automated screening procedure for high-throughput generation of antibody fragments. Biotechniques Suppl, 30–37 (2002). 148. Turunen, L., Takkinen, K., Soderlund, H. & Pulli, T. Automated panning and screening procedure on microplates for antibody generation from phage display libraries. J. Biomol. Screen. 14, 282–293 (2009). 149. Lou, J. et al. Antibodies in haystacks: how selection strategy influences the outcome of selection from molecular diversity libraries. J. Immunol. Methods 253, 233–242 (2001). 150. Storz, U. IP issues in the therapeutic antibody industry. in Antibody Engineering (eds. Kontermann, R. & Dübel, S.) 517–581 (Springer, 2010). 151. Kreitman, R.J. et al. Phase I trial of recombinant immunotoxin anti-Tac(Fv)-PE38 (LMB-2) in patients with hematologic malignancies. J. Clin. Oncol. 18, 1622–1636 (2000). 152. Fellouse, F.A. et al. High-throughput generation of synthetic antibodies from highly functional minimalist phage-displayed libraries. J. Mol. Biol. 373, 924–940 (2007). 153. Hamilton, S.R. & Gerngross, T.U. Glycosylation engineering in yeast: the advent of fully humanized yeast. Curr. Opin. Biotechnol. 18, 387–392 (2007). 154. Moghaddam, A. et al. Identification of scFv antibody fragments that specifically recognise the heroin metabolite 6-monoacetylmorphine but not morphine. J. Immunol. Methods 280, 139–155 (2003). 155. Dorsam, H. et al. Antibodies to steroids from a small human naive IgM library. FEBS Lett. 414, 7–13 (1997). 156. Pope, A. et al. In vitro selection of a high affinity antibody to oestradiol using a phage display human antibody library. Immunotechnology 2, 209–217 (1996). 157. Hemminki, A., Niemi, S., Hautoniemi, L., Soderlund, H. & Takkinen, K. Fine tuning of an anti-testosterone antibody binding site by stepwise optimisation of the CDRs. Immunotechnology 4, 59–69 (1998). 158. Saviranta, P. et al. Engineering the steroid-specificity of an anti-17beta-estradiol Fab by random mutagenesis and competitive phage panning. Protein Eng. 11, 143–152 (1998).

254

159. Chames, P., Coulon, S. & Baty, D. Improving the affinity and the fine specificity of an anti-cortisol antibody by parsimonious mutagenesis and phage display. J. Immunol. 161, 5421–5429 (1998). 160. Bikker, F.J., Mars-Groenendijk, R.H., Noort, D., Fidder, A. & van der Schans, G.P. Detection of sulfur mustard adducts in human callus by phage antibodies. Chem. Biol. Drug Des. 69, 314–320 (2007). 161. Szent-Gyorgyi, C. et al. Fluorogen-activating single-chain antibodies for imaging cell surface proteins. Nat. Biotechnol. 26, 235–240 (2008). 162. Watanabe, H., Nakanishi, T., Umetsu, M. & Kumagai, I. Human anti-gold antibodies: biofunctionalization of gold nanoparticles and surfaces with anti-gold antibodies. J. Biol. Chem. 283, 36031–36038 (2008). 163. Dadaglio, G., Nelson, C.A., Deck, M.B., Petzold, S.J. & Unanue, E.R. Characterization and quantitation of peptide-MHC complexes produced from hen egg lysozyme using a monoclonal antibody. Immunity 6, 727–738 (1997). 164. Krogsgaard, M. et al. Visualization of myelin basic protein (MBP) T cell epitopes in multiple sclerosis lesions using a monoclonal antibody specific for the human histocompatibility leukocyte antigen (HLA)-DR2-MBP 85–99 complex. J. Exp. Med. 191, 1395–1412 (2000). 165. Mutuberria, R. et al. Isolation of human antibodies to tumor-associated endothelial cell markers by in vitro human endothelial cell selection with phage display libraries. J. Immunol. Methods 287, 31–47 (2004). 166. Cohen, C.J., Denkberg, G., Lev, A., Epel, M. & Reiter, Y. Recombinant antibodies with MHC-restricted, peptide-specific, T-cell receptor-like specificity: new tools to study antigen presentation and TCR-peptide-MHC interactions. J. Mol. Recognit. 16, 324–332 (2003). 167. Engberg, J., Krogsgaard, M. & Fugger, L. Recombinant antibodies with the antigenspecific, MHC restricted specificity of T cells: novel reagents for basic and clinical investigations and immunotherapy. Immunotechnology 4, 273–278 (1999). 168. Stryhn, A. et al. Shared fine specificity between T-cell receptors and an antibody recognizing a peptide/major histocompatibility class I complex. Proc. Natl. Acad. Sci. USA 93, 10338–10342 (1996). 169. Villa, A. et al. A high-affinity human monoclonal antibody specific to the alternatively spliced EDA domain of fibronectin efficiently targets tumor neovasculature in vivo. Int. J. Cancer 122, 2405–2413 (2008). 170. Pini, A. et al. Design and use of a phage display library. Human antibodies with subnanomolar affinity against a marker of angiogenesis eluted from a twodimensional gel. J. Biol. Chem. 273, 21769–21776 (1998). 171. Schliemann, C. & Neri, D. Antibody-based vascular tumor targeting. Recent Results Cancer Res. 180, 201–216 (2010). 172. Rothlisberger, D., Pos, K.M. & Pluckthun, A. An antibody library for stabilizing and crystallizing membrane proteins–selecting binders to the citrate carrier CitS. FEBS Lett. 564, 340–348 (2004). 173. Uysal, S. et al. Crystal structure of full-length KcsA in its closed conformation. Proc. Natl. Acad. Sci. USA 106, 6644–6649 (2009). 174. Osbourn, J., Groves, M. & Vaughan, T. From rodent reagents to human therapeutics using antibody guided selection. Methods 36, 61–68 (2005). 175. Jespers, L.S., Roberts, A., Mahler, S.M., Winter, G. & Hoogenboom, H.R. Guiding the selection of human antibodies from phage display repertoires to a single epitope of an antigen. Bio/Technology 12, 899–903 (1994). 176. Xie, M.H., Yuan, J., Adams, C. & Gurney, A. Direct demonstration of MuSK involvement in acetylcholine receptor clustering through identification of agonist ScFv. Nat. Biotechnol. 15, 768–771 (1997). 177. Ellmark, P., Andersson, H., Abayneh, S., Fenyo, E.M. & Borrebaeck, C.A. Identification of a strongly activating human anti-CD40 antibody that suppresses HIV type 1 infection. AIDS Res. Hum. Retroviruses 24, 367–373 (2008). 178. Roth, A. et al. Anti-CD166 single chain antibody-mediated intracellular delivery of liposomal drugs to prostate cancer cells. Mol. Cancer Ther. 6, 2737–2746 (2007). 179. Liu, B. et al. Recombinant full-length human IgG1s targeting hormone-refractory prostate cancer. J. Mol. Med. 85, 1113–1123 (2007).

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

Articles

Chemoproteomics profiling of HDAC inhibitors reveals selective targeting of HDAC complexes

© 2011 Nature America, Inc. All rights reserved.

Marcus Bantscheff1,3, Carsten Hopf1,3, Mikhail M Savitski1, Antje Dittmann1, Paola Grandi1, Anne-Marie Michon1, Judith Schlegl1, Yann Abraham1, Isabelle Becher1, Giovanna Bergamini1, Markus Boesche1, Manja Delling1, Birgit Dümpelfeld1, Dirk Eberhard1, Carola Huthmacher1, Toby Mathieson1, Daniel Poeckel1, Valérie Reader2, Katja Strunk1, Gavain Sweetman1, Ulrich Kruse1, Gitte Neubauer1, Nigel G Ramsden2 & Gerard Drewes1 The development of selective histone deacetylase (HDAC) inhibitors with anti-cancer and anti-inflammatory properties remains challenging in large part owing to the difficulty of probing the interaction of small molecules with megadalton protein complexes. A combination of affinity capture and quantitative mass spectrometry revealed the selectivity with which 16 HDAC inhibitors target multiple HDAC complexes scaffolded by ELM-SANT domain subunits, including a novel mitotic deacetylase complex (MiDAC). Inhibitors clustered according to their target profiles with stronger binding of aminobenzamides to the HDAC NCoR complex than to the HDAC Sin3 complex. We identified several non-HDAC targets for hydroxamate inhibitors. HDAC inhibitors with distinct profiles have correspondingly different effects on downstream targets. We also identified the anti-inflammatory drug bufexamac as a class IIb (HDAC6, HDAC10) HDAC inhibitor. Our approach enables the discovery of novel targets and inhibitors and suggests that the selectivity of HDAC inhibitors should be evaluated in the context of HDAC complexes and not purified catalytic subunits. Protein lysine acetylation is a key mechanism in the epigenetic control of gene expression and the regulation of cell metabolism1–3, and protein deacetylases are potential targets for treating cancer and a range of autoimmune and neurodegenerative diseases4. The first mammalian histone deacetylase (HDAC) was discovered in 1996 by a chemical biology approach using an immobilized, microbially derived compound as affinity matrix5. Based on sequence phylogeny and function, there are four distinct classes of HDAC: class I (HDAC1, 2, 3 and 8), class IIa (HDAC4, 5, 7 and 9), class IIb (HDAC6 and 10) and class IV (HDAC11) represent Zn2+-dependent amidohydrolases, whereas class III comprises the mechanistically diverse NAD+-dependent sirtuins6. HDACs form the catalytic core of megadalton complexes involved in chromatin modification and gene repression. Four such molecular machines have been characterized to date. Whereas the CoREST, NuRD and Sin3 complexes contain an HDAC1-HDAC2 dimer as core, the NCoR complex is formed around HDAC3 (ref. 7). The roles of these complexes are diverse and often cell-type specific. Although more data are emerging regarding their role in the determination of cell fate, their functions in tissue homeostasis are less well understood8. The CoREST complex couples histone deacetylation to demethylation to repress neuronal genes9, the NuRD complex links deacetylation to a chromatin-remodeling ATPase and promotes gene silencing10, and the Sin3 complex represses genes downstream of various developmental pathways11. The NCoR complex is the major corepressor for nuclear receptors12,13. Class IIa HDACs exhibit low enzymatic

activity and are proposed to have “modification reader” or s caffold f unctions14,15. Class IIb HDACs exhibit mostly nonepigenetic functions in regulating protein folding and turnover16. Small-molecule HDAC inhibitors were discovered by their ability to induce redifferentiation of transformed cells17. Suberoylanilide hydroxamic acid (SAHA; vorinostat, Zolinza) and romidepsin (Istodax) are approved for the treatment of cutaneous T-cell lymphoma, and valproate is in clinical use as an anticonvulsant. Several HDAC inhibitors are in development for a number of indications but clinical development has been hampered by a lack of target selectivity. This increases the risk of toxic liabilities and also limits the use of these compounds as research tools18. The perceived lack of selectivity of HDAC inhibitors may originate from the optimization of lead compounds using standard industry assays involving recombinant enzymes or protein fragments. These seem unlikely to properly reflect the native conformation and activity of the target and its physiological context owing to incorrect protein folding, post-translational modifications and the absence of regulatory subunits. Remarkably, purified class I HDACs exhibit increased activity in the presence of interacting proteins13,19. Most HDAC inhibitors adhere to a distinctive pharmacophore comprising a ‘cap’, which binds to the rim of the substrate channel, a spacer spanning the channel, and a Zn2+-chelating function. A photoaffinity analog of SAHA was shown to label not only HDACs but also the proteins RCOR1, MBD3 and MTA1/2. This indicates that these proteins are close to the active site, and suggests that the cap conveys inhibitor selectivity20.

1Cellzome AG, Heidelberg, Germany. 2Cellzome Ltd., Chesterford Research Park, Cambridge, United Kingdom. 3These authors contributed equally to this work. Correspondence should be addressed to M.B. ([email protected]) or G.D. ([email protected]).

Received 18 November 2010; accepted 17 December 2010; published online 23 January 2011; doi:10.1038/nbt.1759

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

255

Articles consist of a moiety that binds to a ligand pocket conserved within the target class under investigation, and a functional group for immobilization enabling efficient enrichment of bound proteins for analysis24. HDACs share a conserved substrate pocket, and most hydroxamate inhibitors are nonselective25. We synthesized a target class–specific probe matrix by derivatizing sepharose with analogs of the hydroxamates SAHA and givinostat (ITF2357). The probe matrix was exposed to cell extracts, and aliquots of the sample were treated with excess inhibitor, which competes with the immobilized probes for target protein binding. The reduction in protein capture that resulted from inhibitor treatment was quantified by isobaric tagging of tryptic peptides and tandem mass spectrometry analysis (MS/MS) of the combined peptide pools26. For each identified protein, the decrease of the reporter ion signals relative to the vehicle control reflects the competitive binding of the ‘free’ inhibitor to its target. The results comprise binding data for both direct enzyme targets and proteins residing in a complex with the target, as these are predicted to have matching IC50 profiles

Recent advances in chemoproteomics enabled binding studies of small-molecule enzyme inhibitors to endogenous proteins in cells and tissues21–23. Here we extend the chemoproteomics approach from the monitoring of individual target proteins to the analysis of inhibitors binding to native megadalton protein complexes, with a view to discovering novel targets, complexes and inhibitors. We found that HDAC inhibitors targeted known and novel protein complexes that are precisely defined by matching half-maximal inhibitory concentration (IC50) values for a given inhibitor for all complex subunits, and we used quantitative immuno affinity purifications to confirm the composition of the complexes. Inhibitor selectivity data for native drug target complexes deviated from literature values obtained using isolated recombinant enzymes, indicating an unexpected degree of selectivity of certain HDAC inhibitors.

a

1

O

Probe matrix

O

H N

2

Vehicle control

3

NH

O N H

O

OH

O

4 TMT 131

OH

N H

5

6

Protein does not bind to drug

b

100

50

IP of protein A

6

m/z

12

12

7 12 8 12 9 13 0 13 1

0

TMT 130

Protein A binds to target 100

Protein B binds to target 100

Protein A Protein B

50

TMT 129

50

7 12 8 12 9 13 0 13 1

m/z

log [inhibitor]

12

12

log [inhibitor]

m/z

0

6

12

6 12 7 12 8 12 9 13 0 13 1

0

Target binds to drug Target

100

TMT 128 50

log [inhibitor]

6 12 7 12 8 12 9 13 0 13 1

0 12

Inhibitor concentration

TMT 127

m/z

Protein C binds to target Protein D binds to target

100

100

Protein C Protein D

50 50 log [inhibitor]

6 12 7 12 8 12 9 13 0 13 1

9 13 0 13 1

12

8

7 12

12

log [inhibitor]

m/z

0

6

12

TMT 126

0

12

© 2011 Nature America, Inc. All rights reserved.

RESULTS Synthesis of a target class–specific HDAC probe matrix Target class–directed chemical probes provide tools for the identification of drug targets directly in cells and tissues. Typically, probes

m/z

IP of protein D

Figure 1 Mapping of HDAC drug target complexes in chemical space and in proteome space. (a) Chemoproteomics competition binding assay to profile HDAC inhibitor target complexes in cell extract. (1) A probe matrix is generated by derivatizing sepharose with analogs of nonselective HDAC inhibitors (left, SAHA, right, givinostat). (2) Cell extract is incubated with vehicle or with drug over a range of concentrations. (3) The ‘free’ drug competes with the immobilized probes for drug-binding sites on target-protein complexes. White hexagon, inhibitor drug. (4) Captured proteins are trypsinized and each peptide mixture is tagged with a distinct isobaric tandem mass tag (TMT). (5) Tagged samples are pooled and analyzed by LC-MS/MS. Each peptide gives rise to six reporter signals in the MS/MS spectrum. (6) When free drug outcompetes protein capture, signal intensities relative to the vehicle control decrease for each peptide originating from this protein. Complexes formed by the target and associated proteins are defined by matching inhibition (IC50) curves. (b) Definition of target protein complexes in biological space by quantitative co-IP. Data generated from the same cell extracts are used to deconvolute protein complexes formed around the drug target.

256

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

Articles a

NH2

Aminobenzamide

Tacedinaline (CI-994) O O

N H N H

NH2

NH

O

NH S

N H

O S

O O

O NH

10

0.01

0.1

1

10

0.01

0.1

1

10

100

99

0 1

Residual binding

© 2011 Nature America, Inc. All rights reserved.

O

1

1.0

10

100 1,000

1

10

100 1,000

1

10

100 1,000

1.0 95

1 0 1

Romidepsin

0.1

10

100 1,000

1

10

100 1,000

1.0

0

HDAC1 HDAC2 HDAC3 HDAC6 HDAC8 HDAC10

0.001 0.01

1

GATAD2A LSD1 MTA3 NCOR1 RCOR1 SIN3A TBL1XR1

0.1

1

[Compound] (µM)

0.001 0.01

0.1

1

10

100 1,000

BZW2 DNTTIP1 ISOC2 MIDEAS PDXK

0.001 0.01

[Compound] (µM)

0.1

1

[Compound] (µM)

0

PC I− 3 Va MC 40 lp −1 51 PCroic 293 R I−2 aci o d Ta mi 478 ce de 1 d ps En ina in tin line M BM ost oc L− at et 21 in Sc os 0 rip tat ta Be SA id lin HA Pa A os no pic tat b i D in di Tr ac os n ic in tat ho os st ta at t in A

N H

O

0 0.01

SAP30 SIN3A SAP30L SIN3B WIZ BRMS1 HDAC6 LOC153364 ISOC2 TBL1XR1 TBL1X NCOR2 NCOR1 GPS2 HDAC3 RCOR1 HMG20A LSD1 GSE1 RCOR3 HMG20B RCOR2 HDAC1 HDAC2 MIER3 MTA3 MTA1 MBD3 MIER2 PHF21A GATAD2B GATAD2A RBBP4 MTA2 ZMYM2 DNTTIP1 CDYL MIDEAS RREB1 MIER1 ZMYM3 RERE EHMT2 MBD2 CHD4 CDK2AP1

CoREST

O

H N

98

NCOR1

BML210

b

Other proteins

Relative potency

Hydroxamic acid

Residual binding

OH

Residual binding

N H

O

Complex subunits

1.0

SIN3

O

H N

Residual binding

HDACs

SAHA

Figure 2 HDAC inhibitor drug targets and target complexes are defined by chemoproteomics profiling of drugs and compounds used as research tools. (a) Representative concentration-inhibition profiles of SAHA, BML-210 (aminobenzamide analog of SAHA), tacedinaline and romidepsin were determined in K562 cell extract as outlined in Figure 1. Inhibitors were pre-incubated with cell extracts at 4 °C before addition of the probe matrix, with the exception of the aminobenzamides BML-210 and tacedinaline, which were pre-incubated at 22 °C. Profiles are grouped in three plots for each inhibitor: HDACs (left), components of CoREST, NuRD, Sin3 and NCoR complexes (middle), and examples of other proteins either representing novel direct targets or complex components (right). Previously known complex associations are represented in a color code. Profiles of additional inhibitors are depicted in Supplementary Figure 3. (b) Bidirectional hierarchical clustering of the concentration-inhibition data for 16 inhibitors versus 1,251 proteins (each targeted by at least one inhibitor). Only the area of the clustering around HDACs is shown. For better comparison of selectivities, average pKdapp values were transformed into relative affinities scaling from 0 to 1 for each inhibitor. Statistically significant clusters are highlighted in blue and brown representing >95% and >99% unbiased bootstrap probability, respectively (Supplementary Fig. 7).

across a set of drugs. We confirmed associations of proteins in complexes using immunoaffinity purifications (Fig. 1). Quantitative mapping of protein binding to the probe matrix We first tested the probe matrix with recombinant HDACs 1–11 purified from Sf9 insect cells. With the exception of HDAC1, the enzymes were found to bind to the matrix. However, binding was only partially reduced by excess SAHA, indicating that the bulk of the purified enzyme exhibited low activity (Supplementary Table 1). HDAC3, the only enzyme purified together with a cofactor (NCoR2), was most susceptible to competition with excess SAHA. The activity of the enzymes in a peptide deacetylation assay (Supplementary Fig. 1) may arise from a small fraction of properly folded protein or from contamination with insect-cell activities15,27. Consistent with the binding data, HDAC1 showed the lowest activity, whereas HDAC3, used as a complex with NCoR2, was most active. In contrast to the recombinant proteins, all class I and class IIb HDACs and many known HDAC complex subunits were specifically captured when the SAHA- or givinostat-derivatized matrix was used to probe endogenous HDACs in cell extracts from the myelogenous leukemia line K562 (Supplementary Fig. 2 and Supplementary Data Set 1). The ability to competitively inhibit the binding of a protein to the probe matrix with excess ‘free’ inhibitor distinguishes specific binding from nonspecific background. From the ~2,600 proteins identified, 267 proteins exhibited substantially reduced matrix nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

binding when excess SAHA or trichostatin A (TSA) was added to the lysate, with their corresponding reporter ion intensities reduced by 50% or more. These were designated as potential inhibitor targets, and included the six class I and class IIb HDACs, and 29 proteins known to associate with class I HDACs in the CoREST, NuRD, Sin3 and NCoR complexes. In addition, many proteins previously not associated with deacetylase function bound to the probe matrix specifically, implicating them either as novel inhibitor targets or as components of target complexes. For instance, 5,10-methenyltetrahydrofolate synthetase and hydroxysteroid (17-beta) dehydrogenase 4 were specifically captured by givinostat-conjugated, but not SAHA-conjugated, matrix. This suggests that they are targets of givinostat but not SAHA. Next, the subproteome binding to the SAHA matrix was differentially mapped in nuclear versus cytosolic fractions of Jurkat cells, and in a panel of human cell lines and mouse tissues (Supplementary Fig. 2 and Supplementary Data Set 1). Most proteins in the subproteome appeared to be ubiquitously expressed. As expected, the majority was enriched in the nuclear fraction. Because of the function of class I HDACs in cell division 28, we conducted differential mapping of the subproteome in HeLa cells arrested in mitosis or G1/S phase, compared to nonarrested cells. Notably, three proteins were specifically captured by the probe matrix in greater amounts from mitotic cells compared to nonmitotic cells, and thus may constitute a novel HDAC complex. These are DNTTIP1 (deoxynucleotidyltransferase-interacting protein), C14ORF43, 257

Articles a protein of unknown function, which we dubbed MIDEAS (for mitotic deacetylase-associated SANT domain protein), and the putative histone acetylase CDYL.

inhibitors at five concentrations, typically ranging from 40 nM to 10 µM. For low- or very high-potency compounds, concentrations were adjusted accordingly. Subsequently, samples were incubated with the probe matrix, captured proteins were quantitatively mapped by MS/MS and a set of IC50 values was determined for each inhibitor (Fig. 2a and Supplementary Fig. 3). To assess reproducibility, we carried out several replicate experiments per inhibitor profile using targeted data acquisition29 (Supplementary Data Set 2). The data were sufficiently reproducible to discriminate fairly small (twofold) differences in IC50 values. Remarkably, we found statistically significant differences between distinct complexes containing

Proteomics target profiling of HDAC inhibitors The probe matrix offers a unique tool to probe a subproteome of putative targets with drugs under close to physiological conditions. We selected a set of 16 structurally diverse HDAC inhibitors, including approved drugs, compounds in clinical development and small molecules with potential as research tools (Supplementary Table 2). Aliquots of K562 cell extract were incubated with vehicle or different

Table 1 Kdapp values (in mM) for selected HDAC inhibitors of molecular targets and target complexes, as determined using chemoproteomic binding profiling

© 2011 Nature America, Inc. All rights reserved.

Romidepsin Tacedinaline BML-210 SAHA preincub. Trichostatin A Valproic acid Belinostat reincub. at 4 °C preincub. at 22 °C preincub. at 4 °C at 4 °C max conc. preincub. at 22 °C preincub. at 4 °C preincub. at 4 °C p 10 µM max conc. 30 µM max conc. 300 µM max conc. 100 nM max conc. 600 µM max conc. 1 µM max conc. 20 mM Classification

Protein

Kdappa

s.d.b

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Kdapp

s.d.

Class I HDAC

HDAC1 HDAC2 HDAC3 HDAC8

0.29 0.35 0.26 >30c

0.02 0.04 0.08 –

14.9 10.7 2.8 > 300

1.0 0.2 0.4 –

0.008 0.010 0.017 > 0.1

0.001 0.001 0.001 –

0.28 0.35 0.29 > 10

0.05 0.08 0.06 –

13.0 11.9 6.0 240

1.8 1.4 0.3 137

0.010 0.012 0.041 >1

0.001 0.002 0.007 –

745 975 9,350 12,675

147 249 3949 734

Class II HDAC

HDAC6 HDAC10

0.53 >30

0.14 –

> 300 > 300

– –

> 0.1 > 0.1

– –

0.13 5.26

0.04 2.34

> 600 > 600

– –

0.109 >1

0.025 >20,000 – >20,000

CoREST complex

GSE1 LSD1 HMG20A HMG20B PHF21A RCOR1 RCOR2 RCOR3 ZMYM2 ZMYM3

0.29 0.29 0.31 0.33 0.25 0.20 0.21 0.26 0.36 0.39

0.09 0.07 0.09 0.10 0.04 0.03 0.06 0.08 0.06 0.24

9.2 9.2 9.8 12.6 31.8 8.5 18.2 10.5 43.3 134

0.9 0.5 0.4 0.5 1.0 0.5 5.5 0.3 32.2 90.9

0.006 0.006 0.005 0.006 0.012 0.005 0.008 0.005 0.013 0.035

0.001 0.001 0.001 0.001 – 0.001 0.000 0.001 0.005 0.010

0.22 0.17 0.22 0.25 0.27 0.24 0.13 0.21 0.22 0.58

0.06 0.05 0.04 0.04 0.05 0.06 0.02 0.05 0.07

15.5 16.9 14.7 18.4 68.6 15.0 18.0 20.6 112 161

0.7 1.2 1.2 1.8 5.6 1.6 2.2 1.7 73 32

0.008 0.007 0.006 0.008 0.013 0.007 0.008 0.007 0.012 0.015

0.002 0.001 0.001 0.001 – 0.001 0.001 0.001 0.002

NuRD complex

CDK2AP1 CHD4 GATAD2A GATAD2B MBD2 MBD3 MTA1 MTA2 MTA3

2.63 – 0.56 0.62

59.5 – 0.09 0.11

– – 8.8 1.7

– 0.024 0.016 0.014

– 0.011 0.001 0.001

2.9

41.6 24.9

3.1 4.1

0.013 0.011 0.010 0.012

0.001 0.004 0.001 0.002

45.9 – 79.6 53.6 81.5 46.9 45.5 119 48.3

2.5 – 3.1 4.3

0.09 0.11 0.23 0.07

– 0.41 0.38 0.41 0.66 0.33 0.28 0.56 0.28

– – 0.09 0.10

0.68 0.53 0.60 0.41

– 64.1 42.9 33.0 41.9 38.2

3.9 6.8 34 7.9

0.020 0.057 0.020 0.032 0.023 0.020 0.021 0.028 0.019

0.004 0.002

BRMS1 SAP18 SAP30 SAP30L SIN3A SIN3B

0.41d – 0.09 0.20 0.10 0.16

– – 0.01

0.043

– – 0.17 0.17 0.14 0.26

– – 0.04

> 600

0.036 0.027

– – 0.002 – 0.009 0.016

0.03 0.14

> 600 > 600

– – – – – –

0.022 – 0.025 0.049 0.025 0.024

– >20,000 – – – – 0.003 10,485 3,911 0.015 13,236 4,056 0.006 12,671 4,281 15,690 2,739

NCoR complex

GPS2 NCOR1 NCOR2 TBL1X TBL1XR1

0.31 0.25 0.35 0.29 0.27

0.11 0.07 0.02 0.13 0.09

2.9 1.8 1.9 2.6 2.3

0.1 – 0.1 0.1 0.1

0.016 0.017 0.017 0.018 0.014

0.001 0.002 0.002 0.002 0.001

0.20 0.26 0.21 0.26 0.27

0.07 0.06 0.06 0.05 0.04

6.6 5.7 6.8 4.9 4.9

0.5 0.7 1.2 0.3 0.3

0.027 0.024 0.026 0.029 0.025

0.005 0.000 0.001 0.002 0.003

6,068 3,598 4614 4,673 2,956

4,825 365 2,471 1,160 951

DNTTIP1

DNTTIP1 MIDEAS

0.22 0.31

0.07 0.05

55.5 162

4.4 78.2

0.014 0.013

0.001 0.002

0.18 0.18

0.03 0.09

119 156

20 28

0.014 0.013

0.002 0.003

1,187 1,110

286 371

ELM-SANT domain proteins

MIER1 MIER2 MIER3 RERE

0.24 0.59 0.11 0.49

0.03 0.22 0.01 0.12

21.4 26.3 6.3 222

3.7 – – 48.2

0.023

0.003 – 0.001 0.005

0.25 – 0.12 0.32

0.05 0.04 0.20

36.1 37.0 – 272

3.9 – – 57

0.016 – 0.007 0.025

0.002 – – 0.009

1,423 – 954 4,336

892 – 181 1,131

> 30 2.14 1.90 > 30 3.90

– 0.26 0.07 – 0.57

– – – – –

0.66 2.72 > 10 0.06 > 10

0.06 0.40 – 0.01 –

– – – – –

>1 0.96 >1 >1 >1

– 0.02 – – –

Sin3 complex

Selected ALDH1A2 HDACi off-targets BZW2 CBR1 ISOC2 PPP3CA

0.01 0.05

> > > > > >

> > > > >

300 300 300 300 300 300

300 300 300 300 300

– – – – – –

– – – – –

0.029

0.009 0.039 > > > > >

0.1 0.1 0.1 0.1 0.1

0.04 0.11 0.26 0.04

> 600

> > > > >

600 600 600 600 600

0.002 0.023 0.002

– –

594 568 493 748 509 516 536 534 1251 8,23

158 174 95 202 – 130 172 123 327

2,897 1,219 1,429 1,153 1,483 1,223 1,048 1,099

8641 231 212 249 618 140 180

10,528 5,352 >20,000 – >20,000 – >20,000 – >20,000 –

aNo

Kdapp listed: protein was not detected in sample. bNo s.d. listed: protein was only detected in one sample or not at all. c >, no inhibition at maximum compound concentration tested (in at least two experiments). ditalics, IC50 value listed (no Kdapp was determined).

258

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

a

b

Baits:

SIN3

1.0

1.0

Relative enrichment in Sin3A in IP

0.5 ident. in IP: HDAC (class I)

H D H AC D 1 AC LS 2 M D1 T SI A3 N D 3A N TR TT I C ER P1 D F EH YL 1 H MT D 2 A TB C L1 3 XR 1

Enrichment in IP vs ctrl IgG HDAC1 HDAC2 HDAC3 CoREST GSE1 HMG20A HMG20B LSD1 PHF21A RCOR1 RCOR2 RCOR3 ZMYM2 ZMYM3 ZNF217 NuRD CDK2AP1 CHD3 CHD4 GATAD2A GATAD2B MBD2 MBD3 MTA1 MTA2 MTA3 NuRD/Sin3 RBBP4 (shared) RBBP7 Sin3 ARID4A ARID4B BRMS1 BRMS1L* ING1* ING2* SAP30 SAP30L SAP130 SIN3A SIN3B SUDS3 ELM-SANT MIER1 MIER2 MIER3 RERE TRERF1 MIDEAS DNTTIP1 DNTTIP1 HATs CDYL Histone EHMT1 methylation EHMT2 WDR5 WIZ NCoR GPS2 NCOR1 NCOR2 TBL1X TBL1XR1 Histones HIST1H1C HIST1H4A HIST2H2BA

c

1.0

Enrichment IP

29

0.8

9 15

0.6

1

0.4 2

0.2

5

4

0

6

7

–0.2

3

CoREST

Relative enrichment in TBL1XR1 in IP

1.0

NCoR

7

4 3

5

0.8

6

0.6 0.4 0.2 13 28 1 14 16 12 915 17

0 –0.2

1 HDAC1 2 HDAC2 3 HDAC3 4 TBL1XR1 (bait) 5 NCOR1 6 TBL1X 7 GPS2 8 MTA3 (bait) 9 CDK2AP1 10 CHD3 11 CHD4 12 GATAD2A 13 GATAD2B 14 MBD2 15 MBD3 16 MTA1 17 = MTA2

NuRD

–0.4 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1.0 Relative enrichment in MTA3 in IP

19 11 9 20 1 6 15 28 5 22 21 1 14 26 161730 2 7 24 4 37

29

26 19 7 5 27 15 17212022 1 28 11 372 14 16 6 9 39 45 402413 41 388 4 42 23 18 432544 12 10 31 30

32 34 33 3 36

13

35 7

27

HDAC1

0.5

1 HDAC1 2 HDAC2 3 LSD1 (bait) 4 HMG20A 5 RCOR3 6 RCOR1 7 GSE1 8 SIN3A (bait) 9 SAP30 10 ARID4B 11 SAP30 12 SUDS3A 13 BRMS1 14 RBBP4 15 RBBP7

11 12 13 8 14 10

–0.4 –0.4 –0.2 0 0.2 0.4 0.6 0.8 Relative enrichment in LSD1 in IP

31

18

23

HDAC2

0

1

1.0 1

9

11

1

29

4 2

HDAC3

0

7

Enrichment IP

Figure 3 Deconvolution of protein complexes by co-IP analysis confirms the identification of novel HDAC complexes. (a) HDAC complexes identified by both chemoproteomics profiling and co-IP–MS/MS analysis of HDAC complexes. IPs were performed from K562 cells using antibodies for HDAC1, 2 and 3, known complex components (the CoREST subunit LSD1, the NuRD subunit MTA3, the Sin3 subunit SIN3A, and the NCoR-subunit TBL1XR1) and examples of novel HDAC interacting proteins. * denotes previously reported complex components not captured by the SAHA matrix. The color code indicates enrichment E of immunoprecipitated proteins as compared to mock-IP experiments (scales from −1 to 1, E = 0 denotes equal abundance, see Online Methods). (b) Examples of the quantitative mapping of immunoaffinity-purified protein complexes by MS/MS. Purifications conducted with two different antibodies each, and corresponding isotype controls, were combined after PAGE, trypsinization and isobaric tagging. Quantification data are shown as plots of relative enrichment in immunoprecipitates of Sin3 versus LSD1 (upper panel), and MTA3 versus TBL1XR1 (lower panel). Each square represents a protein with its size scaled according to the number of sequenceto-spectrum matches. (c) HDAC protein complexes in chemical and protein space. For each protein identified in chemoproteomics and co-IP experiments, enrichment in the IP samples is plotted against the average relative affinity data across all inhibitors tested. Target proteins are represented in red (class I HDACs), blue (CoREST components), green (NCoR components), purple (NuRD components), light blue (Sin3 components), pink (MiDAC components) and yellow (ELM-SANT proteins). The square size indicates the confidence of the interaction with the immunopurified protein complex (large squares: FDR < 0.05, medium-sized squares: 0.05 < FDR < 0.15, c.f. Supplementary Figs. 8 and 9 and Supplementary Table 4).

14

22 15 28 21 19 2 18 16 17

0

1 24

1 23

26

Protein HDAC1 HDAC2 HDAC3 GSE1 HMG20A HMG20B LSD1 PHF21A RCOR1 RCOR2 RCOR3 ZMYM2 ZMYM3 CDK2AP1 CHD4 GATAD2A GATAD2B MBD2 MBD3 MTA1 MTA2 MTA3

Group HDAC (class I) CoREST

NuRD

No. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

Protein BRMS1 SAP30 SAP30L SIN3A SIN3B RBBP4 RBBP7 DNTTIP1 MIDEAS GPS2 NCOR1 NCOR2 TBL1X TBL1XR1 MIER1 MIER2 MIER3 RERE CDYL EHMT1 EHMT2 RREB1 WIZ

Group Sin3

No. 23 24 25 26 27 NuRD/Sin3 28 29 DNTTIP1 30 31 NCoR 32 33 34 35 36 ELM-SANT 37 38 39 40 Other 41 42 43 44 45

35 25 28

36

3

1

33 32

20

5

29 6

LSD1

0.5

MTA3

0

1

1.0 Enrichment IP

© 2011 Nature America, Inc. All rights reserved.

Articles

SIN3A

0

1

1

41

30

30

1

2

37

43

42 45

28

31

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

0

38

1

HDAC1 and HDAC2 (Supplementary Fig. 4 and Supplementary Data Set 3). The quantiDNTTIP1 0.5 fication of relative protein amounts sequestered 0 Relative affinity by the probe matrix upon sequential incubations enabled us to determine apparent dissociation constants (Kdapp) from the IC50 values. The resulting deviation between Kdapp and IC50 values was less than twofold for 99% of the proteins (Supplementary Data Set 4). It is interesting to compare the target profiles of SAHA and its analog BML-210, which are identical except for the replacement of the hydroxamate by an aminobenzamide group in BML-210. This change causes a general drop of potency for class I HDACs (and complete loss of potency for class IIb), concomitant with an increase in selectivity for HDAC3 relative to HDAC1 or HDAC2 (Fig. 2a). Some compounds of the aminobenzamide class were reported to exhibit slow binding kinetics for class I HDACs, in particular for an HDAC3-NCoR2 complex30. We explored the effect of preincubation of aminobenzamide inhibitors with the cell extract at 22 °C, or prolonged preincubation at 4 °C. Under either condition we observed a more pronounced inhibition of HDAC3 binding to the probe matrix (Supplementary Fig. 5). Consequently, we performed additional

7

TBL1XR1

0

41

37 2 1 2

TRERF1 1

0

Relative affinity

CDYL 1

0

Relative affinity

EHMT2 1

0

Relative affinity

1

profiling experiments for all aminobenzamides using preincubation at 22 °C and found that inhibition of all the class I HDACs was affected, with the greatest change in inhibitor potency for HDAC1 and HDAC2 (up to fivefold for some compounds; Supplementary Fig. 3). The data set comprises concentration-inhibition profiles for 16 compounds, assayed with 1,251 proteins. Combinations with at least twofold reduction in binding by any of the inhibitors are shown (Fig. 2b). The set of proteins included the class I and IIb HDACs, components of HDAC complexes, and putative novel complex components or targets (data for selected inhibitors listed in Table 1, additional inhibitors in Supplementary Table 3 and MS data in Supplementary Data Set 2). Because the profiles were generated using high maximum inhibitor concentrations, it is unlikely that all of these proteins are physiologically relevant targets. The inhibition curves typically displayed a Hill coefficient of ~1 (indicating stoichiometric binding of one inhibitor molecule per enzyme molecule), with 259

260

a

HeLa extract

Probe-matrix bound

Vehicle Aph. Noc. Aphidicolin Nocodazole

Vehicle DNTTIP1 HDAC1

RFU b 3,500

Nocodazole Nocodazole, TSA 10 µM Vehicle Vehicle, TSA 10 µM

3,000 2,500 2,000

*

1,500 1,000 500

3A N SI

(N 3 TA M

(S

uR

in

D

3)

)

) ES T oR (C 1

1 IP TT N

LS D

(M

tro

iD

lI

AC

gG

)

0

D

Deconvolution of HDAC target complexes To differentiate between novel direct targets and novel components of HDAC complexes, we carried out a series of co-immunoprecipitations (co-IP) with quantitative MS/MS analysis. We evaluated a set of 27 antibodies and ultimately selected 14 directed against three class I HDACs, four known complex protein subunits (LSD1 from CoREST, MTA3 from NuRD, SIN3A from Sin3, TBL1XR1 from NCoR) and four potential HDAC complex components identified in the chemoproteomics profiles (DNTTIP1, the ELM-SANT protein TRERF1, the histone acetylase CDYL and the histone methylase EHMT2). The co-IP data sets comprise many known complex components and new targets or components already identified in the chemoproteomics profiles (Fig. 3a and Supplementary Data Set 5). Co-IP samples from two different antibodies and corresponding control IgG samples were combined for quantitative MS/MS analysis, such that specifically co-immunopurified proteins were clearly discriminated from background and differences in complex compositions were directly quantified. Notably, the display of the SIN3Aversus LSD1-precipitated proteins indicated a preference of the Sin3 complex for HDAC1 relative to HDAC2 (Fig. 3b). However, most co-IP samples contained large numbers of co-purifying proteins, thus obscuring the identity of true interactors. For instance, ~300 proteins were at least twofold enriched with two different HDAC2 antibodies relative to control IgG samples. Evidently, HDAC inhibitor target complexes contained in the co-IP data sets should also be present in the chemoproteomics data (Fig. 2b). Hence we analyzed the overlap between these orthogonal data sets and devised a confidence score to statistically assess the association of proteins with the target protein complexes (Fig. 3c, Supplementary Fig. 8 and

Supplementary Table 4). For each protein, the relative enrichment in each co-IP and the average relative potency values across all inhibitors were normalized to a scale from 0 to 1, and the confidence score was defined as the sum of squares of both values, scaling between 0 and 2. The calculation of this score for each protein in the experimental and a corresponding randomized data set enabled determination of high-confidence complex associations with very low false-discovery rates (Supplementary Fig. 9). The ELM-SANT domain proteins MIER1, MIER2, MIER3 and RERE were co-purified with HDAC1 and/or HDAC2 but not with the CoREST, NuRD, Sin3 or NCoR complexes. This is consistent with their inhibition profiles not matching any of the known complexes (Fig. 2b). These proteins therefore likely represent components of distinct HDAC complexes formed around ELM-SANT scaffolds31. MIER1, MIER2 and HDAC2 (but not HDAC1) were co-purified with the putative histone acetylase CDYL. MIER1, HDAC2 and CDYL were also found in the EHMT2 co-IP, suggesting that the reported CDYL-EHMT2 complex32 contains MIER1, MIER2 and HDAC2. The MiDAC complex was confirmed in the DNTTIP1 co-IP, which comprised HDAC1, HDAC2, MIDEAS and TRERF1, an ELM-SANT protein related to MIDEAS, which also copurified with HDAC2. Immunoaffinity purification of TRERF1 itself confirmed its association with DNTTIP1 and HDAC1/2, but not with MIDEAS. This suggests that TRERF1 and MIDEAS represent alternative scaffolds for related complexes. To further characterize the MiDAC complex, we assessed the expression of DNTTIP1 and HDAC1 in HeLa cells arrested in mitosis or in early S-phase. No differences in expression were detected (Fig. 4a).

on

the notable exception of romidepsin, which consistently yielded a Hill coefficient of 2 (Supplementary Fig. 6). Bidirectional hierarchical clustering of the complete data set clearly outlines several HDAC complexes defined by their inhibition profiles relative to the major chemical compound classes (Fig. 2b and Supplementary Fig. 7). In the chemical dimension, the clustering is driven by the major chemotypes with several hydroxamate subclusters that differed in their effect on class II HDACs. In agreement with published data, we found that peptidic and hydroxamate compounds are substantially more potent than aminobenzamides. A notable observation was the unexpected degree of selectivity of aminobenzamide inhibitors, which show a preference for the HDAC3-NCoR complex. Clustering in the protein dimension is driven by the association of proteins in complexes, because all subunits of a target complex exhibit Kdapp values for a given inhibitor. It should nonetheless be noted that proteins known to reside in two or more complexes (e.g., HDAC1, HDAC2, RBBP4/7 or LSD1 (ref. 7) are predicted to exhibit Kdapp values representing aggregates between the individual complexes. The data show that complex subunits remained associated with the inhibitor target proteins during the assay procedure, and that they affect the inhibitor-binding properties of the catalytic HDAC subunits. Whereas the HDAC1/2-containing CoREST and NuRD complexes showed similar inhibition profiles across the compound set, there were marked differences in the Sin3 inhibition profiles, with the aminobenzamides and valproate showing a much lower potency for the Sin3 complex compared to CoREST and NuRD. The clustering delineates additional HDAC complexes formed around the ELM-SANT domain proteins MIDEAS, MIER1, MIER2, MIER3 and RERE. Notably, MIDEAS, DNTTIP1 and CDYL, which were captured in greater amounts by the probe matrix from mitotic cells compared to nonmitotic cells, were clustered in close proximity, suggesting the existence of a distinct mitotic deacetylase complex (MiDAC).

C

© 2011 Nature America, Inc. All rights reserved.

Articles

Antibody

Figure 4 Class I HDACs and DNTTIP1 form a mitotic deacetylase complex (MiDAC). (a) Cell cycle-dependent association of DNTTIP1 with the SAHA probe matrix was probed by western blot analysis. Lanes 1–3; identical expression levels of DNTTIP1 and HDAC1 in HeLa cells treated with aphidicolin (induces G1/S-phase arrest), nocodazole (induces arrest in mitosis) or vehicle. Lanes 4–9; increased amounts of DNTTIP1 is captured by the SAHA matrix from nocodazole-treated cells (lanes 6, 7) compared to aphidicolin (lanes 4, 5) or vehicle (lanes 8, 9). (b) Deacetylase activity assay of immunoaffinity-precipitated HDAC complexes demonstrates increased mitotic activity of a DNTTIP1-containing complex (MiDAC) but not of LSD1 (CoREST), MTA3 (NuRD) and SIN3A (Sin3)-containing complexes. Values are displayed as relative fluorescence units (RFU ± s.d.; N = 3; *P < 0.001 (Student’s t-test)). The fluorescence signal is reduced to background by 10 µM trichostatin A.

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

Articles Taken together with the fact that more DNTTIP1 was captured by the SAHA matrix from mitotic cells than from nonmitotic cells, the data indicated an increase in complex formation during mitosis, rather than an increase in the expression of constituent proteins. Next, we measured the deacetylase activity of the immunoprecipitated MiDAC complex (again isolated with the DNTTIP1 antibody), and compared it to the CoREST complex (LSD1 antibody), NuRD complex (MTA3 antibody) and Sin3 complex (SIN3A antibody). Whereas all samples possessed substantial deacetylase activity, only the MiDAC complex exhibited greater activity in divisionarrested cells (Fig. 4b).

b

c

A SA H A PC I-2 Ta 47 ce 81 d R ina om li i ne PC dep I-3 sin Bu 40 fe 51 x Va am a lp ro c at e

TS

VC

SAHA

Cmpd

4

4

Tacedinaline

H2A 1 × ac

3 × ac H2B 1 × ac

PCI−24781

3 × ac 4 × ac H3

1 × ac 2 × ac

H4

0

K me STGGK ac APR

2

0

1 × ac 2 × ac 3 × ac 4 × ac

Histones 3.725

GK ac GGK ac GLGK ac GGAK ac R 4

2

0

VC TS A

Avg. log2 FC vs. unmod 0.000

Fold increase (log2 scale)

PCI−34051

2

2 × ac

2 × ac

Romidepsin

K ac STGGK ac APR

S PC AH Ta -2 A ce 47 d 81 R ina om lin id e PC eps I-3 in Bu 40 fe 51 xa Va ma lp c ro at e

Overlay

Fold increase (log2 scale)

Acetylated tubulin

Fold increase (log2 scale)

Acetylated histone H3

Vehicle

a

Valproate

© 2011 Nature America, Inc. All rights reserved.

Cell-based profiling of HDAC inhibitors To investigate the correlation of proteomics target profiles with substrate selectivity, a subset of the reference inhibitors was subjected to cell-based tubulin and histone modification assays. K562 and HeLa cells were treated with vehicle or compounds, including nonselective inhibitors (SAHA, PCI-24781), class I selective inhibitors (tacedinaline, romidepsin, valproate) and the HDAC8 inhibitor

PCI-34051. Cell viability was monitored and drug effects were detected by antibodies for acetylated tubulin as the major substrate of HDAC6 (ref. 16) and histones H3 and H4 as the major class I HDAC substrates by immunofluorescence and western blot analysis (Fig. 5a and Supplementary Fig. 10). The nonselective HDAC inhibitors increased steady-state acetylation of tubulin and histones manifested by the staining of acetylated microtubules and punctuate nuclear staining of acetylated histones. The class I selective HDAC inhibitors stimulated histone acetylation but did not affect tubulin, as expected. Aliquots of vehicle-treated and drug-treated cells were also compared by differential mapping of histone acetylation and methylation marks using quantitative high-resolution MS/MS33 (Fig. 5b and Supplementary Data Set 6). The results confirmed the range of activities observed in chemoproteomics profiling and indicate a pronounced abundance of hyperacetylated histone peptides after treatment with nonselective HDAC inhibitors, in particular TSA and romidepsin. In contrast, valproate exhibited a more selective effect, in particular less acetylation of H3K9 in peptides containing acetylated K14 (Fig. 5c).

Figure 5 Differential effects of HDAC inhibitors on histone and tubulin acetylation. Immunofluorescence analysis of histone H3 (K9ac/K14ac) and tubulin acetylation in HeLa cells treated for 4 h with vehicle, SAHA (10 µM), tacedinaline (50 µM), PCI-24781 (20 µM), PCI-34051 (100 µM), romidepsin (1 µM) or valproate (2mM). (a) Mapping of histone acetylation in K562 cells treated with HDAC inhibitors by LC-MS/MS. Cells were treated with TSA (10 µM), SAHA (5 µM), PCI-24781 (2 µM), tacedinaline (50 µM), romidepsin (1 µM), PCI-34051 (20 µM), bufexamac (100 µM) or valproate (2 mM) for 6 h. Histones were extracted from cells and acetylated peptides were quantified after isobaric tagging. (b) Heat map showing abundance of peptides with single or multiple acetylated lysines as dependent on inhibitor treatment. (c) Abundance of differently modified variants of the Histone H3.3 peptide 9-17 and the fully acetylated H4-peptide 5-19. Triplicate experiments were performed and error bars represent s.e.m. (Supplementary Data Set 6).

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

261

Articles

60 40 20 0

OH

O

80

NH O

Bufexamac

60 40 20 0

80

b

O H2N

O

100

HN

1.2

HN

H2N

N

60

Residual binding

80

100

Percent inhibition HDAC3

100

Percent inhibition HDAC6

Percent inhibition HDAC2

a

Cpd AA-1 Cpd AA-2

40 20

1.0 0.8 0.6

HDAC1 (NI) HDAC2 (NI) HDAC3 app (KD = 341 µM) HDAC6 app (KD = 10.7 µM) app HDAC8 (KD = 235 µM) app HDAC10 (KD = 12.3 µM)

0.4 0.2

0

0 40

60

80 100

Percent inhibiton HDAC1 Acetylated histone H3K9

Acetylated tubulin

20

40

60

80 100

0

Percent inhibition HDAC1 Overlay

20

40

60

80 100

1

10 100 1,000 [Bufexamac] (µM)

Percent inhibition HDAC1 100

d

50

100

0 0.01

Percent inhibition / viability

AA-2 Bufexamac

© 2011 Nature America, Inc. All rights reserved.

Control

c

0

0.1 1.0 10 [Cpd AA-2] (µM)

100

100 IC50 = 2.9 µM 50

ac-H4K5 ac-H3K9/K14 ac-tubulin Viability

IFN-α secretion [% inhibition]

20

Percent inhibition / viability

0

IC50 = 8.9 µM

50

0 0.1

1.0 10 [Bufexamac] (µM)

100

0 0.01

0.1 1.0 10 [Bufexamac] (µM)

100

Figure 6 The nonsteroidal anti-inflammatory drug bufexamac is a novel class IIb HDAC inhibitor. (a) Screen of a focused compound library against HDACs 1, 2, 3 and 6 using a chemoproteomics binding assay with the SAHA matrix in whole cell extract from Jurkat and Ramos cells. The plots outline inhibition relative to HDAC1 for HDAC6, HDAC3 and HDAC2, as quantified by antibodies on dot-blot arrays. The compound concentration was 10 µM and chemical structures of selective hit compounds are shown. (b) HDAC selectivity profile of bufexamac in K562 cells, measured as outlined in Figure 1. (c) Treatment of HeLa cells with bufexamac elicits hyperacetylation of tubulin, whereas treatment with the o-aminoanilide AA-2 leads to hyperacetylation of histones. Cultured cells were treated with vehicle or drug for 4 h, and cells were analyzed by immunofluorescence microscopy and by western blot analysis using antibodies for acetylated tubulin (EC 50 = 2.9 µM) and acetylated histones H3 (K9) and H4 (K5), respectively. (d) Treatment of peripheral blood mononuclear cells with bufexamac inhibits the secretion of IFN-α (EC50 = 8.9 ± 4.9 µM, three independent experiments).

Similarly, we observed fourfold less peracetylated H4(5-19) peptide in valproate-treated cells than in TSA-treated cells. Bufexamac as a novel class IIb HDAC inhibitor To discover novel selective HDAC inhibitors, we developed a high-throughput adaptation of the chemoproteomics protocol in which we replaced MS/MS detection with multiplexed fluorescent antibody detection on ‘dot blot’ arrays. The method was applied to the screening of a focused compound library in whole cell extracts of Jurkat and Ramos cells for inhibitors of HDAC1, 2, 3 and 6. Several hits were obtained with a few compounds displaying a notable degree of selectivity (Fig. 6a). Two aminobenzamide fragments were identified as hit compounds exhibiting selectivity for HDAC3. Bufexamac, a nonsteroidal anti-inflammatory drug with an unknown mechanism of action 34, preferentially affected HDAC6. Bufexamac was subjected to quantitative proteomics profiling as described above for the reference HDAC inhibitors set, which confirmed its selectivity for HDAC6 and HDAC10, the other class IIb isoform, in addition to several non-HDAC targets (Fig. 6b and Supplementary Data Set 2). The results are consistent with tubulin immunofluorescence and western blot data, which showed 262

a much larger amount of acetylated tubulin, the major HDAC6 substrate, but not of acetylated histones as substrates of class I HDACs (Figs. 5b and 6c). The cellular potency for tubulin deacetylation correlated with the potency for one of the drugs’ anti-inflammatory effects, the secretion of interferon (IFN)-α in peripheral blood mononuclear cells (Fig. 6d). DISCUSSION Gene transcription and its epigenetic regulation are controlled by megadalton protein complexes35,36. Therefore, the action of drugs which modulate epigenetic mechanisms should be considered in the context of the multiprotein complexes they target. We developed an affinity capture method combined with multiplexed protein quantification by mass spectrometry to probe the interaction of drug molecules with drug targets in cells or tissue under conditions that preserve the integrity of protein complexes. To our knowledge, this is the first demonstration that small molecules exhibit different affinities toward (that is, they “recognize”) different protein complexes containing the same catalytic subunit. The strategy has the potential to be extended to other classes of pharmacological target, and enables the discovery of drug leads and their molecular targets as functional protein complexes. The biological VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

Articles activity of compounds is assessed without use of recombinant purified proteins or protein overexpression. The major prerequisite is a probe matrix which binds to a sub-proteome which is characterized by a shared chemical ligand space, typically based on a substrate- or cofactor-binding site. We developed a probe matrix that captures the class I and class IIb HDACs, and the majority of previously reported subunits of HDAC complexes, by binding to the substrate pocket. Class IIa HDACs were not identified, presumably because they exhibit low catalytic activity and low affinity for the hydroxamate probes14. Moreover, the probe matrix binds to many other enzymes which potentially represent targets sharing a similar chemical ligand space, including other metallo enzymes, as well as other proteins which may be associated with enzymes in protein complexes. A basic application of a probe matrix is the differential expression profiling of a subproteome across a range of biological samples and conditions. In line with a pleiotropic function most proteins binding to the hydroxamate matrix displayed minor differences across a panel of cell lines and tissues. However, two proteins lacking an obvious small-molecule binding site were captured predominantly from mitotically arrested cells (DNTTIP1 and MIDEAS), in line with the formation of a specific mitotic HDAC complex. We confirmed and extended this finding by using inhibitor profiling, co-IP and enzyme assay data. A powerful application of the probe matrix is the profiling of drugs and lead molecules interacting with proteins and protein complexes in cells. We demonstrated that robust quantitative data are obtained by high sensitivity liquid chromatography (LC)/LC-MS/MS to mea sure protein binding to the matrix in whole cell extract as a function of the concentration of competing “free” inhibitor. Unbiased bidirectional hierarchical clustering of the proteomics target profiles of 16 inhibitors (Fig. 2b) guides (i) the classification of drugs in selectivity clusters, which are predicted to exhibit similar pharmacological effects, (ii) the grouping of protein targets in chemical space, and (iii) the assignment of targets to protein complexes. In order to distinguish individual targets that share a structurally similar ligand binding space from proteins associated in a complex, the chemoproteomics clustering is correlated with co-IP mapping of endogenous complexes in the same cell type37. Our data set clusters 16 inhibitors in terms of their effects on 1,251 proteins that specifically interact with the probe matrix. The clustering of inhibitors is driven by the major chemotypes represented by hydroxamates and aminobenzamides, with several hydroxamate sub-clusters, and reveals an unexpected degree of selectivity for inhibitors previously perceived as nonselective25,38,39. However, much of the published data is inconsistent, raising issues with the enzyme assays employed. Notably, a recent carefully controlled extensive enzyme kinetic study of HDAC inhibitors also reported a higher degree of inhibitor selectivity14. Remarkably the compounds in the aminobenzamide cluster showed several distinctive features. Extending previous findings30, we observed slow binding to class I complexes, in particular HDAC1/2-dependent CoREST and NuRD complexes. Moreover, we found a characteristic selectivity profile with a preference for the HDAC3-NCoR complex and no or minor effects on the HDAC1/2-dependent Sin3 complex. It is tempting to speculate whether this selectivity profile may contribute to a more favorable toxicology profile or to reduced clinical efficacy40. Similarly, the clinically used drug valproate also affected the Sin3 complex to a lesser degree than other class I complexes. We did not find major differences when we assessed the effect of inhibitors with different selectivities on global histone acetylation. However, little is known about site specificity of the different complexes and their relative nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

c ellular activities, which are likely cell-type and gene specific. To study these effects our methodology could be extended to include chromatin immunoprecipitation performed with the targets and antibodies validated in the co-IP studies. A number of non-HDAC targets are potently affected by several hydroxamate HDAC inhibitors but do not appear to be components of HDAC complexes, given that their inhibition profiles do not match that of any HDAC (Fig. 2), and because they were not enriched in the co-IP set (Fig. 3a). These proteins may represent off-targets sharing a similar chemical ligand space. Examples are the basic leucine zipper/W2 domain protein BZW2 and the isochorismatase domain protein ISOC2, and several other Zn2+-dependent metalloenzymes. In the protein dimension, the clustering data delineate target protein complexes, as proteins exhibiting matching inhibition profiles across the inhibitor panel are likely to be physically associated. This is evidenced by the excellent clustering of the four major HDACcontaining complexes (Fig. 2b). To our knowledge, this is the first time that small-molecule binding data are used to characterize target protein complexes. We thus extended our data by conducting an extensive co-IP analysis of endogenous HDAC complexes from the same cell extract. A few previously reported class I HDAC complex components that did not bind to the probe matrix were identified in the co-IP samples and may represent interactions that are sensitive to inhibitor binding, for example, the ING2 subunit of the Sin3 complex41. The co-IP results confirmed additional HDAC1/HDAC2 complexes delineated in the analysis of the chemoproteomics data. These complexes are built around ELM-SANT domain subunits that are phylogenetically related to corepressor components of NuRD and CoREST complexes31. Several of such complexes exist with each containing a single ELM-SANT scaffold, such as MIER1, MIER2, MIER3, RERE, TRERF1 or MIDEAS, a previously unannotated gene product with homology to the REST corepressor. One function of HDAC complexes is likely the coordination of deacetylation with other epigenetic modifications. The CoREST complex couples HDACs to the demethylase LSD19, and MIER1 and RERE were shown to scaffold HDACs with the EHMT methyltransferases31, and our inhibition profiles confirmed these complexes as HDAC inhibitor targets. The composition of these HDAC complexes was deconvoluted further by the co-IP data, in particular the MiDAC complex formed by HDAC1/2, MIDEAS and DNTTIP1. DNTTIP1 is a DNA binding protein that has been described to modulate the activity of terminal deoxynucleotidyl transferase (TDT), a specialized DNA polymerase that incorporates nontemplated nucleotides to the 3′ end of DNA templates to mediate the junctional diversity of immunoglobulin genes42. However, we did not consistently identify an association of TDT with the MiDAC complex, suggesting a TDT independent function of the complex in cell division. The inhibition profiles also implicated the REST corepressor CDYL32 as a component of MiDAC, in line with the increase in CDYL captured from mitotic cells by the SAHA matrix. CDYL co-immunopurified with HDAC2 but not with the MiDAC subunit DNTTIP1, and hence further analysis is required to clarify whether it is a component of MiDAC or of an alternative complex. The fourth class I enzyme, HDAC8, is more difficult to assign to a complex, because it is only targeted by a few inhibitors. We did not identify a suitable antibody to characterize HDAC8 by co-IP. The HDAC8 inhibitor PCI-3405143 was the only compound in our panel that was specific for a single HDAC. The class IIb enzymes HDAC6 and HDAC10 were only inhibited by hydroxamate type compounds and both do not appear to form robust complexes as no 263

© 2011 Nature America, Inc. All rights reserved.

Articles strong associations with other proteins in the inhibition profiles were detected. HDAC6 has recently been implicated in chromatin regulation44 but the HDAC6/HDAC10 inhibitor bufexamac did not affect the acetylation of histones, suggesting that class IIb deacetylases do not play a direct role of in histone modification. Our chemoproteomics methodology can be adapted to high throughput screening by using an antibody-based readout to reduce sample requirements and process time. We conducted a screen of a focused compound library for selective inhibitors. The screen identified the hydroxamate drug bufexamac, an NSAID with an unknown mechanism of action34 as a class IIb selective inhibitor. Its profile was unique among the set of inhibitors studied. The drug induced tubulin hyperacetylation in drug concentrations matching its antiinflammatory effect. Therefore, inhibition of HDAC6 may contribute to the clinical efficacy of bufexamac. In conclusion, we have shown that a chemoproteomics strategy based on small-molecule inhibitors can be applied to discover and classify molecular complexes around drug target proteins, which has not, to our knowledge, been previously shown. The approach confirms and extends orthogonal protein-protein interaction mapping. We have demonstrated the utility of this strategy in drug discovery by measuring distinctive target profiles for clinical HDAC inhibitors in cell extracts, and employed it in screening for novel inhibitors. The data support the value of drug discovery strategies based on target proteins in their biological context. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Accession numbers. PRIDE database (http://www.ebi.ac.uk/pride): mass spectrometry data set accession numbers 15345–15472. Note: Supplementary information is available on the Nature Biotechnology website. Acknowledgments This work was supported by a grant from the German Bundesministerium für Bildung und Forschung (Spitzencluster BioRN, Verbundprojekt Inkubator/ Teilprojekt INE-TP01) to Cellzome AG. We are grateful to N. Garcia-Altrieth, M. Jundt, M. Löttgers, J.-I. Huber, M. Klös-Hudak, J. Krause, B. Kröh, A. Podszuweit, T. Rudi and K. Weis for expert technical assistance, to C. Gemünd and V. Wolowski for the development of software and database tools, and to F. Weisbrodt for help with the figures. We would like to thank T. Edwards, O. Rausch and D. Simmons for suggestions and support. AUTHOR CONTRIBUTIONS A.D., D.E., A.-M.M., and K.S. performed biochemical and cell biological experiments; V.R. synthesized and sourced compounds; D.P. performed the interferon assay; I.B. analyzed histone modifications; B.D., M.D. and M. Boesche prepared peptide samples and operated mass spectrometers; M. Bantscheff, M.M.S., T.M. and G.S. established and conducted mass spectrometry data handling processes; M.M.S., Y.A., C. Huthmacher and J.S. contributed data analysis and visualization; M. Bantscheff, C. Hopf, P.G. and G.D. analyzed data, planned and supervised experiments, and conceptualized the project; G.B., U.K., G.N. and N.G.R. contributed ideas and supported the work; and M. Bantscheff and G.D. wrote the paper. 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. Kouzarides, T. Chromatin modifications and their function. Cell 128, 693–705 (2007).

264

2. Choudhary, C. et al. Lysine acetylation targets protein complexes and co-regulates major cellular functions. Science 325, 834–840 (2009). 3. Zhao, S. et al. Regulation of cellular metabolism by protein lysine acetylation. Science 327, 1000–1004 (2010). 4. Karberg, S. Switching on epigenetic therapy. Cell 139, 1029–1031 (2009). 5. Taunton, J., Hassig, C.A. & Schreiber, S.L. A mammalian histone deacetylase related to the yeast transcriptional regulator Rpd3p. Science 272, 408–411 (1996). 6. Gregoretti, I.V., Lee, Y.M. & Goodson, H.V. Molecular evolution of the histone deacetylase family: functional implications of phylogenetic analysis. J. Mol. Biol. 338, 17–31 (2004). 7. Yang, X.J. & Seto, E. The Rpd3/Hda1 family of lysine deacetylases: from bacteria and yeast to mice and men. Nat. Rev. Mol. Cell Biol. 9, 206–218 (2008). 8. Cunliffe, V.T. Eloquent silence: developmental functions of Class I histone deacetylases. Curr. Opin. Genet. Dev. 18, 404–410 (2008). 9. You, A., Tong, J.K., Grozinger, C.M. & Schreiber, S.L. CoREST is an integral component of the CoREST-human histone deacetylase complex. Proc. Natl. Acad. Sci. USA 98, 1454–1458 (2001). 10. Tong, J.K., Hassig, C.A., Schnitzler, G.R., Kingston, R.E. & Schreiber, S.L. Chromatin deacetylation by an ATP-dependent nucleosome remodelling complex. Nature 395, 917–921 (1998). 11. Zhang, Y., Iratni, R., Erdjument-Bromage, H., Tempst, P. & Reinberg, D. Histone deacetylases and SAP18, a novel polypeptide, are components of a human Sin3 complex. Cell 89, 357–364 (1997). 12. Karagianni, P. & Wong, J. HDAC3: taking the SMRT-N-CoRrect road to repression. Oncogene 26, 5439–5449 (2007). 13. Guenther, M.G., Barak, O. & Lazar, M.A. The SMRT and N-CoR corepressors are activating cofactors for histone deacetylase 3. Mol. Cell. Biol. 21, 6091–6101 (2001). 14. Bradner, J.E. et al. Chemical phylogenetics of histone deacetylases. Nat. Chem. Biol. 6, 238–243 (2010). 15. Lahm, A. et al. Unraveling the hidden catalytic activity of vertebrate class IIa histone deacetylases. Proc. Natl. Acad. Sci. USA 104, 17335–17340 (2007). 16. Boyault, C., Sadoul, K., Pabion, M. & Khochbin, S. HDAC6, at the crossroads between cytoskeleton and cell signaling by acetylation and ubiquitination. Oncogene 26, 5468–5476 (2007). 17. Marks, P.A. & Breslow, R. Dimethyl sulfoxide to vorinostat: development of this histone deacetylase inhibitor as an anticancer drug. Nat. Biotechnol. 25, 84–90 (2007). 18. Bolden, J.E., Peart, M.J. & Johnstone, R.W. Anticancer activities of histone deacetylase inhibitors. Nat. Rev. Drug Discov. 5, 769–784 (2006). 19. Zhang, Y. et al. Analysis of the NuRD subunits reveals a histone deacetylase core complex and a connection with DNA methylation. Genes Dev. 13, 1924–1935 (1999). 20. Salisbury, C.M. & Cravatt, B.F. Activity-based probes for proteomic profiling of histone deacetylase complexes. Proc. Natl. Acad. Sci. USA 104, 1171–1176 (2007). 21. Bantscheff, M. et al. Quantitative chemical proteomics reveals mechanisms of action of clinical ABL kinase inhibitors. Nat. Biotechnol. 25, 1035–1044 (2007). 22. Ong, S.E. et al. Identifying the proteins to which small-molecule probes and drugs bind in cells. Proc. Natl. Acad. Sci. USA 106, 4617–4622 (2009). 23. Sharma, K. et al. Proteomics strategy for quantitative protein interaction profiling in cell extracts. Nat. Methods 6, 741–744 (2009). 24. Bantscheff, M., Scholten, A. & Heck, A.J. Revealing promiscuous drug-target interactions by chemical proteomics. Drug Discov. Today 14, 1021–1029 (2009). 25. Khan, N. et al. Determination of the class and isoform selectivity of small-molecule histone deacetylase inhibitors. Biochem. J. 409, 581–589 (2008). 26. Bantscheff, M. et al. Robust and sensitive iTRAQ quantification on an LTQ Orbitrap mass spectrometer. Mol. Cell. Proteomics 7, 1702–1713 (2008). 27. Jones, P. et al. Probing the elusive catalytic activity of vertebrate class IIa histone deacetylases. Bioorg. Med. Chem. Lett. 18, 1814–1819 (2008). 28. Kruhlak, M.J. et al. Regulation of global acetylation in mitosis through loss of histone acetyltransferases and deacetylases from chromatin. J. Biol. Chem. 276, 38307–38319 (2001). 29. Savitski, M.M. et al. Targeted data acquisition for improved reproducibility and robustness of proteomic mass spectrometry assays. J. Am. Soc. Mass Spectrom. 21, 1668–1679 (2010). 30. Chou, C.J., Herman, D. & Gottesfeld, J.M. Pimelic diphenylamide 106 is a slow, tight-binding inhibitor of class I histone deacetylases. J. Biol. Chem. 283, 35402–35409 (2008). 31. Wang, L., Charroux, B., Kerridge, S. & Tsai, C.C. Atrophin recruits HDAC1/2 and G9a to modify histone H3K9 and to determine cell fates. EMBO Rep. 9, 555–562 (2008). 32. Mulligan, P. et al. CDYL bridges REST and histone methyltransferases for gene repression and suppression of cellular transformation. Mol. Cell 32, 718–726 (2008). 33. Savitski, M.M., Mathieson, T., Becher, I. & Bantscheff, M. H-score, a mass accuracy driven rescoring approach for improved Peptide identification in modification rich samples. J. Proteome Res. 9, 5511–5516 (2010). 34. Trommer, H. et al. Examinations of the antioxidative properties of the topically administered drug bufexamac reveal new insights into its mechanism of action. J. Pharm. Pharmacol. 55, 1379–1388 (2003). 35. Alberts, B. The cell as a collection of protein machines: preparing the next generation of molecular biologists. Cell 92, 291–294 (1998). 36. Gavin, A.C. et al. Proteome survey reveals modularity of the yeast cell machinery. Nature 440, 631–636 (2006).

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

Articles 41. Smith, K.T., Martin-Brown, S.A., Florens, L., Washburn, M.P. & Workman, J.L. Deacetylase inhibitors dissociate the histone-targeting ING2 subunit from the Sin3 complex. Chem. Biol. 17, 65–74 (2010). 42. Kubota, T., Maezawa, S., Koiwai, K., Hayano, T. & Koiwai, O. Identification of functional domains in TdIF1 and its inhibitory mechanism for TdT activity. Genes Cells 12, 941–959 (2007). 43. Balasubramanian, S. et al. A novel histone deacetylase 8 (HDAC8)-specific inhibitor PCI-34051 induces apoptosis in T-cell lymphomas. Leukemia 22, 1026–1034 (2008). 44. Wang, Z. et al. Genome-wide mapping of HATs and HDACs reveals distinct functions in active and inactive genes. Cell 138, 1019–1031 (2009).

© 2011 Nature America, Inc. All rights reserved.

37. Malovannaya, A. et al. Streamlined analysis schema for high-throughput identification of endogenous protein complexes. Proc. Natl. Acad. Sci. USA 107, 2431–2436 (2010). 38. Beckers, T. et al. Distinct pharmacological properties of second generation HDAC inhibitors with the benzamide or hydroxamate head group. Int. J. Cancer 121, 1138–1148 (2007). 39. Blackwell, L., Norris, J., Suto, C.M. & Janzen, W.P. The use of diversity profiling to characterize chemical modulators of the histone deacetylases. Life Sci. 82, 1050–1058 (2008). 40. Farias, E.F. et al. Interference with Sin3 function induces epigenetic reprogramming and differentiation in breast cancer cells. Proc. Natl. Acad. Sci. USA 107, 11811–11816 (2010).

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

265

© 2011 Nature America, Inc. All rights reserved.

ONLINE METHODS

Reagents. All reagents were purchased from Sigma unless otherwise noted below. Antibodies were purchased from the following suppliers: sc-7872 (HDAC1), sc-81599 (HDAC2), sc-17795 (HDAC3), sc-11405 (HDAC8), sc-81325 (MTA3), sc-100908 (TBL1XR1), sc-81082 and sc-166296 (DNTTIP1) and sc47778 (β-actin) from Santa Cruz; 05-814 (HDAC2), 05-813 (HDAC3), 07-505 (HDAC8) from Millipore; ab46985 (HDAC1), ab3479 (Sin3A), ab70039 (DNTTIP1); NB100-40825 (EHMT2), NB100-81655 (TRERF1) and NB100-81654 (TRERF1) from Novus Biologicals; ab5188 (CDYL) and ab61236 (H4-AcK5) from Abcam; H-3034 (HDAC-3) and T-6793 (Ac-tubulin) from Sigma; H00009425-M02 (CDYL) and H00010013-M01 (HDAC6) from Abnova; no. 2184 (LSD1) from Cell Signaling Technologies; and 382158 (H3AcK9/K14) from Calbiochem. Secondary antibodies labeled with IRDye 680 and IRDye 800 were from LICOR, and antibodies labeled with Alexa 480 and Alexa 594 were from Invitrogen. The HDAC activity assay was purchased from ActiveMotif. Reference compounds were purchased from the following suppliers: vorinostat (SAHA), belinostat (PXD-101), dacinostat (LAQ-824), panobinostat (LBH-589), PCI-24781, and entinostat (MS-275) from Selleck; trichostatin A, PCI-34051, bufexamac and apicidin from Sigma; scriptaid and tacedinaline (CI-994) from Tocris; MC-1293 and BML-210 from Enzo Life Sciences; MGCD-0103 from Chemietek; romidepsin (FK-228) from ACC Corp; and valproic acid from Calbiochem. All other compounds were synthesized as described in Supplementary Synthetic Procedures. Cell culture. Jurkat E6.1, HL60, Ramos and HeLa cells were purchased from American Type Culture Collection; K562 cells were purchased from DSMZ. Jurkat E6.1 cells were cultured in RPMI1640 supplemented with 4.5 g/l glucose, 10 mM HEPES, 1 mM sodium pyruvate and 10% FCS. Ramos cells were cultured in RPMI1640 containing 10% FCS. K562 cells were cultured in RPMI medium containing 10% FCS. Cells were expanded to maximal 1 × 106 cells/ml. HeLa cells were cultured in minimum essential media (MEM) supplemented with 1 mM pyruvate, 0.1 mM nonessential amino acids and 10% FCS. For indirect immunofluorescence assays, the FCS content was reduced to 2%. For cell cycle arrest of HeLa cells in G1/S phase or mitosis, cells were treated for 16 h with 15 µg/ml aphidicolin (Sigma) or with 0.3 µM nocodazole (Sigma). Control HeLa cells were treated with DMSO for 16 h. Preparation of cell lysates. Frozen cell pellets were homogenized in lysis buffer (50 mM Tris-HCl, 0.8% Igepal-CA630, 5% glycerol, 150 mM NaCl, 1.5 mM MgCl2, 25 mM NaF, 1 mM sodium vanadate, 1 mM DTT, pH 7.5). One complete EDTA-free protease inhibitor tablet (Roche) per 25 ml was added. The sample was dispersed using a Dounce homogenizer, kept rotating for 30 min at 4 °C and spun for 10 min at 20,000g at 4 °C. The supernatant was spun again for 1 h at 145,000g. The protein concentration was determined by Bradford assay (BioRad), and aliquots were snap frozen in liquid nitrogen and stored at −80 °C.

c ompound collections from Asinex (http://www.asinex.com/chemsearch.html) and Enamine (http://www.enamine.net/), using a training set of 140 known HDAC inhibitors. Competition binding assays using the SAHA matrix were done essentially as described above but adapted to a 96-well format. We used 1mg of cell lysate and 5 µl of beads per well. Compounds from the screening library including reference compounds as standards were added at 20 µM and 100 µM final concentration from 50× DMSO stocks. Each plate contained eight positive (TSA, 50 µM) and eight negative controls (2% DMSO). Beads were eluted in SDS sample buffer (100 mM Tris pH 7.4, 4% SDS, 20% glycerol, 0.01% bromophenol blue, 50 mM DTT) and spotted in duplicate on nitrocellulose membranes (600 nl/spot) using an automated liquid dispenser (Fluid). After drying, the membranes were rehydrated in 20% ethanol, and processed for detection with specific antibodies as indicated. Spot intensities were quantified using a LiCOR Odyssey scanner and percentage inhibition was calculated using positive and negative controls as 100% and 0% inhibition, respectively. Quantitative co-IP. Antibodies were tested for suitability in co-IP assays by immunoprecipitation-western blot analysis procedures. For western blot analysis we used a LI-COR Odyssey System. Suitable antibodies (40–100 µg) were coupled to 100 µl AminoLink resin (Thermo Fisher Scientific). Cell lysate samples (10 mg) were incubated with prewashed immuno resin on a shaker for 2 h at 4 °C. Beads were washed in lysis buffer containing 0.4% Igepal-CA630 and lysis buffer without detergent. Bound proteins were eluted in 100 µl 2× SDS sample buffer. Protein samples were reduced, alkylated and separated by SDS-PAGE. To provide a specificity control for quantitative LC-MS analysis, IgG from the same species was used in an analogous ‘mock IP’ carried out in parallel from an aliquot of the same lysate sample. Typically, four IP reactions, which were subsequently combined in a single iTRAQ sample for MS/MS analysis, were done in parallel, two with different antibodies directed against the same (or different) antigen(s) and two ‘mock IP’ samples. Enzymatic deacetylation assays. The enzyme activity of purified recombinant HDAC1, HDAC2, HDAC3-NCoR and HDAC6 (1 µg of protein per well) was measured using the Active Motif HDAC Assay Kit in 96-well format following the manufacturer’s instructions. Fluorescence measurements (340 nm excitation/460 nm emission) were recorded with an Analyst HT plate reader (Molecular Devices) in triplicates. Each time series was performed in duplicate. For the determination of enzymatic activity in endogenous HDAC complexes, cell extract samples (375 µl each at 4 mg/ml protein concentration) were prepared from either nocodazole- or vehicle-treated HeLa cultures and were incubated with 15 µg of each antibody for 2 h at 4 °C. Subsequently, 60 µl Protein G beads equilibrated in lysis buffer containing 0.4% NP40 were added and, following incubation for 1 h at 4 °C, beads were washed twice with lysis buffer followed by 10 volumes of Assay buffer (HDAC Assay Kit). We added 10 µl of beads per well and deacetylation activity was determined as described above after 150 min in the absence or presence of 10 µM trichostatin A. Each point was measured in triplicate. Statistical significance was assessed by unpaired Student’s t-test.

Proteomics-based inhibitor profiling. Each inhibitor profiling experiment is denoted by an experiment identifier (internal diameter) number. A list of all profiling experiments is provided in Supplementary Table 5 including all experimental parameters like preincubation times, inhibitor concentrations, isobaric tagging schemata and MS method used. Affinity profiling assays were carried out as described previously21 with minor modifications. Derivatized sepharose beads (35 µl beads per sample) were equilibrated in lysis buffer and incubated with 1 ml (5 mg protein) cell lysate, which had been preincubated with test compound or vehicle for 45 min, on an end-overend shaker for 1 h. Incubation was done at 4 °C for all compounds. In addition, experiments were also performed at 22 °C for the aminobenzamide compounds (tacedinaline, entinostat, BML-120 and mocetinostat). In some experiments different preincubation times (0 to 240 min) or temperatures (4 °C or 22 °C) were tested, to assess the influence of binding kinetics on the selectivity profiles. Beads were transferred to disposable columns (MoBiTec), washed with lysis buffer containing 0.2% NP-40 and eluted with 50 µl 2× SDS sample buffer. Proteins were alkylated with 200 mg/ml iodoacetamide for 30 min, separated on 4–12% NuPAGE (Invitrogen), and stained with colloidal Coomassie.

Sample preparation for MS. Gels were cut into slices across the entire separation range and subjected to in-gel digestion21. For acquisition of doseresponse inhibitor data in one single multiplexed run, TMT (Thermo-Fisher Scientific) tags were used because they allow the acquisition of six-point data. For immunoaffinity purifications, a maximum of four samples was compared, and iTRAQ reagents (Applied Biosystems) were used for reasons of economy and coverage45. Peptide extracts were labeled with iTRAQ or TMT in 40 mM triethylammoniumbicarbonate, pH 8.53. After quenching of the reaction with glycine, labeled extracts were combined. For compound profiling experiments extracts from vehicle-treated samples were labeled with TMT reagent 131, and combined with extracts from compound-treated samples labeled with TMT reagents 126–130, fractionated using reversed-phase chromatography at pH 12, dried and acidified before LC-MS/MS analysis.

Compound screening. The compounds for screening were selected either for their potential to be zinc chelators or based on a similarity search of the

LC-MS/MS analysis. Samples were dried in vacuo and resuspended in 0.1% formic acid in water and aliquots of the sample were injected into a

nature biotechnology

doi:10.1038/nbt.1759

© 2011 Nature America, Inc. All rights reserved.

nano-LC system (Eksigent 1D+) coupled to LTQ-Orbitrap mass spectrometers (Thermo-Finnigan). Peptides were separated on custom 50 cm × 75 µM (internal diameter) reversed-phase columns (Reprosil) at 40 °C. Gradient elution was performed from 2% acetonitrile to 40% acetonitrile in 0.1% formic acid over 2–3 h. LTQ-Orbitrap XL and Orbitrap Velos instruments were operated with XCalibur 2.0/2.1 software. Intact peptides were detected in the Orbitrap at 30.000 resolution. Internal calibration was performed using the ion signal from (Si(CH3)2O)6H+ at m/z 445.120025 (ref. 46). Data-dependent tandem mass spectra were generated for up to six peptide precursors using a combined CID/HCD approach47 or using HCD at a resolution of 7,500 for histone modification data. For CID, up to 5,000 ions (Orbitrap XL) or up to 3,000 ions (Orbitrap Velos) were accumulated in the ion trap within a maximum ion accumulation time of 200 msec. For HCD, target ion settings were 50,000 (Orbitrap XL) and 25,000 (Orbitrap Velos), respectively. Peptide and protein identification. Mascot 2.0 (Matrix Science) was used for protein identification using 10 p.p.m. mass tolerance for peptide precursors and 0.8 Da (CID) or 20 mDa (HCD) tolerance for fragment ions. Carbamidomethylation of cysteine residues and iTRAQ/TMT modification of lysine residues were set as fixed modifications and S,T,Y phosphorylation, methionine oxidation, N-terminal acetylation of proteins and iTRAQ/TMT modification of peptide N termini were set as variable modifications. The search database consisted of a customized version of the International Protein Index database combined with a decoy version of this database created using a script supplied by Matrix Science. Unless stated otherwise, we accepted protein identifications as follows: (i) for single spectrum to sequence assignments, we required this assignment to be the best match and a minimum Mascot score of 31 and a 10× difference of this assignment over the next best assignment. Based on these criteria, the decoy search results indicated <1% false-discovery rate (FDR); (ii) for multiple spectrum to sequence assignments and using the same parameters, the decoy search results indicate <0.1% FDR. For protein quantification a minimum of two sequence assignments matching to unique peptides was required. FDR for quantified proteins was <<0.1%. For analysis of histone peptides additional variable modifications considered were acetylation of lysine, mono- di- and trimethylation of lysine, mono- and dimethylation of arginine. Localization of post-translational modifications on histone peptides was validated by remapping the de-isotoped and de-convoluted tandem MS spectra to b and y ions expected from the peptide hit within 20 p.p.m. mass accuracy. Only spectrum to sequence matches for which fragment ions support localization on only one amino acid and that yielded an H score > 6 indicating 99% confidence33 were considered for further analysis29. For identification of peptides with acetylated lysine the presence of the 126.0913 immonium ion was required. Peptide and protein quantification. Centroided iTRAQ/TMT reporter ion signals were computed by the XCalibur software operating and extracted from MS data files using customized scripts. Only peptides unique for identified proteins were used for relative protein quantification and are referred to in Supplementary Data Sets 1–6 as unique peptide assignments (UPA). Further, quantification spectra matching to UPAs were filtered according to the following criteria: mascot ion score > 15, signal to background ratio of the precursor ion > 4, signal to interference > 0.5 (ref. 29). Reporter ion intensities were multiplied with the ion accumulation time yielding an area value proportional to the number of reporter ions present in the mass analyzer. For compound competition binding experiments, fold-changes are reported based on reporter ion areas in comparison to vehicle control and were calculated using sum-based bootstrap algorithm. Fold-changes were corrected for isotope purity as described and adjusted for interference caused by co-eluting nearly isobaric peaks as estimated by the signal-tointerference measure29. The heat maps in Supplementary Figure 2a–d are based on the accumulated reporter ion responses for each protein divided by its molecular weight (10 percentile bins). This enables accurate relative quantification between different conditions for the same protein and similarly to spectrum count based methods gives an estimation of the relative abundance across different proteins48. Fractional abundance was calculated by the reporter ion response in condition i divided by the summed reporter ion response across all conditions:

doi:10.1038/nbt.1759

FA(i) =

A(i)

n

∑ j =1 A( j)

Dose-response curves were fitted using R (http://www.r-project.org/) and the drc package (http://www.bioassay.dk), as described previously21. IC50 values were confirmed in replicate experiments using targeted data acquisition for a subset of proteins29. To compare selectivities of compounds displaying different absolute potencies relative potencies were calculated as (pIC50 – min(pIC50))/(max(pIC50) – min(pIC50)) for each experiment. Apparent dissociation constants (Kdapp) were derived from IC50 values as described23 and relative affinities were determined as described for relative potencies. For the assessment of data robustness (Supplementary Fig. 4), seven replicate samples of trichostatin A treated samples were analyzed as previously described. IC50 values for 21 proteins were calculated for each experiment and normalized using the average pIC50 determined in each experiment. Then average and s.d. of the pIC50 for each protein across the seven samples were calculated. A pair-wise, two sided t-test was performed between pIC50 values of each protein with those of every other protein. In addition P-values, average pIC50, and differences between pIC50 were calculated after combining the proteins into complexes. For immunoprecipitations, the enrichment E was calculated as (A(IP) – A(mock IP))/(A(IP) + A(mock IP)) and scales between −1 and 1. ‘A’ represents the summed-up reporter ion response for the protein of interest. When immunoprecipitation experiments against two different bait proteins (IP1, IP2) were analyzed in a single iTRAQ experiment, relative enrichment (Fig. 3a) was calculated as RE(IP1) = (A(IP1) – A(mock IP))/(A(IP1) + A(IP2) + A(mock IP)). RE of 0.5 means that the protein was precipitated in both IPs equally well if no signal was detected in the mock IP. C score–based determination of complex components using compound profiling and immunoprecipitation data sets. For each protein identified in an immunoprecipitation experiment against one of the bait proteins (7,000 unique bait/protein pairs) enrichment was normalized to values between 0 and 1. Similarly average pIC50s retrieved from compound profiling experiments were normalized to values between 0 and 1 (5,683 unique proteins). The average normalized pIC50 (anpIC50) value was calculated across all compounds included for each protein. The anpIC50 of each protein is linked to the normalized enrichment (nPD) values determined in each immunoprecipitation experiment. The C-score for the 6,263 unique bait/protein combinations observed in compound profiling and immunoprecipitation experiments was calculated as the sum of squares of the anpIC50 and the nPD values. The resulting value we dubbed C-score and it scales from 0 to 2. Similarly, C-score values for individual compound/bait pairs were calculated. We tested the discrimination power of the C-score by scrambling all anpIC50 values and reassigning them to the 5,683 unique proteins. After that the linking of competition data and the pull-down data (again a total of 6,263 matches) as well as the C-score calculation were performed exactly as above. In a next step we used this randomized C-score data set to determine the significance threshold for separating random hits from experimental databases on a cumulative FDR. For each C-score value X, the ratio of number of bait/protein pairs from the random matching and from the experimental matching that have a C-score ≥ X were calculated. This enabled us to determine the subset of high confidence interactions (FDR < 0.05: C-score > 1.14; FDR < 0.15: C-score >1.0 and max(c-score) > 1.2) (Supplementary Fig. 9). Heat map generation. Heat maps and t-tests were performed using the R-package and Tableau. For unbiased hierarchical clustering of compound profiling data, all quantified proteins identified with at least 9 independent experiments were considered. Protein and compound clustering was based on relative affinities averaged over replicate experiments using the Euclidean distance measure and the complete linkage method provided in R. Only those proteins were considered for which binding to the SAHA matrix was inhibited by at least one compound (that is, an IC50 value was determined) and that were identified in experiments related to more than half of the tested compounds. The significance of observed clusters was tested with the R package pvclust, which calculates two P-values for each cluster on the basis of a bootstrap resampling

nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

technique: approximately unbiased P-value (AU) and bootstrap probability value (BP)49. Clusters were reported as significant for AU > 95 indicating 95% probability of not being a random cluster. Cell-based assays. For cell-based protein acetylation assays, HeLa or K562 cells (96-well format, 5 × 104 cells per well) were treated with compounds for 6 h. Cells were washed with cold PBS and lysed directly in SDS-sample buffer, followed by denaturation at 95 °C. Lysates (10 µl) were resolved on SDS-gels, transferred to PVDF membranes and analyzed for tubulin and histone acetylation by immune-detection using IRDye-labeled secondary antibodies and an Odyssey scanner (LiCOR). Data analysis of the quantified bands was performed using Excel and GraphPad Prism. For the IFN-α secretion assay, human peripheral blood mononuclear cells were isolated with Histopaque 1077 (Sigma) from fresh human donor blood and were plated at a concentration of 0.5 × 106 cells/ml. Cells were incubated with test compounds for 45 min before stimulation with plasmacytoid dendritic cell–specific TLR9 agonist ODN2216 (Invivogen). IFN-α released into cell supernatants was measured in triplicate after 16 h by Flex-Set IFNalpha (BD Biosciences) by flow cytometry (FACSCalibur, BD Biosciences). For all cell-based assays viability was assessed in parallel (MTT kit, Roche). For histone modification experiments 1E6 K562 cells were treated with HDAC inhibitors for 6 h. After cell harvest, samples were subjected to histone enrichment as previously described50 and separated using SDS gel electrophoresis. For indirect immunofluorescence analysis, HeLa cells were plated to subconfluency on polylysine-coated glass chamber slides and treated after recovery

nature biotechnology

with the indicated compounds for 4 h. Samples were fixed in cold methanol, permeabilized with Triton-X100, blocked in 1% BSA and treated for immunodetection of acetylated tubulin and acetylated histone H3. Cells were counterstained for nucleic acids using 4′,6-diamidino-2-phenylindole (DAPI). Multichannel fluorescence microscopy was performed on an Olympus IX70 microscope. Images were acquired using a monochrome CCD camera (CoolSNAP HQ Digital) and analyzed with MetaMorph (Universal Imaging Corporation). The instrument was adjusted to ensure proper comparison of levels of acetylated tubulin and acetylated histone H3 before and after inhibitor treatment.

45. Thingholm, T.E., Palmisano, G., Kjeldsen, F. & Larsen, M.R. Undesirable chargeenhancement of isobaric tagged phosphopeptides leads to reduced identification efficiency. J. Proteome Res. 9, 4045–4052 (2010). 46. Olsen, J.V. et al. Parts per million mass accuracy on an Orbitrap mass spectrometer via lock mass injection into a C-trap. Mol. Cell. Proteomics 4, 2010–2021 (2005). 47. Kocher, T. et al. High precision quantitative proteomics using iTRAQ on an LTQ Orbitrap: a new mass spectrometric method combining the benefits of all. J. Proteome Res. 8, 4743–4752 (2009). 48. Sanders, S.L., Jennings, J., Canutescu, A., Link, A.J. & Weil, P.A. Proteomics of the eukaryotic transcription machinery: identification of proteins associated with components of yeast TFIID by multidimensional mass spectrometry. Mol. Cell. Biol. 22, 4723–4738 (2002). 49. Suzuki, R. & Shimodaira, H. Pvclust: an R package for assessing the uncertainty in hierarchical clustering. Bioinformatics 22, 1540–1542 (2006). 50. Shechter, D., Dormann, H.L., Allis, C.D. & Hake, S.B. Extraction, purification and analysis of histones. Nat. Protoc. 2, 1445–1457 (2007).

doi:10.1038/nbt.1759

letters

Generation of anterior foregut endoderm from human embryonic and induced pluripotent stem cells

© 2011 Nature America, Inc. All rights reserved.

Michael D Green1, Antonia Chen1, Maria-Cristina Nostro2, Sunita L d’Souza1, Christoph Schaniel1, Ihor R Lemischka1, Valerie Gouon-Evans1, Gordon Keller2 & Hans-Willem Snoeck1 Directed differentiation of human embryonic stem (hES) cells and human induced pluripotent stem (hiPS) cells captures in vivo developmental pathways for specifying lineages in vitro, thus avoiding perturbation of the genome with exogenous genetic material. Thus far, derivation of endodermal lineages has focused predominantly on hepatocytes, pancreatic endocrine cells and intestinal cells1–5. The ability to differentiate pluripotent cells into anterior foregut endoderm (AFE) derivatives would expand their utility for cell therapy and basic research to tissues important for immune function, such as the thymus; for metabolism, such as thyroid and parathyroid; and for respiratory function, such as trachea and lung. We find that dual inhibition of transforming growth factor (TGF)-b and bone morphogenic protein (BMP) signaling after specification of definitive endoderm from pluripotent cells results in a highly enriched AFE population that is competent to be patterned along dorsoventral and anteroposterior axes. These findings provide an approach for the generation of AFE derivatives. Directed differentiation of pluripotent stem cells into a variety of cell types opens a promising avenue for cell replacement therapy and provides a powerful tool for basic research4. ES cells are derived from the inner cell mass of the blastocyst and can be maintained in a pluri potent state by defined conditions. Furthermore, adult somatic cells can be reprogrammed into a pluripotent state (hiPS cells), paving the way for the generation of patient-specific pluripotent cells6. Most efforts at generating endoderm derivatives from human pluripotent cells have focused on midgut (pancreatic endocrine cells) and posterior foregut (hepatocyte) cell types 1–4,7. However, several clinically relevant cell types originate from AFE, the most rostral aspect of the endoderm. The caudal region of the AFE gives rise to trachea and lung8. The generation of respiratory tissue may allow cellular replacement therapies for respiratory disease. Rostral to the lung field is the pharyngeal endoderm, which forms four paired outcroppings, called pharyngeal pouches9. These develop into spe cific organs: eustachian tube and tympanic membrane (first pouch), palatine tonsils (second pouch), thymus (anterior third pouch), par athyroids (dorsal third and fourth pouch) and parafollicular C cells of

the thyroid (fourth pouch). The thyroid body develops from the floor of the pharynx9. Several of these organs are excellent candidates for cellular replacement therapy. The thymus is the site of production of T lymphocytes10. Thymic function is severely affected by allogeneic hematopoietic stem cell transplantation, because of both pretrans plant conditioning regimens and post-transplant graft-versus-host disease, leading to profound defects in T-cell reconstitution11. As the thymus involutes with age, older transplant recipients would particularly benefit from thymic replacement therapy. In addition, the thymus is congenitally absent in nude/severe combined immuno deficient (SCID) and DiGeorge syndromes10. Moreover, autoimmune, congenital or acquired hypothyroidism and genetic or iatrogenic hypoparathyroidism may be treatable by replacement with tissues derived from ES or iPS cells. Efforts at generating AFE-derived cells have met with little success, although upregulation of thymus12,13, thyroid14, parathyroid15 or lung markers16 in differentiating ES cells either after specification of defini tive endoderm or in mixed-lineage embryoid bodies has been reported. However, these reports relied on drug selection, did not quantify the efficiency of induction or did not determine the presence of alternative lineages. Because cell fate is established through sequential and increas ingly lineage-restricted progenitors4,8, we hypothesized that the failure to specify these cell types with high efficiency in vitro is due to the inability to induce AFE from endoderm. Therefore, we developed a strategy to generate AFE from human pluripotent cells. Definitive endoderm, one of the three germ layers of the embryo proper, is induced from ES cells by high concentrations of activin A, mimicking nodal signaling during gastrulation6. Examination of this process in the hES cell line HES2 by quantitative PCR revealed a transcriptional cascade in which the primitive streak marker MIXL1 and then the endodermal transcription factors SOX17 and FOXA2 are upregulated (Fig. 1a)1,2,7,8,17. After 4 d of exposure to activin A, >95% of the cells expressed the definitive endoderm mark ers CXCR4, c-KIT and EPCAM (Fig. 1b)17. After gastrulation, the definitive endoderm forms a tube with distinct anteroposterior axis identity8. Within definitive endoderm, the pluripotency marker SOX2 reemerges as a foregut marker, whereas CDX2 identifies hindgut18. After activin A removal at day 5 of culture, we observed an increase in both CDX2 and SOX2 expression (Fig. 1a), suggesting the generation

1Department of Gene and Cell Medicine and Black Family Stem Cell Institute, Mount Sinai School of Medicine, New York, New York, USA. 2Division of Stem Cell and Developmental Biology and McEwen Centre for Regenerative Medicine, Ontario Cancer Institute, Toronto, Ontario, Canada. Correspondence should be addressed to H.-W.S. ([email protected]).

Received 5 November 2010; accepted 25 January 2011; published online 27 February 2011; doi:10.1038/nbt.1788

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

267

7

8

b 96.1%

5

c-KIT

3

10

102 0

10

3

10

102 0 0 102 103 104 105

0 102 103 104 105

CXCR4

CXCR4

SOX2 0.6

*

2.0

PAX9

*

0.5 0.4

P3 R sF

G

N

24 1– ns

*

1.0

PAX6 1.25

*

0.8

1.00

0.6

0.75

0.4

0.50

0.2

0.25

0

0

mRNA (ratio to β-actin)

tio di

BRACHYURY

d1

d5

G

G G

N

d5

O N d5

G

R sF

IN /S B G + IN sF /S R B P3

0 P3

0.1

0

IN /S O B G + GIN sF /S R B P3

0.2

0.5

O

1.0

N

*

0.3

d5

1.5

e mRNA (ratio to β-actin)

2.5

O

d

97.6%

5

10

4

104

mRNA (ratio to β-actin)

1.0 0.5 d0 d5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

*

2.0

SOX2

1.5 1.0 0.5 0 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

d0 d5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

CDX2

d0 d5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

0.6

*

0.5

PAX9

0.4 0.3 0.2 0.1 0

d0 d5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

0.6

*

0.5

TBX1

0.4 0.3 0.2 0.1 0

d0 d5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

CREB313

CEBPA

TBX1

SB NO -4 GG 31 IN 54 / 2 co H nd ep iti ati on c s

SB NO -4 GG 31 IN 54 / 2 co H nd ep iti ati on c s

SB NO -4 GG 31 IN 54 / 2 co H nd ep iti ati on c s

SB NO -4 GG 31 IN 54 / 2 co H nd ep iti ati on c s

mRNA (ratio to β-actin)

SB NO -4 GG 31 IN 54 / 2 co H nd ep iti ati on c s

mRNA (ratio to β-actin)

SB NO -4 GG 31 IN 54 / 2 co He nd pa iti tic on s

SB OG -4 G 31 IN 54 / 2 co Hep nd at iti ic on s

N

N SB OG -4 G 31 IN 54 / 2 co Hep nd at iti ic on s

268

SB NO -4 GG 31 IN 54 / 2 co H nd ep iti ati on c s

f

B Y- MP 27 4 63 (1 2 ng (µ /m M l) )

2 0.25 * 0.2 Figure 1 Induction of AFE 0.20 markers in NOGGIN/SBNOGGIN/SB-431542 0.15 431542-treated definitive 1 0.1 0.10 endoderm. (a) Expression * or 0.05 of FOXA2, MIXL1, SOX17, * 0 0 0 Hepatic conditions SOX2 and CDX2 mRNA during 1 2 3 4 5 6 7 8 9 activin A–mediated induction Time (days) of definitive endoderm in hES cells. Data expressed ODD1 CDX2 EVX1 PAX9 SOX2 FGF8 1.00 1.25 0.3 1.25 * 2 0.5 as quantification of mRNA * * 1.00 1.00 0.4 0.75 normalized to β-ACTIN (also 0.2 0.75 0.75 0.3 known as ACTB), scaled 0.50 1 0.50 0.50 0.2 0.1 proportionally to maximum 0.25 0.25 0.25 0.1 induction. Cytokines were added * * * 0 0 0 0 0 0 as indicated on top of the figure (bar). (b) Representative flow cytometric analysis of definitive endodermal markers CXCR4, C-KIT and EPCAM at day 5 of activin A induction. Two biologically independent experiments are shown. (c) Expression of FOXA2, SOX2, CDX2, PAX9 and TBX1 mRNA on day 9 in cultures treated on day 5 after induction of definitive endoderm (see upper left panel), with the factors listed in the lower left panel (n = 3 biological replicates; *, significantly different from all other conditions, P < 0.0001; one-way ANOVA). d0, prior to start of differentiation; d5, day 5. (d) Expression of SOX2 and PAX9 on day 9 in cultures treated on day 5, after induction of definitive endoderm, with NOGGIN/SB-431542 (SB) in the presence or absence of sFRP3 (*, P < 0.05, n = 3 biological replicates). (e) Expression of BRACHYURY and PAX6 mRNA at day 9 in hES cells differentiated as previously described to neurectoderm (day 1 addition of NOGGIN/SB-431542), or after induction of endoderm (endoderm induction until day 5, followed by addition of NOGGIN/SB-431542). For BRAYCHURY, day 3.5 hES cells exposed to activin A and undergoing gastrulation served as a positive control (*, P < 0.0001, n = 3 experiments consisting each of three biological replicates). (f) Expression of ODD1, CDX2, EVX1, CREB313, CEBPA, TBX1, PAX9, SOX2 and FGF8 mRNA in day 9 cultures treated in parallel with either NOGGIN/SB-431542 or cultured in hepatic conditions after induction of definitive endoderm until day 5 (n = 3 experiments consisting each of three biological replicates). mRNA (ratio to β-actin)

© 2011 Nature America, Inc. All rights reserved.

EPCAM

10

mRNA (ratio to β-actin)

4 5 6 Time (days)

1.5

0

mRNA (ratio to β-actin)

3

FOXA2

2.0

2.5

mRNA (ratio to β-actin)

2

G IN /S B O d1 G a G c IN t. /S A B

1

O G A G IN B ctiv d5 /S bF MP in B N A 4 G O d1 F (0 (1 G a (1 .5 00 G 0 ng n IN ct. ng /m g/ /S A /m l) ml d3 B ) l) .5 ac t. A

0

Concentration N/A N/A N/A 10 µM 300 ng/ml, 50 ng/ml 200 ng/ml 200 ng/ml 10 µM 50 µM 0.1 µM 1 µM 10 µM 200 ng/ml, 200 ng/ml 100 ng/ml 50 ng/ml 10 ng/ml, 5 ng/ml 50 ng/ml 500 ng/ml 200 ng/ml, 10 µM 50 ng/ml 10 ng/ml 50 ng/ml 1 ng/ml 10 ng/ml 10 ng/ml 200 ng/ml, 10 µM

9

N

0.2

8

G

0.4

Factors ES cells Day 5 act. A induction Nothing Rock Inhibitor (Y-27632) Cerbereus + Lefty NOGGIN Dkk SB-431542 DEAB ATRA ATRA ATRA NOGGIN + Dkk WNT3a BFGF BMP4 + BFGF FGF10 FGF10 NOGGIN + SB-431542 Wnt5a FGF8b FGF8b FGF4 FGF4 BMP4 NOGGIN + cyclopamine

7

O

0.6

5 6 Time (days)

N

0.8

Condition d1 d5 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

4

d5

SOX17 SOX2 CDX2

3

mRNA (ratio to β-actin)

2

d1

1

2.5

C on

B Y- MP 27 4 63 (1 2 ng (1 /m 0 l) µM )

B Y- MP 27 4 63 (1 2 ng (1 /m 0 l) µM ) Ac BM t. bF P A ( G 4 ( 10 F 0 0 (1 .5 ng 0 ng /m ng /m l /m l) ) l) S di eru m ffer med e fre ia ntia e tio n

FOXA2 MIXL1

1.0

c

N

Percent of maximum

a

A B ct. bF MP A ( G 4 ( 10 F 0 0 (1 .5 ng 0 ng /m ng /m l /m l) ) l)

letters

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

letters a

DAPI

SOX2

DAPI/SOX2

DAPI

© 2011 Nature America, Inc. All rights reserved.

DAPI

b

FOXA2

Figure 2 Immunofluorescence analysis of NOGGIN/SB-431542-treated definitive endoderm. (a) Immunofluorescence for FOXA2, SOX2, CDX2, PAX9, FOXG1 and TBX1 of day 9 in HES2 definitive endoderm cultures treated on day 5 with NOGGIN/SB-431542. Scale bar, 50 µm (upper); 100 µm (lower). (b) Expression of FOXA2 and SOX2 in HFD9 hiPS cultures in the same conditions. Scale bar, 25 µm.

SOX2/FOXA2

DAPI/PAX9

TBX1

DAPI/FOXG1

DAPI/TBX1

DAPI

SOX2

FOXA2

DAPI/SOX2

DAPI/FOXA2

FOXA2/SOX2

of a mixture of anterior and posterior definitive endoderm. Therefore, we examined which signals added after induction of definitive endo derm favored anterior (SOX2+) and suppressed posterior (CDX2+) endoderm generation. Following the generation of a CXCR4+EPCAM+ population in embryoid bodies exposed to activin A, the embryoid bodies were dissociated and plated as a monolayer. We tested the addition of 24 combinations of morphogens and inhibitors at day 5 (Fig. 1c), and used expression of FOXA2, SOX2, CDX2, TBX1 (endoderm anterior to the stomach)9,18,19 and PAX9 (pharyngeal endoderm)10,19 as read outs of cellular identity at day 9 of culture. Only in the combined presence of NOGGIN, a physiological inhibitor of BMP signaling, and SB-431542, a pharmacological inhibitor of activin A/nodal and TGF-β signaling, was SOX2 expression induced, CDX2 expression suppressed and FOXA2 expression maintained. Furthermore, only this condi tion induced strong expression of TBX1 and PAX9 (Fig. 1c). During the activin A–induction stage, cell number increased 4.5- ± 1.9-fold, and during the NOGGIN/SB-431542 stage, the cells expanded another 1.4- ± 0.4-fold. Notably, NOGGIN/SB-431542 treatment was equally potent in two hiPS cell lines (HDF2 and HDF9), with induction of SOX2, PAX9 and TBX1 (Supplementary Fig. 1). Multiple FGF family members and WNT3a, consistent with their functions in development5,7,20, posteriorized definitive endoderm, as shown by increased CDX2 expression (Fig. 1c). However, WNT antago nism through addition of soluble Frizzled-related protein 3 (sFRP3) was not sufficient to induce SOX2 (Fig. 1c). Furthermore, sFRP3 did not synergize with NOGGIN/SB-431542, and even appeared nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

detrimental for the induction of PAX9 and SOX2 (Fig. 1d). The timing of the addition of NOGGIN/SB-431542 was critical, as only treat ment immediately after the generation of a uniform CXCR4+c-KIT+ or CXCR4+EPCAM+ population at day 5/6 induced a SOX2+FOXA2+ population at day 9. Earlier administration abrogated gastrulation, and later administration failed to downregulate the posterior marker CDX2 (data not shown). FOXA2 is also expressed in the notochord (mesoderm), and FOXA2 and SOX2 are co-expressed by the hindbrain floorplate (neurecto derm)21,22. Furthermore, direct application of NOGGIN/SB-431542 to hES cells without prior endoderm induction by activin A leads to a neuroectodermal fate23. Therefore, we assayed for the presence of these alternative fates. As expected, the neuroectodermal marker PAX6 was expressed in cultures where NOGGIN/SB-431542 was added at day 1, whereas BRACHYURY, a marker of the notochord and of gastrulating cells, was expressed during early endoderm induction (Fig. 1e). Neither BRACHYURY nor PAX6 were expressed in defini tive endoderm exposed to NOGGIN/SB-431542 (Fig. 1e), indicating that NOGGIN/SB-431542 treatment of activin A–induced definitive endoderm specifies only AFE. To further assess whether the NOGGIN/SB-431542-induced endodermal cells were distinct from previously described endodermal lineages, we compared day 9 NOGGIN/SB-431542-treated cultures with day 9 cultures grown under conditions favoring a hepatic (poste rior foregut) fate. The latter has been previously shown to require BMP4 and bFGF after activin A induction of endoderm2. The expression of CDX2, the hindgut marker EVX1, the liver markers CREB313 and CEBPA, as well as ODD1, a stomach domain marker18, was higher in the ‘hepatic’ conditions than in the NOGGIN/SB-431542 conditions, and the reverse was true for the anterior markers TBX1, PAX9, SOX2 and FGF8, a marker within the endoderm specific for pharyngeal pouch endoderm24 (Fig. 1f). Therefore, NOGGIN/SB-431542 treatment specifies AFE cells that are distinct from those specified in hepatic conditions. Application of NOGGIN/SB-431542 to activin A–induced defini tive endoderm yielded colonies of densely packed cells surrounding an empty lumen-like or cyst-like opening. More than 90% of the cells were found in such colonies if plated at high density. Virtually all cells co-expressed SOX2 and FOXA2 (Fig. 2a, for HES2 cells, Fig. 2b for HDF9 iPS cells, Supplementary Fig. 2a for HDF9 and HDF2 hiPS cells, and Supplementary Fig. 2b for HES2 cells in Matrigel culture), although rare cells expressed only FOXA2 (Fig. 2a and Supplementary Fig. 2a, arrows). All colonies stained positive for TBX1, PAX9 and the pharyngeal endoderm marker FOXG1 (Fig. 2a, HES2 cells). The typical colonies observed in NOGGIN/SB-431542-treated cultures were never seen when cells were cultured in media without added factors (Supplementary Fig. 2c). In these conditions, >95% of cells expressed FOXA2, but only rarely were SOX2+FOXA2+ cells observed (Supplementary Fig. 2c, arrows). Colonies with this morphology were also never observed in hepatic conditions (Supplementary Fig. 2d). Comparative immunofluorescence analysis of HES2-derived endodermal cells cultured in parallel in either NOGGIN/SB-431542 or hepatic conditions revealed that only NOGGIN/SB-431542 cul tures were characterized by strong SOX2, PAX9 and TBX1 expression (Supplementary Fig. 2d). 269

B Y- MP 27 4 63 (1 2 ng /m A l) BM ctiv bF P in G 4( A( F 0 1 (1 .5 00 0 ng n ng /m g/ /m l) ml ) l)

letters

d

a

WNT3a (100 ng/ml), FGF10 (10 ng/ml), NOGGIN/ KGF (10 ng/ml) BMP4 (10 ng/ml), EGF (10 ng/ml) SB-431542

1

b

2

3

4

PAX9

AIRE

SFTPC

e DAPI/FOXA2

DAPI/SOX2

NKX2.1 8 6 4 2 0

*

NKX2.5 10 8 6 4 2 0

*

PAX1 14 12 10 8 6 4 2 0

*

12 P63

12.5 10.0 7.5 5.0 2.5 0

13

*

NOGGIN/ WNT3a (100 ng/ml), FGF10 (10 ng/ml), KGF (10 ng/ml) BMP4 (10 ng/ml), SB-431542 EGF (10 ng/ml) (WKFBE)

FOXA2/SOX2

M e SB NO dia -4 GG 31 IN 54 / co H 2 nd ep iti ati on c s

M e SB NO dia -4 GG 31 IN 54 / co H 2 nd ep iti ati on c s

e SB NO dia -4 GG 31 IN 5 / co H 42 nd ep iti ati on c s

mRNA (ratio to β-actin)

Media Figure 3 Functional characteristics of Hepatic NOGGIN/SB-4315428 9 10 11 12 13 1 2 3 4 5 6 7 Time (days) induced AFE cells. 20× NKX2.5 NKX2.1 * PAX1 (a) H/E staining of a * 7 7 Conversion Fold 3 * g Days Factors 6 6 efficiency expansion teratoma derived after 5 5 hiPS/hES 2 1 4 4 BMP4/Y-27632 5 weeks from HES2 3 3 2 1 2 2 transplanted under 4.6 ± 1.9 1 1 Activin A 0 0 0 the kidney capsule + + −/− of NOD/SCIDIl2rg > 90% CXCR4 c-KIT Endoderm 5 NOGGIN/ mice. The three SB-431542 right-hand panels + + 92 ± 2% FOXA2 SOX2 Anterior foregut 7 DAPI NKX2.1 DAPI/NKX2.1 show neurectoderm (neural rosette), 8.9 ± 3.3 endoderm (intestinal WKFBE epithelium) and mesoderm (cartilage), + Ventral anterior 10× 37 ± 6% NKX2.1 13 foregut respectively. Scale bar, 50 µ. (b) H/E staining of a growth arising 5 weeks after transplantation of NOGGIN/SB-431542-induced AFE cells derived from HES2 cells under the kidney capsule of immunocompromised mice. Scale bar, 50 µm. (c) Immunofluorescence analysis of the tissue from b stained for FOXA2, PAX9, AIRE and SFTPC. Scale bar, 50 µm. (d) Expression of SOX2, NKX2.1, NKX2.5, PAX1 and P63 in HES2-derived cells generated in the two conditions schematically represented on top of the panel (n = 6 culture wells from two independent experiments; *, significantly different from NOGGIN/SB431542; P < 0.05) WKFBE: WNT3a, KGF, FGF10, BMP4 and EGF. (e) Expression of NKX2.1, NKX2.5 and PAX1 in HES2-derived cells generated in the three conditions schematically represented on top of the panel (n = 4 to 6 culture wells from three independent experiments, *, significantly different from the other conditions; P < 0.05). (f) Expression of FOXA2 (green) and SOX2 (red) 2 d after treatment of activin A–induced definitive endoderm with NOGGIN/SB-431542 (blue, DAPI). Scale bar, 50 µm. (g) Schematic overview of the efficiency of induction of ventral AFE. WKFBE: WNT3a, KGF, FGF10, BMP4 and EGF. (h) Immunofluorescence for NKX2.1 in differentiated HDF9 hiPS cells after sequential treatment with activin A, NOGGIN/SB431542 and WKFBE according to the scheme in g. Scale bar, 50 µm.

M

© 2011 Nature America, Inc. All rights reserved.

f

*

5

SB NO B -4 G G Y- MP 31 IN 27 4 54 / ( 63 1 2 ng W 2 /m KF Ac l) BE B tiv bF MP in SB NO G 4( A( G 4 F 0 1 31 GIN (1 .5 00 5 / 0 ng n ng /m g/ W 42 KF /m l) ml ) l) BE SB NO G -4 G 31 IN 54 / W 2 KF BE SB NO -4 G G 31 IN 54 / W 2 KF BE SB NO -4 G G 31 IN 54 / W 2 KF BE

c

FOXA2

mRNA (ratio to β-actin)

SOX2 2.5 2.0 1.5 1.0 0.5 0

or NOGGIN/SB-431542 6 7 8 9 10 11 Time (days)

h

Collectively, these expression data show that NOGGIN/SB-431542 specifies a highly enriched population of cells with AFE phenotype in activin A–induced definitive endoderm. These findings are con sistent with the fact that mice null for the BMP antagonist Chordin display anterior truncations25 and with the observation that activin A–induced endoderm contains a large fraction of CDX2+ posterior endoderm (Fig. 1a). To determine the potential in vivo of cells cultured in NOGGIN/ SB-431542 conditions, we transplanted 106 cells under the kidney capsule of NOD/SCIDIl2rg−/− mice. Whereas undifferentiated HES2 cells generated teratomas containing cells derived from all three germ layers (Fig. 3a), NOGGIN/SB-431542-treated cells produced growths lacking identifiable ectodermal or mesodermal elements (Fig. 3b). We observed multiple luminal structures, lined either by pseudostratified epithelium (typical of upper airway epithelium) or a more disorganized epithelium containing one to three layers of nuclei (Fig. 3b). The latter consistently stained for surfactant 270

rotein-C (SFTPC), a marker specific for type II alveolar cells in p the lung (Fig. 3c and Supplementary Fig. 3a). In hES cell–derived teratomas, no SFTPC staining was observed (data not shown). The remainder of the cells stained almost uniformly for FOXA2. However, except in the luminal structures, FOXA2 was confined to the cytoplasm, possibly owing to differentiation into FOXA2− terminal AFE derivatives or to abnormal FOXA2 regulation in a xenograft (Fig. 3c). Islands of cells expressing PAX9, as well as rare regions showing discrete nuclear speckles of AIRE (specific for medullary thymic epithelial cells 10), were also detected (Fig. 3c). In hES-derived teratomas, PAX9 was only observed in zones of cartilage formation and AIRE expression was not observed (data not shown). Collectively, these data suggest that the developmental potential of NOGGIN/SB-431542-induced definitive endoderm is largely limited to AFE derivatives in these conditions. Next, we attempted to further differentiate these cells. AFE under goes dorsoventral patterning, resulting in specification of lung buds, VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

letters Activin A (100 ng/ml) BMP4 (0.5 ng/ml) bFGF (10 ng/ml)

BMP4 (1 ng/ml) Y-27632

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

GATA6

2.5 2.0 1.5 1.0 0.5 0

11

12

13

FOXJ1

2

*

1

W

KF BE KF BE AT R + A

0

W

*

10

KF BE KF B AT E R + A

W

W

FGF8 + SHH WKFBE

FGF8

NOGGIN/ SB-431542

SHH WFKBE WKFBE + ATRA (0.5 µM) WKFBE + ATRA (0.5 µM)

1 2 3 SFTPC

10

19 0

15

11

0.

7 Time (days)

0.

6

0.

5

05

WNT3a + FGF10 + KGF

0

trachea and pharyngeal pouches in response to WNT, BMP and FGF signals from the ventral b mesoderm10,26. At this stage, SOX2 expression remains higher dorsally, whereas NKX2.1 (lung BMP4 Activin A (100 ng/ml) and thyroid field8,26), NKX2.5 (transiently (1 ng/ml) BMP4 (0.5 ng/ml) bFGF (10 ng/ml) Y-27632 expressed in the ventral pharyngeal endo derm27) and PAX1 (within endoderm specifi cally expressed in the pharyngeal pouches28) are specific ventral markers. Extended treat ment with NOGGIN/SB-431542 until day 13 resulted in continued expression of SOX2, sug 1 2 3 4 gestive of a dorsal fate (Fig. 3d) and consistent with the fact that Noggin is expressed dorsally in the AFE, whereas BMP4 is expressed ventrally26. In contrast, replac ing NOGGIN/SB-431542 with WNT3a, KGF, FGF10, BMP4 and EGF (all factors, WKFBE) at day 7 of culture resulted in lower expression of SOX2 and induced the ventral markers NKX2.1, PAX1 and NKX2.5 at day 13 (Fig. 3d for HES cells and Supplementary Fig. 3a,b for HDF2 and HDF9 hiPS cells). Expression of the early thyroid marker, PAX8, was not observed, however, suggesting that NKX2.1 induction is indic ative of commitment to a lung, rather than a thyroid, fate (data not shown). Furthermore, P63, a marker of airway progenitor cells26, was strongly induced (Fig. 3d), and the vast majority of the cells expressed the epithelial marker EPCAM (Supplementary Fig. 3c). Addition of individual factors was not sufficient for this transcriptional induction (data not shown). Furthermore, only prior exposure to NOGGIN/SB431542, and not to the hepatic cocktail or to media alone enabled subsequent upregulation of PAX1, NKX2.1 and NKX2.5 by WKFBE (Fig. 3e), demonstrating that NOGGIN/SB-431542 treatment of activin A–induced definitive endoderm is required for differentiation toward a ventral AFE fate. Timing of the WKFBE ventralization stimulus was critical, as only cultures treated at day 7, but not at day 9, were competent to express NKX2.1, PAX1 and NKX2.5 (data not shown). At this time, 92 ± 2% of the cells were FOXA2+SOXA2+ (Fig. 3f,g). Immunofluorescence revealed that after induction in WKFBE, 37 ± 6% of cells expressed NKX2.1 (Fig. 3h and Supplementary Fig. 3d). During NOGGIN/SB-431542 followed by WKFBE treatment of activin A–induced endoderm, cells expanded an additional 8.95- ± 3.3-fold (Fig. 3g). Thus, NOGGIN/SB-431542-induced AFE is uniquely com petent to respond to ventralization signals in vitro. Exposure of NOGGIN/SB-431542-induced AFE to WKFBE did not result in expression of terminal differentiation markers for thymus, parathyroid, thyroid or lung at day 13 or day 19 of culture (data not shown). As these cells had the potential to give rise to SFTPC+ cells in vivo, we attempted to achieve lung specification in vitro. Consistent with a critical role for retinoic acid in early lung development29, addi tion of retinoic acid to the WKFBE cocktail decreased the expression of the pharyngeal pouch marker PAX1, but increased FOXP2, NKX2.1, GATA6 and FOXJ1, a constellation of markers suggestive of a lung

FOXP2

0.5 0.4 0.3 0.2 0.1 0

9

W

KF B AT E R + A

KF BE KF BE AT R + A

0

*

7 8 Time (days)

W

*

1

NKX2.1

10 8 6 4 2 0

2

6

KF BE KF BE AT R + A

PAX1

3

5

W

4

W

3

W

2

W

© 2011 Nature America, Inc. All rights reserved.

WNT3a (100 ng/ml), FGF10 (10 ng/ml), KGF (10 ng/ml) BMP4 (10 ng/ml), EGF (10 ng/ml) (WKFBE)

NOGGIN/ SB-431542

WKFBE + ATRA (0.5 µM) 1

KF BE

a

mRNA (ratio to β-ACTIN)

Figure 4 Induction of lung and pharyngeal pouch markers from ventral AFE generated in vitro. (a) Expression of PAX1, NKX2.1, FOXP2, GATA6 and FOXJ1 in HES2-derived cells generated in the two conditions schematically represented on top of the panel (n = 4 to six culture wells from three independent experiments, *, significantly different from WKFBE conditions; P < 0.05). WKFBE: WNT3a, KGF, FGF10, BMP4 and EGF. (b) Induction of SFTPC and GCM2 mRNA in ventralized AFE in the presence of factors indicated in the figure.

GCM2

fate26 (Fig. 4a). To enhance SFTPC induction, we added combinations of signaling agonists and antagonists at day 11 to AFE ventralized in the presence of retinoic acid. Among the >400 combinations exam ined, WNT3a + FGF10 + FGF7 induced high levels of SFTPC mRNA (Fig. 4b) at day 19, consistent with the developmental observations that FGF10 and Wnt signaling are critical for distal lung development26. To assess whether NOGGIN/SB-431542-induced AFE could generate pharyngeal pouch derivatives, we sought to pattern the cultures ven tralized with WKFBE in the absence of retinoic acid. Consistent with the requirement of sonic hedgehog (SHH) and FGF8 for parathyroid development30, addition of FGF8 or SHH to AFE cultures induced the parathyroid-specific marker GCM2 (Fig. 4b). The effects of SHH and FGF8 were not additive, mirroring in vivo epistasis studies showing that Shh is upstream of Fgf8 in mouse pharyngeal pouch development30. Although we did not test all temporal and signaling permutations exhaustively, these data suggest that NOGGIN/SB-431542-induced cells are capable of differentiating into downstream lineages, includ ing the lung field and pharyngeal pouches. Collectively, our data show that dual inhibition of BMP and TGF-β signaling in hES/hiPS cell-derived definitive endoderm specifies a highly enriched AFE population, providing an in vitro approach for the directed differentiation of human pluripotent cells into cell types and tissues derived from the AFE in vivo. 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 This work was supported by NYSTEM grant NO8G-422 to H.-W.S. AUTHOR CONTRIBUTIONS M.D.G. performed all experiments with assistance of A.C. M.-C.N., V.G.-E., S.L.S. and G.K. advised and assisted with induction of definitive endoderm. S.L.S., I.R.L. and C.S. generated and characterized the hiPS lines, respectively. M.D.G. and H.-W.S. designed the experiments and wrote the manuscript.

271

letters 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/.

© 2011 Nature America, Inc. All rights reserved.

Published online at http://www.nature.com/naturebiotechnology/. Reprints and permissions information is available online at http://npg.nature.com/ reprintsandpermissions/.

1. D’Amour, K.A. et al. Production of pancreatic hormone-expressing endocrine cells from human embryonic stem cells. Nat. Biotechnol. 24, 1392–1401 (2006). 2. Gouon-Evans, V. et al. BMP-4 is required for hepatic specification of mouse embryonic stem cell-derived definitive endoderm. Nat. Biotechnol. 24, 1402–1411 (2006). 3. Cai, J. et al. Directed differentiation of human embryonic stem cells into functional hepatic cells. Hepatology 45, 1229–1239 (2007). 4. Murry, C.E. & Keller, G. Differentiation of embryonic stem cells to clinically relevant populations: lessons from embryonic development. Cell 132, 661–680 (2008). 5. Spence, J.R. et al. Directed differentiation of human pluripotent stem cells into intestinal tissue in vitro. Nature 470, 105–109 (2010). 6. Yamanaka, S. A fresh look at iPS cells. Cell 137, 13–17 (2009). 7. Gadue, P., Huber, T.L., Paddison, P.J. & Keller, G.M. Wnt and TGF-beta signaling are required for the induction of an in vitro model of primitive streak formation using embryonic stem cells. Proc. Natl. Acad. Sci. USA 103, 16806–16811 (2006). 8. Zorn, A.M. & Wells, J.M. Vertebrate endoderm development and organ formation. Annu. Rev. Cell Dev. Biol. 25, 221–251 (2009). 9. Graham, A. Deconstructing the pharyngeal metamere. J. Exp. Zoolog. B Mol. Dev. Evol. 310, 336–344 (2008). 10. Rodewald, H.R. Thymus organogenesis. Annu. Rev. Immunol. 26, 355–388 (2008). 11. Jenq, R.R. & van den Brink, M.R. Allogeneic haematopoietic stem cell transplantation: individualized stem cell and immune therapy of cancer. Nat. Rev. Cancer 10, 213–221 (2010). 12. Lai, L. & Jin, J. Generation of thymic epithelial cell progenitors by mouse embryonic stem cells. Stem Cells 27, 3012–3020 (2009). 13. Hidaka, K. et al. Differentiation of pharyngeal endoderm from mouse embryonic stem cell. Stem Cells Dev. 19, 1735–1743 (2010). 14. Lin, R.Y., Kubo, A., Keller, G.M. & Davies, T.F. Committing embryonic stem cells to differentiate into thyrocyte-like cells in vitro. Endocrinology 144, 2644–2649 (2003).

272

15. Bingham, E.L., Cheng, S.P., Woods Ignatoski, K.M. & Doherty, G.M. Differentiation of human embryonic stem cells to a parathyroid-like phenotype. Stem Cells Dev. 18, 1071–1080 (2009). 16. Wang, D., Haviland, D.L., Burns, A.R., Zsigmond, E. & Wetsel, R.A. A pure population of lung alveolar epithelial type II cells derived from human embryonic stem cells. Proc. Natl. Acad. Sci. USA 104, 4449–4454 (2007). 17. Yasunaga, M. et al. Induction and monitoring of definitive and visceral endoderm differentiation of mouse ES cells. Nat. Biotechnol. 23, 1542–1550 (2005). 18. Sherwood, R.I., Chen, T.Y. & Melton, D.A. Transcriptional dynamics of endodermal organ formation. Dev. Dyn. 238, 29–42 (2009). 19. Peters, H., Neubüser, A., Kratochwil, K. & Balling, R. Pax9-deficient mice lack pharyngeal pouch derivatives and teeth and exhibit craniofacial and limb abnormalities. Genes Dev. 12, 2735–2747 (1998). 20. Li, Y. et al. Sfrp5 coordinates foregut specification and morphogenesis by antagonizing both canonical and noncanonical Wnt11 signaling. Genes Dev. 22, 3050–3063 (2008). 21. Wood, H.B. & Episkopou, V. Comparative expression of the mouse Sox1, Sox2 and Sox3 genes from pre-gastrulation to early somite stages. Mech. Dev. 86, 197–201 (1999). 22. Weinstein, D.C. et al. The winged-helix transcription factor HNF-3 beta is required for notochord development in the mouse embryo. Cell 78, 575–588 (1994). 23. Chambers, S.M. et al. Highly efficient neural conversion of human ES and iPS cells by dual inhibition of SMAD signaling. Nat. Biotechnol. 27, 275–280 (2009). 24. Vitelli, F. et al. A genetic link between Tbx1 and fibroblast growth factor signaling. Development 129, 4605–4611 (2002). 25. Bachiller, D. et al. The role of chordin/Bmp signals in mammalian pharyngeal development and DiGeorge syndrome. Development 130, 3567–3578 (2003). 26. Morrisey, E.E. & Hogan, B.L. Preparing for the first breath: genetic and cellular mechanisms in lung development. Dev. Cell 18, 8–23 (2010). 27. Tanaka, M., Schinke, M., Liao, H.S., Yamasaki, N. & Izumo, S. Nkx2.5 and Nkx2.6, homologs of Drosophila tinman, are required for development of the pharynx. Mol. Cell. Biol. 21, 4391–4398 (2001). 28. Wallin, J. et al. Pax1 is expressed during development of the thymus epithelium and is required for normal T-cell maturation. Development 122, 23–30 (1996). 29. Chen, F. et al. A retinoic acid-dependent network in the foregut controls formation of the mouse lung primordium. J. Clin. Invest. 120, 2040–2048 (2010). 30. Moore-Scott, B.A. & Manley, N. Differential expression fo sonic hedgehog along the anterior-posterior axis regulates patterning of pharyngeal pouch endoderm and pharyngeal endoderm-derived organs. Dev. Biol. 278, 323–335 (2005).

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

ONLINE METHODS

© 2011 Nature America, Inc. All rights reserved.

Cells and culture conditions. hESC (HES2, National Institutes of Health code ES02 from ES Cell International; passages 25–33) were cultured on mouse embryonic fibroblasts plated at 8,000–12,000 cells/cm 2. A medium of DMEM/F12, 20% knockout serum replacement (Gibco), 0.1 mM βmercaptoethanol (Sigma-Aldrich), and 20 ng/ml FGF-2 (R&D) was changed daily. Cells were passaged with trypsin, washed and replated at a dilution of 1:5 to 1:10. HDF2 and HDF9, hiPS cell lines, were cultured as hES2. hES and hiPS cultures were maintained in a 5% CO2/air environment, and hES differentia tions were maintained in a 5% CO2/5% O2/90% N2 environment. Endoderm induction. Mouse embryonic fibroblasts were depleted by a 24 h passage on Matrigel (Gibco) with Y-27632 (10 µM), and embryoid bodies were formed on low-adherence dishes (Costar). During embryoid body for mation and differentiation, HES2 cells were reseeded at the same concen tration (1:1 dilution), whereas HDF2 and HDF9 required a higher seeding (2:1–4:1 concentration) for efficient endoderm generation. Differentiations were performed in a medium of DMEM/F12 (Invitrogen) supplemented with N2 (Gibco), B27 (Gibco), ascorbic acid (50 µg/ml, Sigma), Glutamax (2 mM, Invitrogen), monothioglycerol (0.4 µM, Sigma). The following concentrations of factors were used for primitive streak formation, endoderm induction, anterior/pharyngeal endoderm induction, and subsequent anterior posterior and dorsoventral patterning: human BMP-4, 1 ng/ml and 10 ng/ml; human bFGF, 2.5 ng/ml; human activin A, 100 ng/ml; human Noggin, 200 ng/ml; Y-27632, 1 µM; SB-431542, 10 µM; human FGF10, 10 ng/ml; human FGF7, 10 ng/ml; murine EGF 20 ng/ml; and human WNT3a, 100 ng/ml. Hepatic conditions were as previously described2 and contain BMP-4, 50 ng/ml; bFGF, 10 ng/ml; VEGF, 10 ng/ml; HGF, 10 ng/ml; TGFα, 20 ng/ml; dexamethasone, 40 ng/ml. All factors were purchased from R&D systems, except SB-431542 and Y-27632, which are from Tocris, and dexamethasone (Sigma). The factors were added in the following sequence: day 1, Y-27632; days 1–5, BMP4, bFGF and activin A; days 5–9, Noggin and SB-431542; days 7–19 WNT3a, FGF10, BMP4, FGF7 and EGF. For hepatic differentiation, factors were added in the following sequence: day 1, Y-27632; days 1–5, BMP4, bFGF and activin A; days 5–9, dexamethasone, bFGF, HGF, VEGF, EGF, TGFa and BMP4. For some experiments, 500 µM all-trans retinoic acid (Sigma) was added to the cultures. At day 5, embryoid bodies were trypsinized (Gibco) and plated at 10,000– 50,000 cells/cm2 in gelatin-coated, tissue culture–treated dishes (Fisher). The day 11 screen of morphogens (Fig. 3h) included pairwise and some higher order of magnitude additions of FGFS (FGF10 + FGF7) (R&D Systems), SU-5402 (EMD Chemicals), Wnt5a (R&D Systems), WNT3a (R&D Systems), sFRP1 (R&D Systems), BMP4, Noggin, SHH (R&D Systems), cyclopamine (EMD Chemicals), SB-431542, TGF-β1 (R&D Systems), DAPT (EMD Chemicals), retinoic acid (Sigma), DEAB (Sigma), EGF (R&D Systems), tyrophastin AG-1478 (EMD Chemicals) and WP1066 (EMD Chemicals). Quantitative PCR. Total RNA was extracted using Trizol (Invitrogen), phase lock tubes (5′ Prime) and the RNeasy kit (Qiagen). We treated 1–2 µg of total RNA with DNase I (Qiagen) and reverse transcribed it using random hexam ers and Superscript III (Invitrogen). Amplified material was detected using SybrGreen (Invitrogen). Real-time quantitative PCR was performed on ABI 7900HT thermocycler (Applied Biosystems), with a 10-min step at 95 °C followed by 40 cycles of 95 °C for 15 s and 60 °C for 1 min. All experiments were done in triplicate with SYBR GreenER quantitative PCR SuperMix (Invitrogen).

doi:10.1038/nbt.1788

Denaturing curves for each gene were used to confirm homogeneity of the DNA product. Absolute quantification was obtained in relation to a standard curve of genomic DNA dilution series. Quantified values for the gene of interest were normalized against input determined by the housekeeping genes GAPDH and β-ACTIN. qPCR for each culture well was performed in triplicate. Primer sequences are listed in Supplementary Table 1. Flow cytometry. Day 5 embryoid bodies or day 13 monolayer cultures were dissociated with trypsin/EDTA (2 min). The cells were stained with directly conjugated CXCR4 (Invitrogen), c-KIT (BD Biosciences) and EPCAM (BD Biosciences) in IMDM complemented with 10% serum. Cells were analyzed on a LSRII. Flowjo software (Tree Star) was used for all analyses. Immunofluorescence. Day-7, -9 or -13 cultures were fixed with fresh parafor maldehyde (4%) for 30 min at 25 °C and then washed in PBS. The cells were permeabilized and blocked in a solution with 0.1% saponin, 0.1% BSA, 10% FCS (Gemstar) and 10% fetal donkey serum (Jackson Immunofluorescence). For three-dimensional cultures, cells were embedded at day 5 in Matrigel using the “embedded” and “on-top” assays, as previously described31. Cultures were embedded in Optimal Cutting Temperature (OCT, Tissue Tek) at day 9, extracted, sectioned and processed as above. Primary antibodies were added overnight, and include PAX-9 (Abcam), TBX1 (Abcam), SOX2 (Stemgent, Santa Cruz), CDX2 (Abcam), NKX2-1 (Invitrogen) and AIRE (Santa Cruz). Cells were washed and incubated with donkey anti-mouse whole IgG-Dylight488, donkey anti-goat whole IgG-Cy3, and donkey anti-rabbit whole IgG-Cy5 for 1 h. The cells were washed, and nuclei were stained with DAPI (Invitrogen). Stains were visualized using a fluorescence microscope (Leica DMI 4000B, Wetzlar, Germany). This instrument was fitted with a DFC340 Fx Monochrome Cooled Digital Camera for fluorescent acquisition (Leica) and variable objective lenses (5×–40×) were used. Filter models: A4 UV, 11504135; GFP, 11532366; YFP 1153267; RFP, 11513894; CY3, 11600231. Exposure settings varied, but were set based on hepatic-specified cultures differentiated and stained in parallel. Images were acquired using Leica Application Suite Advanced Fluorescence Software Package AF6000 (Leica) in PBS at 25 °C. Images are shown without render ing or deconvolution. Images were digitally processed using Adobe Photoshop CS4 (Adobe) in accordance with Nature Publishing Guidelines by altering only contrast and brightness, and these manipulations were performed on hepatic specified and experimental conditions simultaneously. Quantification was per formed by counting a minimum of ten random fields at 20× magnification. On most panels, the magnification by the lens objective is listed. Mice. NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ (NOD/SCIDIl2rg−/−) mice were pur chased from Jackson Laboratory. Animals were kept in a specific pathogen-free facility. Experiments and animal care were performed in accordance with the Mount Sinai Institutional Animal Care and Use Committee. One million cells were injected under the kidney capsule. Outgrowths were excised, embedded in OCT and analyzed using hematoxylin and eosin stains for morphology or immunofluorescence for specific antigens as above. Statistical analysis. For statistical analysis, unpaired t-test and when more than two groups were compared, one-way ANOVA was used. Results are expressed as mean ± s.e.m. 31. Lee, G.Y., Kenny, P.A., Lee, E.H. & Bissell, M.J. Three-dimensional culture models of normal and malignant breast epithelial cells. Nat. Methods 4, 359–365 (2007).

nature biotechnology

letters

Implantable magnetic relaxation sensors measure cumulative exposure to cardiac biomarkers

© 2011 Nature America, Inc. All rights reserved.

Yibo Ling1–3,8, Terrence Pong1,4,5,8, Christophoros C Vassiliou2,3,8, Paul L Huang4,6 & Michael J Cima3,7 Molecular biomarkers can be used as objective indicators of pathologic processes. Although their levels often change over time, their measurement is often constrained to a single time point. Cumulative biomarker exposure would provide a fundamentally different kind of measurement to what is available in the clinic. Magnetic resonance relaxometry can be used to noninvasively monitor changes in the relaxation properties of antibody-coated magnetic particles when they aggregate upon exposure to a biomarker of interest. We used implantable devices containing such sensors to continuously profile changes in three clinically relevant cardiac biomarkers at physiological levels for up to 72 h. Sensor response differed between experimental and control groups in a mouse model of myocardial infarction and correlated with infarct size. Our prototype for a biomarker monitoring device also detected doxorubicin-induced cardiotoxicity and can be adapted to detect other molecular biomarkers with a sensitivity as low as the pg/ml range. The physiological levels of molecular biomarkers can be regarded as time-varying continuous signals. However, clinicians seldom take advantage of this temporal information in making diagnostic and prognostic decisions. Biomarker measurements are often made at single time points, which do not adequately capture the dynamics of the underlying signal if they miss transient changes occurring between measurements. For instance, levels of serum cardiac troponin I (cTnI), creatinine kinase (the CK-MB isoform) and myoglobin elevate and return to baseline in a stereotyped manner after acute myocardial infarction (MI). A given measured value could corres pond to either the early or late phase of biomarker release. Here we describe implantable magnetic relaxation sensors that are capable of integrating biomarker levels over time. The signal from such a device corresponds to the entirety of biomarker release long after a pathologic event has occurred, and even after the concentrations have returned to baseline. Most MIs are characterized by symptoms of severe discomfort. However, a significant minority, defined as unrecognized MIs, are accompanied by minimal or no symptoms. The 30-year follow-up of the Framingham Heart Study reported that 28% and 35% of MIs are unrecognized in men and women, respectively1. However, current

standards for detecting unrecognized MIs rely primarily on electrocardiographic surveillance. Results vary markedly between such studies because of differing electrocardiographic criteria2,3. Patients at high risk for unrecognized MIs are followed periodically by their cardiologists but MIs timed between these visits can go unnoticed. A sensor that reports on integrated MI biomarker levels throughout these intervals could therefore be used to identify these previously undetectable infarcts. A new class of nanoparticle-based magnetic resonance contrast agents makes this type of application feasible. Superparamagnetic iron oxide nanoparticles aggregate about analyte molecules and alter the transverse relaxivity (T2) of surrounding water protons4–8. Dubbed magnetic relaxation switches (MRSw), the particles can be functionalized to detect a variety of small molecules including proteins, nucleic acids, oligonucleotides, peptides, receptors, ligands and antibodies8–14. Incorporation of these agents into implantable sensors permits noninvasive biomarker measurements using magnetic resonance relaxometry. We constructed small discrete sensors incorporating MRSw technology to measure cardiac biomarkers and characterized their performance in vivo in a murine model of MI characterized by the release of three clinically validated biomarkers at physiological concentrations. MI detection is particularly well matched for MRSw technology as MI size, a clinically relevant feature, is expected to correspond to cumulative biomarker release. A previous effort to explore the in vivo efficacy of similar sensors was fundamentally limited by the lack of correspondence between naturally occurring cancer and the xenograft tumor model used; the xenograft tumors released biomarkers at concentrations significantly above, and in a temporal fashion that did not resemble, that of physiological tumors15. Myoglobin, cTnI and CK-MB are functional proteins released by the ischemic myocardium after an acute infarct. Their serum dynamics are well known in humans: whereas levels of myoglobin rise and fall rapidly over 24 h after MI, levels of cTnI and CK-MB can remain above baseline for up to a week16. We adapted a left anterior descending (LAD) artery ligation procedure, as described17, to experimentally induce acute myocardial infarction in C57BL6 mice. The sensors were designed to sample biomarkers in a subcutaneous space created within the animal’s flank. If in situ sensing is to be achieved, the intended cardiac targets traditionally measured in

1Harvard-MIT Division of Health Sciences & Technology, Cambridge, Massachusetts, USA. 2Department of Electrical Engineering & Computer Science, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 3The David H. Koch Institute for Integrative Cancer Research, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 4Cardiovascular Research Center, Massachusetts General Hospital, Boston, Massachusetts, USA. 5School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts, USA. 6Cardiology Division, Harvard Medical School, Boston, Massachusetts, USA. 7Department of Materials Science & Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA. 8These authors contributed equally to this work. Correspondence should be addressed to M.J.C. ([email protected]).

Received 23 November 2010; accepted 18 January 2011; published online 13 February 2011; doi:10.1038/nbt.1780

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

273

letters

10,000 cTnl CK-MB 8,000 Myo 6,000

200 150 100

4,000

50

2,000

0

0 0

4

8

12

16

20

Time after LAD ligation (h)

24

Normalized conc.

b

Serum dynamics 250

Myo. conc. (ng/ml)

cTnl and CK-MB conc. (ng/ml)

a

c

cTnl extravasation

** * **

1.00 0.75

Myoglobin extravasation

CK-MB extravasation

**

**

** ** ** ** **

**

0.50

* **

0.25 0

d

MI Sham Control

** **

**

* 0

20

40

60

80

Time after implantation (h)

100 0

20

40

60

80

Time after implantation (h)

100 0

**

20

40

60

80

100

Time after implantation (h)

© 2011 Nature America, Inc. All rights reserved.

Figure 1 Evidence of cardiac biomarker extravasation from serum to the subcutaneous space. (a) cTnI, myoglobin and CK-MB serum profiles after LAD ligation are within the range of literature results and confirm the validity of the MI model used. Approximately 0.5 ml of blood was drawn from each subject, the serum extracted by centrifugation and biomarker levels measured by ELISA at the indicated times after LAD ligation. Results are averages (n = 4), with error bars omitted for uncluttered visualization. (b–d) Three experimental conditions were imposed: control (sensor implantation only), sham (sensor implantation and thoracotomy only) and MI (sensor implantation, thoracotomy and LAD ligation). Each biomarker extravasates, with MI groups exhibiting significantly elevated concentrations, as compared to the corresponding control and sham groups. Presumably, implantation-induced injury caused the low initial cTnI in the setting of high initial myoglobin and CK-MB. The similarity between sham and control groups indicates that any thoracotomy-induced biomarker release is not significantly ‘visible’ in the sensor implant site. Extravasate samples were obtained by flushing the flank with 1 ml PBS at the indicated times after LAD ligation and measured using ELISA. Values are normalized within each panel to the maximum measured concentration. Results are averages ± s.e.m. (n = 4, normalized); P < 0.05 is indicated by black asterisks between MI/sham, red asterisks between MI/ control and green asterisks between sham/control.

serum (Fig. 1a) must be detectable in the subcutaneous space. Serum levels of cardiac biomarkers after acute MI are well characterized in the existing literature18, but their extravasation to the subcutaneous flank had not previously warranted study. We experimentally determined the extravasation dynamics (Fig. 1b–d) under three conditions: MI, sham and control. Whereas MI groups received sensor implantation, thoracotomy and LAD ligation, sham groups received sensor implantation and thoracotomy but no LAD ligation. Control groups received sensor implantation only. These results confirmed that the subcutaneous space is a viable site for cardiac biomarker detection, as biomarker elevations in the MI group differed significantly (P < 0.05) from the sham and control groups. It should therefore be possible to distinguish between the experimental conditions based on measurements acquired from implanted sensors. There are, however, some initial elevations in myoglobin and CK-MB for the control and sham groups that we would expect the sensors to detect. Open chest surgery and subcutaneous device implantation cause substantial noncardiac injury. The early behaviors of these biomarkers are consistent with their differing specificities for cardiac injury; cTnI is highly specific to cardiac damage, but CK-MB is less specific and myoglobin is a marker of general muscle damage. That myoglobin and CK-MB are elevated in the control groups (Fig. 1c,d, green) suggests local implantationinduced trauma will be visible to the implanted sensors. No significant difference can be found between the sham and control groups for any biomarker, suggesting that thoracotomy-induced trauma should not affect the in situ sensor response. The movement of protein biomarkers from the circulation to the subcutaneous space should depend on the chemical properties of the specific biomarker as well as the vascularization and morphology of the implant site. The reliability of subcutaneously implanted sensors may be limited by these factors. Nonetheless, there is little delay seen here between the peaks in the accumulation of serum (Fig. 1a) and extravasate (Figs. 1b–d) biomarkers. Rapid subcutaneous availability of these biomarkers, coupled with a sufficiently rapid sensor response, should enable potential applications of MI detection using this technology. The samples were obtained by flushing the subcutaneous space with 1 ml PBS and were assayed by enzyme-linked immunosorbent assay (ELISA). We considered these measurements a firstorder approximation of biomarker extravasation and used them as a checkpoint for proceeding with sensor implantation. 274

To engineer discrete sensors containing appropriately functionalized MRSw nanoparticles, we enclosed particles in the reservoirs by size-exclusion membranes (Fig. 2a). The dynamic range of each sensor was tuned by adjusting the device transport characteristics and the nanoparticle chemistry. Careful selection of the membrane composition, pore size and surface area permitted control of analyte transport into the device. cTnI sensors were tuned for sensitivity in the 10–100 ng/ml range, whereas sensors to detect myoglobin and CK-MB were tuned to detect in the 100 ng/ml to 1 µg/ml range to reflect the expected subcutaneous levels (Fig. 2b–d). Antibody-antigen binding can provide an extremely strong noncovalent interaction. We tested the proposition that antibody-based MRSw sensors would be irreversible and that the degree of T2 change would correspond to the cumulative analyte exposure. The sensors were exposed to four different constant myoglobin concentrations (Fig. 2e) to investigate the integrative capacity of antibody-based MRSw. The overall response rate is proportional to the concentration outside the device, demonstrating that the sensor behavior is dominated by diffusion transport into the device reservoir. We define exposure as E(tm ) =

tm

∫

C(t )dt , where C(t) is the analyte concentration

t =0

as a function of time and tm is the time of measurement. Plotting the measured T2 against exposure yields a single curve that is independent of the analyte concentration and verifies that the sensors operate as dosimeters in a diffusion-limited regime (Fig. 2f, solid and dotted lines). The diffusion occurs across the full area of the device reservoir and the devices saturate at ~1.5 µg·h/ml. The saturating exposure level can be tuned to the anticipated in vivo biomarker levels by adjusting the area available for diffusion. Reducing the diffusion area by half would, therefore, increase the saturating exposure by a factor of two. We next studied sensor response to three time varying concentration profiles as shown in Figure 2f (inset) (n = 2) because physiologic biomarker levels often follow nonconstant dynamic patterns. These profiles simulate transient biomarker release when the instantaneous biomarker concentration at the time of measurement is zero. The sensor signal persists after the analyte is no longer present, and the measurements (Fig. 2f, red/blue/green data points) coincide with the measured T2 versus exposure profile (Fig. 2f, solid and dotted lines). The results from prolonged exposure to constant and transient VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

letters a

b

c

d

∆T2 (ms)

Figure 2 Use of functionalized MRSw particles, encapsulated within discrete cTnl sensor response Myo sensor response CK-MB sensor response Size-exclusion MRSw in in vitro in vitro in vitro membrane sensors and calibrated in vitro, to measure solution 16 16 16 cumulative exposure to analyte in vitro. 12 12 12 (a) The sensor consists of a reservoir containing 1.6 mm 8 8 8 MRSw particles enclosed by a size-exclusion 4 4 4 membrane. T2 changes are produced when 0 0 0 7.9 mm analytes diffuse across the membrane and 10 100 1,000 10 100 1,000 10,000 1 10 100 1,000 4.1 mm initiate particle aggregation. (b–d) Sensor cTnl conc. (ng/ml) Myoglobin conc. (ng/ml) CK-MB conc. (ng/ml) response as a function of analyte concentration was calibrated to match expected in vivo 100 100 concentrations. The measurements were 75 acquired after an incubation time of 72 h 75 100 for cTnI and 24 h for myoglobin and CK-MB 50 Constant conc. 0 50 50 100 sensors to match the expected duration of 50 ng/ml 50 0 30 elevation for the respective biomarkers. Results 100 25 50 25 25 0 0 are averages ± 95% confidence intervals (n = 0 24 48 72 4). (e) Characterization of sensor saturation Time (h) 0 0 with exposure to constant concentration 0 24 48 0 0.5 1.0 1.5 2.0 Exposure (µg•h/ml) Time (h) profiles. Devices were maintained in a constant concentration of myoglobin solution and the relaxation time T2 was measured every 20 min. The time to saturation depends on the concentration. All of the devices, with the exception of the control, show a similar response and saturate at the same T2 value. The total exposure of the device to myoglobin (measured in units of [µg·h/ml]) is defined as the area under the concentration versus time curve up until the measurement time. (f) Exposure to constant and temporal concentration profiles demonstrate that MRSw sensors function as dosimeters. Plot of T2 versus the constant exposure profiles obtained from e are shown as curves. The symbols represent the average T2 of devices exposed to various temporal concentration profiles shown inset. The curves and symbols lie on top of each other, demonstrating that the response depends exclusively on exposure. Most points have deviations smaller than the symbol size. Measurements were made after several hours of incubation at zero concentration.

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

MI

Conc. (ng/ml)

(Axial view) T2 (ms)

Control

b 14 12 10 8 6 4 2 0

35 34 33 32 31 30 29 28 27 26 25

(Transverse view)

MI Sham Ctrl

*

Explanted sensors

*

* * *

*

(cTnl) ∆T2 (ms)

c

Myoglobin Biomarker

CKMB

R2 = 0.96574

8 6 4 25

30

7

35

40

45

2

R = 0.81258

6 5 4 3

e 16

25

30

35

40

45

2 R = 0.73971

14 12 10 8 6

cTnl

∆T2 vs. infarct size

10

2

d

(Myoglobin) ∆T2 (ms)

a

(CK-MB) ∆T2 (ms)

Figure 3 Sensor response differs markedly between MI and sham/control groups, and its magnitude correlates with the extent of infarction. (a) A T2 map (color bar on the right) superimposed on T2 -weighted images of myoglobin sensors demonstrates the feasibility of MRI-based in situ measurements after 24 h implantation with (MI) or without (control) concomitant LAD ligation at the time of sensor implantation. The images show that the sensors can be measured in either the axial or the transverse plane. (b) The T2 of explanted sensors, as measured by single-sided proton relaxometry (after 72 h implantation for cTnI and 24 h for myoglobin and CK-MB), from the MI group are significantly higher (* indicates P < 0.05 using the Wilcoxon rank-sum two-sided test) than from the control/sham groups for all three biomarkers. Increases in the T2 of myoglobin and CK-MB sensors for both control and sham groups were expected and consistent with the extravasation results. Results are averages ± 95% confidence intervals (n = 36; 6 sensors/subject and 6 subjects/group). (c–e) T2 change of the same explanted sensors from b, if disaggregated into the individual subjects and replotted as a function of infarct size, correlates positively with infarct size for all three cardiac biomarkers. The cumulative release of biomarkers from the infarcted myocardium generates the final T2 sensor value. This characteristic may contribute to the correspondence between sensor reading and infarct size. Results are averages ± 95% confidence intervals (n = 6 sensors/subject).

implanted before any further surgery so that their acute post-implant exposures did not vary between groups. Any differences between the three groups can therefore be attributed to true biomarker extravasation. In situ MRI measurements demonstrated that the sensors can be interrogated at the implant site (Fig. 3a). Most measurements were obtained from explanted sensors using a single-sided relaxometer. Despite the limited sensitivity of the single-sided system (Online Methods), T2 increases of the MI over the sham and control groups for all three biomarkers are evident (Fig. 3b). Wilcoxon rank-sum twosided tests give significant P values (P < 0.05) for cross comparisons between the MI and control groups, and between the MI and sham

∆T2 (ms)

© 2011 Nature America, Inc. All rights reserved.

concentration profiles indicate that the biosensors ultimately function in a manner analogous to radiation dosimeters; they report consistent T2 changes independent of the biomarker exposure profile. The measurement of cumulative analyte release could be useful in detecting biomarker concentrations that are below the level of detection of blood-draw assays. This property also limits antibodyfunctionalized MRSw to fixed lifetimes reached when the binding sites are saturated. Cumulative sensors can thus only be used for finite durations, necessitating careful engineering of the particles and devices to match the chosen application’s required sensitivity and lifetime. We proceeded to implant subcutaneous sensors specific for each biomarker in the flanks of animals from the MI, sham and control groups. It is possible that the sensors might be acutely exposed to intravascular fluid as a result of surgery-induced capillary disruption. Sensors in all three (MI, sham, control) groups were therefore

T2 (ms)

f

T2 (ms)

e

15

20 25 30 35 40 45 INF/LV (percent of myocardium infarcted)

275

letters

350

*

cTnl CK-MB Myoglobin

300 250 200 150 100 50

Doxorubicin

Control

0

b (Myoglobin) ∆T2 (ms)

cTnl conc. (ng/ml)

10 9 8 7 6 5 4 3 2 1 0

Myo. & CK-MB conc. (ng/ml)

a

30 25 20

*

15 10 5 0

Doxorubicin

Control

© 2011 Nature America, Inc. All rights reserved.

Figure 4 An implanted MRSw sensor device can detect the cardiotoxic effect of the chemotherapeutic drug doxorubicin in vivo. (a) Doxorubicin induces a significant (* indicates P < 0.05) increase in serum myoglobin (ELISA measurements). Results are averages ± 95% confidence intervals (n = 4). Wilcoxon rank-sum two-sided tests were used to test for statistical significance. (b) Myoglobin sensors explanted (after 72 h) from a doxorubicin-administered group show significant increases in T2 over those from a control group. Results are averages ± 95% confidence intervals (n = 16; 4 sensors/subject and 4 subjects/group).

groups but not between sham and control groups. The elevations in levels of myoglobin and CK-MB above baseline in the sham and control groups reflect noncardiac injury caused by the initial surgical procedure. Implanted MRSw sensors functionalized against IgG alone showed negligible increases in T2 (data not shown). The infarct zone can be visualized through 2,3,5-triphenyltetrazolium chloride staining and objectively quantified as the fraction of left ventricular volume19. The cumulative release of cardiac biomarkers should be directly proportional to the magnitude of infarction, as the biomarkers are functional proteins directly released from the ischemic tissue. A comparison of sensor readings with infarct size (Fig. 3c–e) shows a consistent trend for all three biomarkers, despite the relatively large errors. The capability to quantitatively measure infarct size has important implications for risk stratification of MI patients20,21. This factor, usually measured indirectly in the clinic through functional tests or imaging, can be directly quantified by the implantable sensors described here. We further studied application of the sensors to discern drug cardiotoxicity. Doxorubicin is a potent anthracycline antibiotic that has found wide clinical use as a cancer chemotherapeutic22,23. Its cardiotoxic effects are well known; after administration, patients exhibit dose-dependent loss of cardiac myocytes accompanied by serum cardiac biomarker elevation24. We confirmed that serum myoglobin increases after doxorubicin administration in a murine model (Fig. 4a), although at an order of magnitude lower concentration than after acute MI (102 ng/ml versus 103 ng/ml). Myoglobin sensors were thus left in vivo for 72 h after implantation (as opposed to 24 h for acute MI). The results show a clear distinction in sensor T2 between the experimental and control groups, thus validating sensor efficacy in assaying drug cardiotoxicity (Fig. 4b). These sensors may potentially be used in the future to establish the cardiac side effects of novel pharmaceuticals. New high-sensitivity troponin assays will lead to an increasing population of patients identified with elevated troponins of nonischemic etiology. Additional assays will need to be developed to help physicians distinguish between acute coronary syndromes and other causes. In vivo diagnostic MRSw measurements may prove to be especially useful in discriminating between troponin elevations caused by transient tachyarrhythmias and sustained ischemic episodes. These integrative sensors are engineered to have lower sensitivity so that they do not saturate upon total exposure to analyte released during and after an MI. The troponin sensor discussed here, for example, 276

has sensitivity in the ng/ml range but could be engineered for even higher sensitivity to the pg/ml range of the newest troponin assays. Future longer-term biomarker accumulation studies should be feasible as long as device saturation characteristics and antibody stability are taken into consideration. Thus, integrative sensors may provide prognostic value as sentinels in high-risk patients and for the detection of unrecognized MIs. The ability to track cumulative biomarkers in vivo can be advantageous for its ability to capture transient events, which are frequently missed with serial testing. Furthermore, this type of sensor can be useful in research as a tool for monitoring biomarkers in small animals in which serial blood draws may not be feasible. MRSw can be functionalized to detect a variety of small molecules and further study will establish the general efficacy of these sensors. The sensors require single time point measurements that fit comfortably into the traditional clinical paradigm because they are practical to obtain, yet provide information on biomarker levels from the time of sensor implantation to when accumulation is measured. We have shown the utility of cumulative biomarker measurements for applications relevant to heart disease. Such data will in all likelihood prove to possess clinical value for other applications as well. The MRSw molecular sensors described here represent a possible path in that direction. We showed that they perform adequately in the detection of pathologically elevated levels of three soluble disease biomarkers that extravasate from the circulation to the implant site. They are stable over several months and can be applied as disease sentinels in MI patients at high risk for short-term recurrence. We are presently establishing their long-term durability and will investigate alternative particle functionalization chemistries or binding moieties as necessary. Methods Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturebiotechnology/. Acknowledgments This work was supported by National Cancer Institute Centers of Cancer Nanotechnology Excellence no. 5 U54 CA119349-12 and CA151844 grants and National Science Foundation Division of Materials Research Award no. 0746264. Y.L. was supported by a National Defense Science and Engineering Graduate fellowship. T.P. was supported by an American Heart Association fellowship. AUTHOR CONTRIBUTIONS Y.L. initiated the project, designed and performed experiments, analyzed data and wrote the manuscript. T.P. conceived experiments, designed and performed animal experiments, analyzed data and wrote the manuscript. C.C.V. contributed ideas, performed experiments, analyzed data and wrote the manuscript. P.L.H. contributed to the design experiments related to clinical relevance, doxorubicin toxicity and myocardial infarction model. M.J.C. was the principal investigator; he initiated the project, conceived experiments and obtained funding. 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. Gutterman, D.D. Silent myocardial ischemia. Circ. J. 73, 785–797 (2009). 2. Sheifer, S.E., Manolio, T.A. & Gersh, B.J. Unrecognized myocardial infarction. Ann. Intern. Med. 135, 801–811 (2001). 3. Ammar, K.A., Kors, J.A., Yawn, B.P. & Rodeheffer, R.J. Defining unrecognized myocardial infarction: a call for standardized electrocardiographic diagnostic criteria. Am. Heart J. 148, 277–284 (2004). 4. Shapiro, M.G., Atanasijevic, T., Faas, H., Westmeyer, G.G. & Jasanoff, A. Dynamic imaging with MRI contrast agents: quantitative considerations. Magn. Reson. Imaging 24, 449–462 (2006). 5. Daniel, K.D. et al. Multi-reservoir device for detecting a soluble cancer biomarker. Lab Chip 7, 1288–1293 (2007).

VOLUME 29 NUMBER 3 MARCH 2011 nature biotechnology

letters 15. Daniel, K.D. et al. Implantable diagnostic device for cancer monitoring. Biosens. Bioelectron. 24, 3252–3257 (2009). 16. Jaffe, A.S., Babuin, L. & Apple, F.S. Biomarkers in acute cardiac disease: the present and the future. J. Am. Coll. Cardiol. 48, 1–11 (2006). 17. Tarnavski, O. et al. Mouse cardiac surgery: comprehensive techniques for the generation of mouse models of human diseases and their application for genomic studies. Physiol. Genomics 16, 349–360 (2004). 18. Thygesen, K., Alpert, J.S. & White, H.D. Universal definition of myocardial infarction. J. Am. Coll. Cardiol. 50, 2173–2195 (2007). 19. Scherrer-Crosbie, M., Rodrigues, A.C.T., Hataishi, R. & Picard, M.H. Infarct size assessment in mice. Echocardiography 24, 90–96 (2007). 20. Bello, D. et al. Infarct morphology identifies patients with substrate for sustained ventricular tachycardia. J. Am. Coll. Cardiol. 45, 1104–1108 (2005). 21. Roes, S.D. et al. Comparison of myocardial infarct size assessed with contrastenhanced magnetic resonance imaging and left ventricular function and volumes to predict mortality in patients with healed myocardial infarction. Am. J. Cardiol. 100, 930–936 (2007). 22. Takemura, G. & Fujiwara, H. Doxorubicin-induced cardiomyopathy: from the cardiotoxic mechanisms to management. Prog. Cardiovasc. Dis. 49, 330–352 (2007). 23. Robert, J. Long-term and short-term models for studying anthracycline cardiotoxicity and protectors. Cardiovasc. Toxicol. 7, 135–139 (2007). 24. Wallace, K.B. et al. Serum troponins as biomarkers of drug-induced cardiac toxicity. Toxicol. Pathol. 32, 106–121 (2004).

© 2011 Nature America, Inc. All rights reserved.

6. Kim, G.Y., Josephson, L., Langer, R. & Cima, M.J. Magnetic relaxation switch detection of human chorionic gonadotrophin. Bioconjug. Chem. 18, 2024–2028 (2007). 7. Taktak, S., Sosnovik, D., Cima, M.J., Weissleder, R. & Josephson, L. Multiparameter magnetic relaxation switch assays. Anal. Chem. 79, 8863–8869 (2007). 8. Tsourkas, A., Hofstetter, O., Hofstetter, H., Weissleder, R. & Josephson, L. Magnetic relaxation switch immunosensors detect enantiomeric impurities. Angew. Chem. Int. Edn Engl. 43, 2395–2399 (2004). 9. Perez, J.M., Josephson, L., O’Loughlin, T., Hogemann, D. & Weissleder, R. Magnetic relaxation switches capable of sensing molecular interactions. Nat. Biotechnol. 20, 816–820 (2002). 10. Perez, J.M., Josephson, L. & Weissleder, R. Use of magnetic nanoparticles as nanosensors to probe for molecular interactions. ChemBioChem 5, 261–264 (2004). 11. Perez, J.M., O’Loughin, T., Simeone, F.J., Weissleder, R. & Josephson, L. DNA-based magnetic nanoparticle assembly acts as a magnetic relaxation nanoswitch allowing screening of DNA-cleaving agents. J. Am. Chem. Soc. 124, 2856–2857 (2002). 12. Sun, E.Y., Weissleder, R. & Josephson, L. Continuous analyte sensing with magnetic nanoswitches. Small 2, 1144–1147 (2006). 13. Wunderbaldinger, P., Josephson, L. & Weissleder, R. Crosslinked iron oxides (CLIO): a new platform for the development of targeted MR contrast agents. Acad. Radiol. 9 Suppl 2, S304–S306 (2002). 14. Zhao, M., Josephson, L., Tang, Y. & Weissleder, R. Magnetic sensors for protease assays. Angew. Chem. Int. Edn Engl. 42, 1375–1378 (2003).

nature biotechnology VOLUME 29 NUMBER 3 MARCH 2011

277

ONLINE METHODS

© 2011 Nature America, Inc. All rights reserved.

Preparation of magnetic nanosensors. NanoMag-CLD superparamagnetic iron oxide nanoparticles (50 nm) with amine-terminated dextran shells (MicroMod) were coupled with monoclonal antibody against goat IgG (Meridian Life Sciences) using a previously described method25. These particles were then derivatized against specific targets by incubation with goat-produced polyclonal antibodies against cTnI (BiosPacific), myoglobin (BiosPacific) and CKMB (BiosPacific). Unless otherwise specified, particles were suspended in PBS with 1% BSA (Sigma-Aldrich) and 0.1% penicillin-streptomycin (Invitrogen) to minimize bacterial contamination and nonspecific adsorption. Device fabrication and implantation. Derivatized particles were encapsulated within small diffusion devices15. Polycarbonate diffusion membranes (SPI Supplies) were affixed by double-sided adhesive to one side of high-density polyethylene cylinders (thickness = 1.6 mm, inner diameter = 4.1 mm, outer diameter = 7.9 mm). The opposing end was closed off by single-sided adhesive (3M) after the reservoir was filled with 25 µl particle solution. Devices (n = 6 per animal) were implanted subcutaneously in the flank. The animals were euthanized and the devices explanted for single-sided relaxometry at specified time points. Devices were sealed with single-sided adhesive and replaced in the implant site for imaging. Myocardial infarction surgery. All experiments were conducted in accordance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) of the Massachusetts General Hospital. Mouse LAD ligation has been previously described17. The procedure was performed under 2% isoflurane anesthesia with supplemental oxygen. The depth of the anesthesia was monitored by tail pinch, constant monitoring of respiratory rate and heart rate. An intercostal approach between the 3rd and 4th intercostal spaces was taken to expose the pericardium. The pericardium was then visualized and opened and the LAD identified under stereo microscopy. A silk ligature was passed underneath and tied around the vessel to induce occlusion. Occlusion was confirmed by observed blanching of the anterior left ventricular wall. The chest cavity was then carefully closed layer by layer. Sham-operated mice underwent the same procedure without tying the LAD ligation suture. Buprenorphine HCl (0.05–0.1 mg/kg) was administered for post-operative care. The mice were monitored continuously for 6 h, then every 8 h and were treated according to IACUC guidelines if they appeared dyspneic or moribund. Independent measurements of biomarker levels. Commercially available ELISAs were used to independently quantify Myoglobin (Life Diagnostics), CK-MB (Oxis International) and Troponin I (Life Diagnostics) levels following control, sham and MI procedures. Each kit was strictly used in accordance with included instructions. Extravasate samples were obtained by flushing the flank with 1 ml PBS at the indicated times after LAD ligation. Doxorubicin cardiotoxicity biomarker levels were characterized and studied by administration of 20 mg/kg doxorubicin HCl intraperitoneally. Infarct size determination. Area at risk and area of infarction were determined by perfusing Evans blue dye into the carotid artery before euthanizing

nature biotechnology

the animal. Hearts were then removed and sectioned perpendicular to the long axis into 1 mm slices using a McIlwain tissue chopper (Sterling) and stained with 1% 2,3,5-triphenyltetrazolium chloride (TTC) for 5 min at 37 °C. Each slice was analyzed for left ventricular area, area at risk and infarct size. Measurement of proton relaxation times. Proton relaxation measurements were determined on a custom-built system using a single-sided NMR relaxometer (0.43 T and 25 °C; Profile NMR MOUSE, ACT Center for Technology). The sensor’s field gradient enables measurements to be made on a single device positioned above the sensor. Transverse relaxation times for the single-sided MR were measured using a Carr Purcell Meiboom Gill sequence with the following parameters: echo time (TE) = 0.035 ms, 5,714 echoes, 16 scans and recovery time (TR) = 3 s. The echo peak intensities were fit to the equation −t

I = I0e T2 using a custom script running on MATLAB (MathWorks). Measurement of device dynamics and dosage response. Devices were prepared as described above and were individually mounted inside a well on a custom-built holder. Each well was then filled with 400 µl of analyte, and the analyte was changed every 4 to 8 h. The holder was mounted on an automated motion stage and measurements were acquired automatically. The total exposure is given by integrating the analyte concentration as a function of time and reported in units of µg·h/ml. Four devices were incubated in a constant concentration (0 ng/ml to 50 ng/ml) and measured at 20-min intervals to ascertain the diffusion-limited response and the onset of saturation. Six devices were exposed for 4–12 h to myoglobin concentrations of between 30 ng/ml and 100 ng/ml (Fig. 2f (inset)). The analyte was then replaced with PBS, and the devices were incubated overnight. The sensor T2 value stabilizes after ~12 h so relaxation times are reported after the overnight incubation. The return to a zero concentration of myoglobin also mimics a transient in vivo event. Magnetic resonance imaging. MRI measurements were made using a 9.4T animal imager (Bruker Biospin). Imaging protocols included a Tri-plane for localizing the implanted devices. Multi-slice multi-echo (MSME) T2-weighted imaging was performed using the following parameters: flip angle = 90°; matrix size (64 × 64); TR = 2330 ms; TE = 16 equally spaced echoes at 4.07 ms intervals ranging from 4.07 ms to 85.12 ms; field of view = 2 cm × 2 cm, slice thickness = 1 mm. Region of interest analysis was performed and T2 fit was performed using a mono-exponential fitting algorithm for the multi-TE data. Regions of interest incorporating the center slice of the devices were analyzed. Statistics. Statistical comparisons were made by means of the Wilcoxon ranksum two-sided test. This nonparametric method was considered appropriate here as no assumptions are made on the distribution of the data. P < 0.05 was taken to be significant.

25. Josephson, L., Tung, C.H., Moore, A. & Weissleder, R. High-efficiency intracellular magnetic labeling with novel superparamagnetic-Tat peptide conjugates. Bioconjug. Chem. 10, 186–191 (1999).

doi:10.1038/nbt.1780

resource

A functionally characterized test set of human induced pluripotent stem cells

© 2011 Nature America, Inc. All rights reserved.

Gabriella L Boulting1–3,12, Evangelos Kiskinis1,2,12, Gist F Croft4,5,12, Mackenzie W Amoroso4,5,12, Derek H Oakley4,5,12, Brian J Wainger6–8, Damian J Williams9, David J Kahler10, Mariko Yamaki1,2, Lance Davidow2, Christopher T Rodolfa3, John T Dimos3,11, Shravani Mikkilineni2,3, Amy B MacDermott9, Clifford J Woolf 6,7, Christopher E Henderson4,5, Hynek Wichterle4,5 & Kevin Eggan1–3 Human induced pluripotent stem cells (iPSCs) present exciting opportunities for studying development and for in vitro disease modeling. However, reported variability in the behavior of iPSCs has called their utility into question. We established a test set of 16 iPSC lines from seven individuals of varying age, sex and health status, and extensively characterized the lines with respect to pluripotency and the ability to terminally differentiate. Under standardized procedures in two independent laboratories, 13 of the iPSC lines gave rise to functional motor neurons with a range of efficiencies similar to that of human embryonic stem cells (ESCs). Although three iPSC lines were resistant to neural differentiation, early neuralization rescued their performance. Therefore, all 16 iPSC lines passed a stringent test of differentiation capacity despite variations in karyotype and in the expression of early pluripotency markers and transgenes. This iPSC and ESC test set is a robust resource for those interested in the basic biology of stem cells and their applications. Reprogramming of somatic cells to iPSCs presents an opportunity to produce previously inaccessible cell types for disease-related studies1. IPSCs can be made from patients and their healthy relatives, allowing the genetic variants that either predisposed them to or protected them from disease to be studied2–5. However, if patient-specific iPSCs are to become a standard resource, it is vital to understand how reliably they generate differentiated derivatives. Among the concerns raised about iPSCs is that reprogramming may be incomplete, resulting in cell lines with variable gene expression or DNA methylation6–8. Indeed, it was reported that when human iPSC lines were differentiated toward a motor neuron identity, they uniformly failed to produce this neural subtype with the efficiency observed using ESCs9. Moreover, it has been suggested that differentiated progeny of iPSCs that harbor reprogramming proviruses have a problematic gene expression signature that can be resolved only by viral excision5. Another open question is whether standard reprogramming, expansion and directed differentiation processes are robust enough to minimize noise caused by donor-to-donor variation, which could obscure disease-specific phenotypes. Finally, it has not been determined whether individual iPSC lines behave similarly from laboratory to laboratory. To address these questions and to determine whether cell-line variability might limit the utility of iPSCs, it is necessary to systematically

examine in parallel a sufficiently large set of cell lines. We derived a set of iPSC lines that includes many sources of variation that might be encountered in the course of modeling development or disease, including age, sex, health status and donor identity. We then compared the ability of the lines to undergo directed differentiation as a stringent test of their pluripotency. We selected motor neurons as a model system because they are an example of the many differentiated human cell types that cannot be obtained by other means, and are specifically affected in amyotrophic lateral sclerosis (ALS)10. After all cell lines were extensively characterized, we found that, like ESCs, most iPSCs were capable of generating functional motor neurons under a standard differentiation protocol, whereas a few lines required more efficient neuralization. The efficiency with which each individual line generated motor neurons was highly reproducible between two different laboratories, indicating that the collection can function robustly as a shared resource. Potential sources of variation between iPSC lines, such as donor age and transgene expression, did not correlate with differentiation efficiency. Likewise, no significant differences were found between three-factor and four-factor lines, or between lines from healthy and ALS patients. However, we found that two parameters may be associated with differing behavior of lines. Donor identity and donor sex were both associated with variation in

1The Howard Hughes Medical Institute, Cambridge, Massachusetts, USA. 2Harvard Stem Cell Institute, Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, Massachusetts, USA. 3Department of Molecular and Cellular Biology, Harvard University, Cambridge, Massachusetts, USA. 4Project A.L.S./Jenifer Estess Laboratory for Stem Cell Research, Columbia University, New York, New York, USA. 5Departments of Pathology, Neurology and Neuroscience, Columbia University, Center for Motor Neuron Biology and Disease (MNC), and Columbia Stem Cell Initiative (CSCI), New York, New York, USA. 6Program in Neurobiology and FM Kirby Neurobiology Center, Children’s Hospital Boston, Boston, Massachusetts, USA. 7Department of Neurobiology, Harvard Medical School, Boston, Massachusetts, USA. 8Department of Anesthesia, Critical Care and Pain Medicine, Massachusetts General Hospital, Boston, Massachusetts, USA. 9Departments of Physiology and Cellular Biophysics, and Neuroscience, Columbia University, New York, New York, USA. 10The New York Stem Cell Foundation, Inc. (NYSCF), New York, New York, USA. 11Present address: iPierian, Inc., South San Francisco, California, USA. 12These authors contributed equally to this work. Correspondence should be addressed to K.E. ([email protected]) or H.W. ([email protected]).

Received 28 October 2010; accepted 19 January 2011; corrected online 11 February 2011; published online 3 February 2011; doi:10.1038/nbt1783

nature biotechnology volume 29 number 3 mARCH 2011

279

resource

© 2011 Nature America, Inc. All rights reserved.

Table 1 Human stem cell lines used for comparative study Cell type

Donor fibroblast

Cell line

ALS diagnosis

Reprogramming factors

Sex

Donor age

Reference

iPS

11

11a

Healthy control

OCT4/SOX2/KLF4

M

36

This report

iPS

11

11b

Healthy control

OCT4/SOX2/KLF4

M

36

This report

iPS

11

11c

Healthy control

OCT4/SOX2/KLF4

M

36

This report

iPS

15

15b

Healthy control

OCT4/SOX2/KLF4

F

48

This report

iPS

17

17a

Healthy control

OCT4/SOX2/KLF4

F

71

This report

iPS

17

17b

Healthy control

OCT4/SOX2/KLF4

F

71

This report

iPS

18

18a

Healthy control

OCT4/SOX2/KLF4

F

48

This report

iPS

18

18b

Healthy control

OCT4/SOX2/KLF4

F

48

This report

iPS

18

18c

Healthy control

OCT4/SOX2/KLF4

F

48

This report

iPS

20

20b

Healthy control

OCT4/SOX2/KLF4

M

55

This report

iPS

27

27b

SOD1G85S

OCT4/SOX2/KLF4

F

29

This report This report

iPS

27

27e

SOD1G85S

OCT4/SOX2/KLF4

F

29

iPS

29

29A

SOD1L144F

OCT4/SOX2/KLF4/c-MYC

F

82

2

iPS

29

29B

SOD1L144F

OCT4/SOX2/KLF4/c-MYC

F

82

2

iPS

29

29d

SOD1L144F

OCT4/SOX2/KLF4

F

82

This report

iPS

29

29e

SOD1L144F

OCT4/SOX2/KLF4

F

82

This report

ES

–

HuES-3

–

–

M

–

23

ES

–

HuES-6

–

–

F

–

23

ES

–

HuES-9

–

–

F

–

23

ES

–

HuES-13

–

–

M

–

23

ES

–

HuES-3 hb9:GFP

–

–

M

–

12

ES

–

RUES1

–

–

M

–

24

Sixteen human iPSC lines were used for comparison with each other and with six ESC lines. IPSC lines include 14 newly generated three-factor lines from two ALS patients and five controls, and two previously published four-factor lines from one ALS patient. This cohort of human stem cell lines allows comparisons to be made between ESCs and iPSCs, between three-factor and four-factor iPSC lines, between male and female lines, between lines derived from the same donor and those derived from another donor, and between cells derived from ALS patients and control donors.

differentiation performance and warrant further study. The test set reported here has already served as the basis for an analysis of the epigenetic influences on stem cell differentiation potential 11. The test set cell lines are available for distribution and should prove to be a valuable resource for many avenues of stem cell research. RESULTS A test set of iPSC lines We assembled a test set of 16 human iPSC lines, including 14 new lines and 2 previously reported lines2 (Table 1). This set comprised lines from seven individuals of both sexes whose ages ranged from 29 to 82 years. All iPSCs were derived by retroviral transduction of skin fibroblasts. Most lines were produced using only three factors (OCT4, KLF4, SOX2) but two lines derived using the additional factor c-MYC2 were included for comparison.. To allow evaluation of the effects of individual genetic background, we included independent lines derived from the same subjects (two lines from each of two donors, three lines from each of two donors and four lines from one donor). Finally, we included lines from both healthy controls (n = 10) and patients with ALS (n = 6), all of which carried mutations in the superoxide dismutase 1 gene (SOD1). To determine whether reprogramming systematically influences stem cell properties, we compared the performance of these iPSC lines to that of six ESC lines (Table 1). To verify that the newly derived iPSC lines were indeed pluripotent stem cells, we assessed expression of the pluripotency markers alkaline phosphatase, NANOG, OCT4, SSEA3, SSEA4, TRA-1-60 and TRA-1-81. In addition to exhibiting cell-surface staining patterns for both SSEA and TRA-1 proteins, all colonies showed distinct nuclear staining when assayed for NANOG and OCT4 immunoreactivity (Fig. 1a and Supplementary Fig. 1a). All iPSC lines generated 280

compact colonies with a morphology (Fig. 1a and Supplementary Fig. 1a) and cell-cycle profile similar to those of ESC controls (Fig. 1b and Supplementary Fig. 1b). We next determined that the test set could differentiate into all three embryonic germ layers as detected by expression of neuron-specific tubulin (TUJ1), smooth muscle actin (αSMA) and endodermal α-fetoprotein (AFP) after 16 d of differentiation in vitro (Fig. 1c and Supplementary Fig. 1c). Lastly, we tested the ability of several lines to generate teratomas in immune-compromised mice. After injection, each line (14/14 tested) formed teratomas containing complex tissue structures characteristic of all three embryonic germ layers (Fig. 1d and Supplementary Fig. 1d). Based on histological criteria these structures included neuronal fibers, hair follicles, melanocytes, keratin pearls, muscle cells, cartilage, glands and goblet cells. Thus our test set resource contains lines that have been extensively characterized and meet the most stringent criteria for pluripotency (Fig. 1e). Most iPSC lines generate electrically active motor neurons using standard procedures To rigorously and quantitatively test the potential of iPSCs to undergo terminal differentiation, we determined the efficiency with which each line could generate spinal motor neurons (Fig. 2). In response to standard procedures involving retinoic acid and induction of the sonic hedgehog pathway12,13 (Fig. 2a), the majority (19/22) of the iPSC and ESC lines generated cells with a neuronal morphology that expressed TUJ1 and the motor neuron marker ISLET 1/2 (ISL) (Fig. 2b). To confirm the reliability of the lines as a resource for the community, we transferred the entire test set from the laboratory in which they were derived (Eggan laboratory) to a geographically distinct laboratory (Project A.L.S. (PALS)/Jenifer Estess Laboratory for Stem Cell Research, Columbia University), which then repeated volume 29 number 3 mARCH 2011 nature biotechnology

resource the directed differentiation experiments. Consistent with the conclusion that the properties of each cell line are reproducible, the same lines were found to generate ISL+ neurons (Fig. 2c). Indeed, the differentiation efficiency for each line (ranging from 4–15% ISL+/total nuclei) showed no difference between the two laboratories (Fig. 2c; ANOVA f = 1.132, P = 0.301, Supplementary Table 1). Thus, the test set will also serve as a robust resource for research in other centers. To confirm that the ISL+ neurons were motor neurons, we quantified the expression of other markers indicative of this neural subtype. We found that the iPSC lines producing the highest percentage of ISL+ neurons also produced the highest percentage of cells expressing the motor neuron–specific transcription factor HB9 (ref. 14; Fig. 2d). On average, iPSC lines generated HB9+ neurons as efficiently as did the established ESC line HuES-13 (Fig. 2h). In addition, in a tested subset of lines, ISL+ neurons were immunopositive for ChAT, the enzyme required for acetylcholine synthesis (Fig. 2e). These differentiated cultures, which also contained a variety of motor neuron progenitors, expressed the neuiPSC 11a ral marker NCAM (Supplementary Fig. 2a), and expressed substantial levels of mRNA encoding the markers HB9, ChAT and CHT1 (Supplementary Fig. 2b). Finally, unlike PAX6+ progenitors in these cultures, ISL+ neurons were never observed to be actively cycling as measured by Ki67 immunostaining (Supplementary Fig. 2c). Although there were quantitative differences in motor neuron generation among the lines (Fig. 2i and Supplementary Table 2), they did not reflect overall differences between iPSC and ESC lines (Fig. 2g and Supplementary Table 1) or iPSC 11a between healthy control and ALS iPSC lines

b

FB no.11

NANOG/ DNA

Cells in S/G2/M (%)

Phase

HuES-13

S/G2/M

100 80

55.8

55.2

60 40 5.3

20

***

0

HuES n=3

Fibroblasts n=7

TRA-1-60 / DNA

iPSC n = 13

Mesoderm

Ectoderm

iPSC 11b

Endoderm

AFP/DNA

iPSC 27e

α-SMA /DNA

Figure 1 Characterization of pluripotency in the test set of iPSC lines. (a) iPSC colonies were morphologically identical to ESC colonies and expressed the pluripotency markers NANOG and TRA-1-60, unlike the patient fibroblasts from which they were derived. FB, fibroblasts. Scale bars, 200 μm. (b) iPSC lines showed cell cycle profiles similar to those of ESCs and different from their parental fibroblasts. The percentage of cells at different stages of the cell cycle was determined by propidium iodide staining and flow cytometry. The percentage of cells in S, G2 and M phase was determined for each cell line and then averaged for each category. ***P < 0.001, mean ± s.d. (c,d) Like ESCs, iPSC lines generated cell types of all three embryonic germ layers (endoderm, AFP; mesoderm, α-SMA; ectoderm, TUJ1) in vitro, as embryoid bodies (c, EBs; scale bars, 100 μm), and when injected into mouse kidney capsules and allowed to form teratomas in vivo (d; scale bars, 50 μm). Representative images of H&E-stained sections are shown for lines 11b and 27e. Glands and goblet cells (endoderm), cartilage and muscle (mesoderm), pigmented neural epithelium and neural rosettes (ectoderm) are shown in the top and bottom panels, respectively, for both lines. (e) Summary chart depicting assays by which iPSC lines in the test set were characterized. Pluripotency assays for 29A and B were previously published2.

d

HuES-13

EBs-phase

c

TUJ1/DNA

© 2011 Nature America, Inc. All rights reserved.

a

(Fig. 2f and Supplementary Table 1). Moreover, four-factor lines did not underperform relative to three-factor lines (Supplementary Table 1). Therefore, most (13/16) iPSC lines in the resource could generate spinal motor neurons with a reproducible efficiency that is equivalent to that of gold standard ESC lines. To demonstrate that the motor neurons we produced were functional, we compared the electrophysiological properties of motor neurons from four iPSC lines with motor neurons from two ESC lines (Fig. 3). We first monitored intracellular Ca2+ dynamics using the Ca2+-sensitive dyes Fura Red AM (Fig. 3a–b) and Fluo-4 AM (Fig. 3c). This was done both in the absence of exogenous stimulation (Fig. 3c,h,j) to monitor spontaneous activity, and after application of either kainate to activate ionotropic glutamate receptors, or KCl to depolarize the membrane and open voltage-gated Ca2+ channels (Fig. 3d–g,i,k). Spontaneous calcium transients were visible in the cell

e 11a 11b 11c 15b 17a 17b 18a 18b 18c 20b 27b 27e 29A iPS- alkaline phosphatase iPS- NANOG iPS- OCT4 iPS- SSEA3 iPS- SSEA4 iPS- TRA-1-60 iPS- TRA-1-81 iPS- cell cycle analysis iPS- in vitro three-germ layer assay iPS- teratoma formation iPS- FC analysis iPS- transgene expression iPS- standard neuronal differentiation iPS- differentiation in both labs iPS- neuralizing differentiation Neurons- transgene expression Neurons- calcium imaging Neurons- patch clamp assays

nature biotechnology volume 29 number 3 mARCH 2011

+ + + + + + + + + + + + + + − + + −

+ + + + + + + + + + + + + − + + − −

+ + + + + + + − + + + + + − − − − −

+ + + + + + + + + + − + + − − + − −

+ + + + + + + + + + − + + − − + − −

+ + + + + + + + + + − + + − − + − −

+ + + + + + + + + + + + + + − + + +

+ + + + + + + + + + + + + + − + − −

+ + + + + + + + + + + + + + − + + −

+ + + + + + + + + + − + + − − + − −

+ + + + + + + + + + + + + + − + + +

+ + + + + + + + + + + + + + + − − −

+ + + + + + + − + + + + + + − − − −

29B 29d 29e

+ + + + + + + − + + + + + + − − − −

+ + + + + + + + + − + + + + − + − −

+ + + + + + + + + − + + + + + − − −

281

a

ESC/iPSC

Embryoid bodies

Day 0

Day 4

ESC medium NIM medium

b

ISL

RA + HAg

Neurobasal medium BDNF + GDNF + CNTF

ISL / TUJ1/ DNA

Day 29

Day 32

Embryoid bodies dissociated into single cells

Fixed and stained

c

iPSC 18c

25

Eggan laboratory PALS laboratory

ISL ( %)

20 15

10 5

HuES-6

0

11a 18a 18b 18c 27b 27e 29a 29b 29d 29e Hu-13 Hu-3

Cell lines

f

g 16

HB9 / TUJ1 / DNA

ISL ( %)

HB9

8

4 0

iPSC 18c

d

ISL ( %)

12

h

12 8 4 0

n=9

ALS iPSC n=4

iPSC n = 13

HuES n=5

6

HB9 (%)

5 4 3 2 1 0

11a 15b 17a 17b 18a 18b 18c 29A 29B 29d HuES-13 n=3 n=1 n=1 n=2 n=3 n=2 n=2 n=1 n=1 n=2 n=2

Cell lines

e

HuES-13

iPSC 29A

i

25

ISL ( %)

20

bodies and processes of multiple cells from each line, even without treatment (Fig. 3c,h,j; Supplementary Fig. 3a and Supplementary Video 1). Upon exposure to kainate, increases in Ca2+ levels were observed in 78% of cells with neuronal morphology (n = 132 cells). Many kainate-responsive cells also exhibited Ca2+ transients upon exposure to KCl (Fig. 3i,k). Immunostaining confirmed that many of the cells that responded to kainate and KCl in these mixed cultures were ISL+ motor neurons (Fig. 3b,i and Supplementary Fig. 3b). To further demonstrate that iPSC-derived neurons express the repertoire of voltage-gated ion channels characteristic of active neurons, we made electrophysiological recordings using whole-cell patch clamping. All cells with a neuronal morphology, derived from HuES-3 hb9:GFP (n = 9 cells), iPSC line 18a (n = 10 cells) and iPSC line 27b (n = 10 cells), showed fast voltage-activated inward currents followed by slow outward currents, consistent with voltage-activated sodium and potassium currents, respectively (Fig. 3l,m). Inward currents (n = 5/5) were blocked by tetrodotoxin (TTX), an inhibitor of voltage-gated sodium channels (Fig. 3n). In addition, depolarizing 282

Adherent cells Day 25

Day 10

HuES-13

Figure 2 iPSCs show similar capacity for directed motor neuron differentiation compared to ESCs. (a) Protocol for directed differentiation of human stem cell lines into motor neurons. Cells were differentiated as embryoid bodies from day 0–29 in media formulations containing morphogens, including retinoic acid (RA), a small molecule agonist of the sonic hedgehog pathway (HAg) and neurotrophic factors BDNF, GDNF and CNTF. Embryoid bodies were dissociated and single cells plated for adherent culture on day 29. On day 32 cultures were analyzed. NIM, neural induction medium. (b) Representative immunostaining results for iPSC (18c) and ESC (HuES-6) cultures show many ISL+ TUJ1+ motor neurons (scale bars, 50 μm). (c) The percentage of all nuclei that were ISL+ was quantified from differentiations performed independently in the Eggan and PALS laboratories. Data sets from lines differentiated in both laboratories are compared here, are highly similar and have reproducible, characteristic percent ISL + efficiencies. 29e and 27e did not differentiate efficiently in either laboratory. Hu-13, HuES13; Hu-3, HuES-3. (d) Efficiency of motor neuron differentiation was also measured by an alternative marker of motor neuron identity, HB9 (scale bars, 50 μm). (e) Many ISL+ motor neurons were also ChAT+, indicating proper maturation toward a cholinergic transmitter phenotype (scale bar, 50 μm). (f,g) iPSC lines from control and ALS patients differentiated into ISL+ motor neurons with similar efficiencies (f), as did ESCs and iPSCs (g). (h) The percentages of HB9+ nuclei were compared for a subset of iPSC lines and HuES-13. Although comparisons again suggest donor- or line-specific differences, iPSC lines were overall equally capable of generating HB9+ motor neurons as HuES-13 (mean ± s.d.). (i) Percent ISL+ data from both laboratories were pooled for each iPSC and ESC line, and comparisons between lines showed generally similar performance, with significant differences between iPSC line 18c and iPSC lines 11a and 11c (P < 0.05). Hu, HuES.

ISL/ ChAT/ DNA

© 2011 Nature America, Inc. All rights reserved.

resource

**

15

10

5

0

11a 11b 11c 15b 17a 17b 18a 18b 18c 20b 27b 27e 29a 29b 29d 29e Hu9 Hu6 Hu13 Hu3 Hu3hb9 GFP Cell lines

stimuli in current-clamp mode elicited single action potentials in both ESC-derived (n = 2) and iPSC-derived neurons (n = 2), as well as repetitive firing in a neuron derived from iPSC line 18a (Fig. 3o). Therefore, we conclude that both ESC- and iPSC-derived neurons generated from the cell lines in the resource are similarly functional at a physiological level. Contribution of other variables to differentiation Although all cell lines were capable of generating motor neurons, we systematically examined some parameters that have been implicated in differentiation efficiency and so would be of interest to potential users of this resource. The majority of iPSC lines (9 out of 15 tested; Supplementary Fig. 4) exhibited genomic stability at both early (p13) and late (p42) passages. The other six lines (29d, 27b, 29e, 11a, 11b, 15b) acquired disparate abnormalities of varying severity at later passages. However, lines that became karyotypically abnormal did not produce motor neurons with a significantly different efficiency compared with normal lines (P = 0.932; Supplementary Table 1). volume 29 number 3 mARCH 2011 nature biotechnology

As it is not known how these chromosomal changes may affect the behavior of any given cell type, these lines should be used with caution in studies making phenotypic comparisons. We and others have reported that reprogramming transgenes can continue to be expressed in patient-specific iPSC lines, but whether they interfere with differentiation has not been fully studied 2,5. We therefore used quantitative real time (qRT)-PCR to quantify relative levels of transcription of the reprogramming transgenes from both their endogenous loci and from the integrated retrovirus (Fig. 4a). In most cases, levels of viral transcription were either undetectable or very low compared with levels of transcripts from the endogenous loci; indeed, viral SOX2 was never detected (Supplementary Fig. 5a). However, a subset of iPSC lines (11b, 11c, 15b, 18b, 18c, 27b, 27e and 29e) continued to express varying levels of viral KLF4 both in the undifferentiated state and after differentiation to motor neurons (Fig. 4a and Supplementary Fig. 5b–d). Moreover, viral OCT4 transcripts were present in three of the iPSC lines (15b, 18c and 27b) both before and after differentiation. Notably, there was no correlation between the total level of aggregated transgene expression in iPSCs and the efficiency of motor neuron differentiation as judged

Fura Red

b

ISL/Fura Red

c

Δ Fluo-4

d

Fluo-4

e

Fluo-4 + KA

f

Fluo-4 + wash

g

Fluo-4 + KCl

iPSC 11a

KA

j

KCl

k

1.00

RUES1 11a 18a 18c 27b NR

2.0

0.75

Signal intensity (515 nm/650 nm)

0.75

0.50

1.5

1.0

2.5

KA

KCl

2.0

0.50

1.5

1.0

0.25

0.25

0

2.5

Signal intensity (515 nm/650 nm)

i

1.00

Signal intensity (515 nm/650 nm)

h

0.5

0.5

0

60

120

180

Time (s)

0

0

10

20

30

0

40

0

60

Time (min)

180

0

0

10

20

30

40

Time (min)

n 2 nA

l

120

Time (s)

1 nA

20 ms

HuES-3 hb9:GFP

m

27b − TTX

2 ms

27b + TTX

o

40 mV

5 nA

Figure 3 ESC- and iPSC-derived neurons are physiologically active. (a) Images of iPSC 11a–derived neurons filled with Fura Red AM and Fluo-4 AM dyes. The Fura Red channel is shown. The field illustrated is that imaged in b–g. Activity of labeled cells is represented in h and i. Scale bar, 100 μm. (b) ISL immunostaining of 11a field in a–g showing ISL+ neurons (star) and ISL– neurons (arrow). (c) Spontaneous electrical activity in cultured iPSC-derived neurons visualized by a ‘subtracted image’ that shows the difference in pixel intensities between two images acquired 1.7 s apart in the Fluo-4 channel. Higher gray values represent increased pixel intensity. (d–g) Identically exposed pseudocolored averages of ten Fluo-4 AM images taken during the control period before addition of kainic acid (KA) (d), after treatment with 100 μM KA (e), after washing following KA administration (f) and after treatment with 50 μM KCl (g). Warmer colors represent increased fluorescence intensity. (h) Plot of Fluo-4/Fura Red intensity ratio in the somata of the two cells indicated by the star and arrow in a–c; only starred cell shows spontaneous activity. (i) Fluo-4/Fura Red intensity ratio of cells in a–c during sequential administration of KA and KCl indicated by bars above graph. (j) Examples of Fluo-4/Fura Red ratios from cell bodies of single spontaneously active cells in cultures of ESC RUES1–derived neurons, and iPSC 11a–, 18a–, 18c– and 27b–derived neurons as well as one example of a nonresponsive (NR), nonactive cell in an RUES1 culture. (k) Response of cells in j to KA and KCl. (l–m) Sample voltage-clamp traces from ESC (l) and iPSC 18a–derived (m) neurons. (n) Blowup of an iPSC 27b–derived neuron recording reveals typical sodium currents (left), which are blocked by 500 nM TTX (right). (o) Current-clamp recordings of single action potentials in ESC and iPSC 27b–derived neurons as well as multiple action potentials in an iPSC 18a–derived neuron.

by the percentage of ISL+ neurons (R2 = 0.1687). In addition, we carried out immunofluorescence staining for OCT4 and the motor neuron marker ISL (Fig. 4b and Supplementary Fig. 5e). Remarkably, a line such as 15b, which showed persistent transgene expression, generated motor neurons with high abundance, even though many of the motor neurons expressed nuclear OCT4. Therefore, although examples of lines that display karyotypic variation and persistent transgene expression are available in the test set, these phenomena had no detectable effect on rates of motor neuron differentiation. Finally, we looked at the contribution of age, sex and donor genotype to the outcome of differentiation in our test set. There was no correlation between donor age and the percentage of ISL+ neurons generated (R2 = 0.0084). However, there was a significant difference in differentiation efficiency between male and female lines (ANOVA P = 0.048, Supplementary Table 3). These sex-specific differences could result from variable processes such as X-chromosome inactivation. Lastly, we compared the ability of independent lines from several of the donors to differentiate into motor neurons. Southern blot analysis (Supplementary Fig. 6a–c) confirmed that lines from donors 11, 18 and 29 arose from distinct reprogramming events. Subsequently,

iPSC 11a

a

Signal intensity (515 nm/650 nm)

© 2011 Nature America, Inc. All rights reserved.

resource

20 ms

iPSC 18a

nature biotechnology volume 29 number 3 mARCH 2011

HuES-3 hb9:GFP

iPSC 27b

200 ms

iPSC 18a

283

resource phology but less than 10% of differentiated cells were TUJ1+ neurons (Fig. 5b). This was significantly lower compared with 18a (Holm-Sidak, T = 5.037, P = 0.002; Supplementary Table 6), and was in contrast to the results of lines that differentiated appropriately, which produced in excess of 25% TUJ1+ cells (Supplementary Fig. 7b). An indication that early blockade of differentiation might occur in some lines was obtained when we used expression of the antigens SSEA3 and TRA-1-60 to quantify the relative proportion of pluripotent cells within each iPSC line by flow cytometry (Supplementary Fig. 8). One of the recalcitrant iPSC lines (27e) showed higher median fluorescence intensity of TRA-1-60 staining than all others (Supplementary Fig. 8d,e and Supplementary Table 7), suggesting that it might be less prone to spontaneous differentiation. As the three recalcitrant lines were nevertheless able to initially form embryoid bodies, we tested whether they could be coaxed into Suboptimal lines are rescued by active neuralization the motor neuron differentiation pathway by pushing them toward a Although the majority of our cell lines reproducibly generated motor neu- more neural fate at the beginning of the differentiation process. We rons, there were three lines (11b, 27e and 29e) that uniformly failed to do combined our embryoid body protocol with dual SMAD inhibition, so in both laboratories. All three formed embryoid bodies (Fig. 5a, n = 3–7 similar to a previous report15, for the first 9 d using SB431542, an inhibindependent experiments per line), but the embryoid bodies from two itor of transforming growth factor-β1 activin receptor-like kinase, and lines—27e and 29e—became cystic and disaggregated (Fig. 5a). This defect LDN193189, a structural analog of the bone morphogenetic protein was reflected in a significant decrease in total yield of differentiated cells inhibitor dorsomorphin15,16. Comparing the three underperforming (P < 0.05; Supplementary Fig. 7a and Supplementary Table 5) and by iPSC lines (11b, 27e, 29e) to two ESC lines, we found that the previously the failure of these two lines to generate TUJ1+ neurons (Supplementary defective lines were all neuralized in this optimized protocol and gave Fig. 7b). A third line—11b—formed embryoid bodies with normal mor- rise to the same high abundance of TUJ1+ cells as did the ESC lines (>75%; Fig. 5c,d). Notably, the three iPSC lines that previously could not generate neurons now robustly produced ISL+ (Fig. 5c,e) a 3.0 2.5 vOCT4 vOCT4 and HB9+ (Fig. 5f) motor neurons by day 21 at 2.5 eOCT4 eOCT4 2.0 levels indistinguishable from those of both the 2.0 1.5 control ESCs (Fig. 5d–f) and the other 13 iPSC 1.5 lines (Fig. 2h,i). Thus, although three lines 1.0 1.0 in our human stem cell resource underper0.5 0.5 formed using a basal differentiation protocol, 0.0 0.0 11a 11b 11c 15b 17a 11a 17b 18a 18b 18c 20b 27e 29d 29e HuES FB 11a 11b 15b 17a 17b 18a 18b 18c 20b 27b 29d HuAvg HuES FB they could be rescued through a neuralizing avg. avg. avg. avg. d32 neurons iPSC lines 16.0 8 protocol to efficiently generate spinal motor vKLF4 vKLF4 14.0 eKLF4 eKLF4 neurons. 12.0 6 Relative mRNA levels

Relative mRNA levels

Relative mRNA levels

Relative mRNA levels

10.0 8.0 6.0 4.0 2.0 0.0

11a 11b 11c 15b 17a 11a 17b 18a 18b 18c 20b 27e 29d 29e HuES FB avg. avg.

4 2 0

iPSC lines

b

HuES-3

iPSC 17a

11a

11b

15b

17a

17b

18a

18b

18c

20b 27b 29d HuAvg HuES FB avg. avg.

d32 neurons

iPSC 15b

ISL/OCT4/DNA

© 2011 Nature America, Inc. All rights reserved.

inspection of the differentiation data (Fig. 2c,h,i) showed that all three iPSC lines from donor 18 produced many motor neurons, whereas the lines from donor 11 performed less well. The difference in differentiation efficiency between line 18c, the best of the three from that donor, and the two lines from donor 11 that generated motor neurons was indeed significant (P < 0.05; Fig. 2i; Supplementary Table 2), and when comparisons between averaged differentiation efficiencies of multiple lines from each donor were made, a significant difference was found (ANOVA P = 0.006; Supplementary Table 4 and Supplementary Fig. 6d). These results further demonstrate that the cell lines included in the test set may provide an opportunity for other researchers to investigate the effects that sex and other donor-specific phenomenon have in directed differentiation.

Figure 4 Persistent transgene expression does not inhibit differentiation. (a) qRT-PCR was used to measure relative levels of transcript from endogenous genes ‘e’ and viral transgenes ‘v’ of the reprogramming factors OCT4 and KLF4 in undifferentiated iPSCs and ESCs, and in day 32 neuron cultures. Transgene expression or silencing in the undifferentiated cells is maintained after differentiation. Relative levels in undifferentiated HuES-3 were set as 1. FB, fibroblasts. (b) Day 32 motor neuron cultures were co-stained for ISL and OCT4. HuES-3– and iPSC 17a–derived cultures, which do not express viral OCT4, did not stain for OCT4. However, iPSC 15b–derived cultures, which do express viral OCT4, contained many OCT4+ ISL+ motor neurons and OCT4+ ISL– cells. Arrow, OCT4+ ISL+; arrowhead, OCT4+ ISL–; chevron, OCT4– ISL+. Scale bars, 50 μm.

284

DISCUSSION To evaluate iPSCs as a research tool, we generated a large panel of cell lines from multiple donors and examined aspects of the cell lines’ pluripotency and ability to generate terminally differentiated motor neurons. The results of our comparisons confirm the remarkable value of iPSC lines for in vitro studies and demonstrate that they can perform as well as standard ESC lines. This observation held true for experiments carried out using standardized procedures in two geographically distinct laboratories. The analyses presented here serve as a quality control for this stem cell resource, while also providing sufficient data on specific aspects of variability to allow investigators to select lines of particular relevance to their research. Our study is not the first to compare human iPSC and ESC lines, but it is the most extensive comparison of their ability to generate a specific terminally differentiated cell thus far. Most studies have used panels of four or fewer iPSC lines17–22, limiting the

volume 29 number 3 mARCH 2011 nature biotechnology

resource

HuES-13

iPSC 11b

iPSC 18a

HuES-3

iPSC 27e

HuES-3 hb9:GFP iPSC 11b iPSC 29e

© 2011 Nature America, Inc. All rights reserved.

iPSC 29d

b possibilities for understanding variability a Day 4 Day 32 Day 29 Day 32 between cell lines or for drawing general TUJ1 Phase DNA DNA conclusions about functional similarities between iPSCs and ESCs. Similar to a previous report that examined four iPSC lines and one ESC line for generation of terminally differentiated motor neurons9, we found that the differentiation efficiencies of individual iPSC lines vary. However, the earlier study showed that the differentiation capacity of iPSCs was inferior to that of ESCs, whereas we found that iPSC lines could be made to differentiate on average as well as ESC lines. Whether the difference in the conclusions of the two studies is due to differences in d the protocols for reprogramming and motor c Day 18 embryoid bodies Day 22 TUJ1/ISL 120 neuron differentiation, or whether it reflects 80 differences in the numbers of samples ana40 lyzed, remains to be determined. Although all cell lines in our test set were 0 11b 27e 29e HuES-3 HuES-3 capable of generating motor neurons, applihb9:GFP cation of the standard protocol for motor e neuron production did reveal significant 20 quantitative differences in the propensity of the lines for terminal differentiation. These 10 differences were highly reproducible, sug0 gesting that they represent intrinsic charac11b 27e 29e HuES-3 HuES-3 hb9:GFP teristics of the lines. Our initial hypothesis f was that the poorly performing lines would be identified by anomalies in standard tests 8 for stem cell quality. However, all cell lines 4 tested expressed pluripotency markers and could form the three germ layers in vitro and 0 11b 27e 29e HuES-3 HuES-3 in teratomas. Moreover, although variations hb9:GFP in karyotype and transgene expression were observed, they were not accurate predictors Figure 5 Suboptimal iPSC lines can be rescued using SMAD inhibition. (a) During standard of differentiation capacity. Fortunately, a differentiation, iPSC lines 27e and 29e showed abnormal embryoid body morphology and survival solution for identifying such predictors has compared to lines that behaved normally (HuES-3 and 29d shown); phase scale bar, 500 μm; DNA now been proposed by a laboratory that used scale bar, 129 μm. (b) Although embryoid bodies from iPSC line 11b had typical morphology, day our test set to search for epigenetic and tran- 32 cultures showed decreased neuronal TUJ1 staining compared to all other normal lines (HuES-13 scriptional differences that correlate with and iPSC 18a shown), scale bar, 129 μm. (c) Representative phase and immunostaining images for differentiation potential 11. Using the lines previously defective iPSC lines 29e, 11b, and control ESC lines HuES-3 and HuES-3-hb9:GFP. Phase we describe here, they developed a scorecard image scale bars are 500 μm, immunostaining image scale bars are 100 μm. (d–f) Quantification of immunostaining in differentiated cultures derived from the three previously problematic iPSC lines for stem cell quality that predicted our motor (11b, 27e, 29e) and ESC controls; percentage of TUJ1+ cells (d); percentage of ISL+ cells (e); and neuron differentiation results (Fig. 2i) with percentage of HB9+ cells (f). Mean ± s.e.m. remarkable precision. We anticipate that one of the major uses of the cell lines provided through this resource will be to model that, once ALS-related phenotypic differences are discovered, they ALS. Notably, our data demonstrate that several conditions that will prove sufficiently reproducible to serve as a foundation for are necessary for reliable disease modeling are met. First, because research on ALS. ALS is not a developmental disease, our finding that iPSCs carrying an ALS-triggering mutation differentiated similarly to those from METHODS healthy controls is as expected. Second, although lines from different Methods and any associated references are available in the online verhealthy donors, taken together, showed donor-related variation in sion of the paper at http://www.nature.com/naturebiotechnology/. differentiation efficiency, the pairwise comparisons did not reach significance. This increases the chances that phenotypic differences we Note: Supplementary information is available on the Nature Biotechnology website. may eventually observe between ALS cases and controls are related to disease. Nevertheless, as we found real line-to-line differences, it ACKNOWLEDGMENTS We thank H. Mitsumoto, J. Montes, P. Kaufmann and J. Andrews for collecting will be essential to confirm that any phenotypes are ALS-related by skin biopsies; K. Koszka, A. Sproul, A. Hon and A. Garcia-Diaz for technical silencing the mutant SOD1. Lastly, the strong concordance between assistance; M. Park, A. Meissner and C. Bock for manuscript assistance, as well the results from two different laboratories reported here suggests as S. Brenner-Morton and T. Jessell for providing Islet antibodies. This work was nature biotechnology volume 29 number 3 mARCH 2011

285

resource funded by Project A.L.S., P2ALS, NYSTEM and the National Institutes of Health (NIH) GO grant 1RC2 NS069395-01. G.L.B. is a Harvard Stem Cell Institute/NIH Trainee. E.K. is an EMBO Postdoctoral Fellow. B.J.W. is supported by NIH Training Grant 5T32GM007592. C.J.W. is supported by grants from the National Institute of Neurological Disorders and Stroke and the National Institute of Child Health and Development. K.E. is a Howard Hughes Medical Institute early career scientist. AUTHOR CONTRIBUTIONS G.F.C., M.W.A. and D.H.O. maintained human fibroblasts. C.T.R. and J.T.D. reprogrammed all iPSC lines. G.L.B. and E.K. expanded all iPSC lines. G.L.B. and E.K. led and contributed equally to all other experiments and analyses in the Eggan laboratory. G.F.C., M.W.A. and D.H.O. led and contributed equally to all other experiments and analyses in the Project ALS laboratory. D.J.K. did FC analysis. A.B.M., D.J.W. and D.H.O. designed and carried out Ca2+ imaging. B.J.W., G.L.B. and C.J.W. did recordings. M.Y. assisted with teratomas. L.D. assisted with quantitative analysis. S.M. assisted with stem cell culture. G.L.B., E.K., K.E., G.F.C., M.W.A., D.H.O., C.E.H. and H.W. conceived the experiments and wrote the manuscript.

© 2011 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. Kiskinis, E. & Eggan, K. Progress toward the clinical application of patient-specific pluripotent stem cells. J. Clin. Invest. 120, 51–59 (2010). 2. 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). 3. Park, I.H. et al. Disease-specific induced pluripotent stem cells. Cell 134, 877–886 (2008). 4. Lee, G. et al. Modelling pathogenesis and treatment of familial dysautonomia using patient-specific iPSCs. Nature 461, 402–406 (2009). 5. Soldner, F. et al. Parkinson’s disease patient-derived induced pluripotent stem cells free of viral reprogramming factors. Cell 136, 964–977 (2009). 6. Chin, M.H. et al. Induced pluripotent stem cells and embryonic stem cells are distinguished by gene expression signatures. Cell Stem Cell 5, 111–123 (2009). 7. Doi, A. et al. Differential methylation of tissue- and cancer-specific CpG island shores distinguishes human induced pluripotent stem cells, embryonic stem cells and fibroblasts. Nat. Genet. 41, 1350–1353 (2009).

286

8. Stadtfeld, M. et al. Aberrant silencing of imprinted genes on chromosome 12qF1 in mouse induced pluripotent stem cells. Nature 465, 175–181 (2010). 9. Hu, B.Y. et al. Neural differentiation of human induced pluripotent stem cells follows developmental principles but with variable potency. Proc. Natl. Acad. Sci. USA 107, 4335–4340 (2010). 10. Kanning, K.C., Kaplan, A. & Henderson, C.E. Motor neuron diversity in development and disease. Annu. Rev. Neurosci. 33, 409–440 (2010). 11. Bock, C. et al. Reference maps of human ES and iPS cell variation enable high-throughput characterization of pluripotent cell lines. Cell published online, doi:10.1016/j. cell.2010.12.032 (3 February 2011). 12. Di Giorgio, F.P., Boulting, G.L., Bobrowicz, S. & Eggan, K.C. Human embryonic stem cell-derived motor neurons are sensitive to the toxic effect of glial cells carrying an ALS-causing mutation. Cell Stem Cell 3, 637–648 (2008). 13. Wichterle, H., Lieberam, I., Porter, J.A. & Jessell, T.M. Directed differentiation of embryonic stem cells into motor neurons. Cell 110, 385–397 (2002). 14. Arber, S. et al. Requirement for the homeobox gene Hb9 in the consolidation of motor neuron identity. Neuron 23, 659–674 (1999). 15. Chambers, S.M. et al. Highly efficient neural conversion of human ES and iPS cells by dual inhibition of SMAD signaling. Nat. Biotechnol. 27, 275–280 (2009). 16. Zhou, J. et al. High-efficiency induction of neural conversion in human ESCs and human induced pluripotent stem cells with a single chemical inhibitor of transforming growth factor beta superfamily receptors. Stem Cells 28, 1741–1750 (2010). 17. Taura, D. et al. Adipogenic differentiation of human induced pluripotent stem cells: comparison with that of human embryonic stem cells. FEBS Lett. 583, 1029–1033 (2009). 18. Tokumoto, Y., Ogawa, S., Nagamune, T. & Miyake, J. Comparison of efficiency of terminal differentiation of oligodendrocytes from induced pluripotent stem cells versus embryonic stem cells in vitro. J. Biosci. Bioeng. 109, 622–628 (2010). 19. Xi, J. et al. Comparison of contractile behavior of native murine ventricular tissue and cardiomyocytes derived from embryonic or induced pluripotent stem cells. FASEB J. 24, 2739–2751 (2010). 20. Armstrong, L. et al. Human induced pluripotent stem cell lines show stress defense mechanisms and mitochondrial regulation similar to those of human embryonic stem cells. Stem Cells 28, 661–673 (2010). 21. Ghosh, Z. et al. Persistent donor cell gene expression among human induced pluripotent stem cells contributes to differences with human embryonic stem cells. PLoS ONE 5, e8975 (2010). 22. Grigoriadis, A.E. et al. Directed differentiation of hematopoietic precursors and functional osteoclasts from human ES and iPS cells. Blood 115, 2769–2776 (2010). 23. Cowan, C.A. et al. Derivation of embryonic stem-cell lines from human blastocysts. N. Engl. J. Med. 350, 1353–1356 (2004). 24. James, D., Noggle, S.A., Swigut, T. & Brivanlou, A.H. Contribution of human embryonic stem cells to mouse blastocysts. Dev. Biol. 295, 90–102 (2006).

volume 29 number 3 mARCH 2011 nature biotechnology

resource Online Methods

© 2011 Nature America, Inc. All rights reserved.

Cells and cell culture. All cell cultures were maintained at 37 °C, 5% CO2. Human fibroblasts were cultured in KO-DMEM (Invitrogen), supplemented with 20% Earl’s salts 199 (Gibco) and 10% hyclone (Gibco), 1× GlutaMax, penicillin/streptomycin (Invitrogen) and 100 μM 2-mercaptoethanol (Gibco). HuES and iPSCs were maintained on gelatinized tissue culture plastic on a monolayer of irradiated CF-1 mouse embryonic fibroblasts (MEFs) (GlobalStem), in hES media23, with substitution of plasmanate (Talechris) by an additional 10% knockout serum replacement (Invitrogen) in PALS laboratory only supplemented with 20 ng/ml of bFGF. Media was changed every 24 h and lines were passaged by trypsinization (0.5% trypsin EDTA, Invitrogen) or dispase (Gibco, 1 mg/ml in hES media for 30 min at 37 °C). Derivation of human fibroblasts and iPSC generation. Human fibroblasts were generated from 3 mm forearm dermal biopsies after informed consent was obtained, as reported previously2. The murine leukemia retroviral vector pMXs containing the human cDNAs for KLF4, SOX2 and OCT4 (ref. 2) were modified to produce higher titer virus by including the Woodchuck Post-transcriptional Responsive Element (WPRE) of FUGW (Addgene plasmid 14883) downstream of the cDNA. Vesicular stomatitis virus (VSV)-g pseudotyped viruses were packaged and concentrated by the Harvard Gene Therapy Initiative at Harvard Medical School. To produce iPSCs, 30,000 human fibroblasts were transduced at an multiplicity of infection of 10–15 with viruses containing all three genes in hES medium with 8 μg/ml polybrene (Sigma-Aldrich). Cells were incubated with virus for 24 h before medium was changed to standard fibroblast medium for 48 h. Cells were subsequently cultured in standard hES medium and iPSC colonies were manually picked based on morphology within 2–4 weeks. Southern blot analysis. Genomic DNA was extracted from day 21 or day 29 motor neuron differentiation samples for each line using QiaAMP DNA Mini kit (Qiagen) according to manufacturer’s protocols, including RNase digestion. 8 μg gDNA was restricted with BglII overnight according to standard protocols and 6.5 μg run on a 0.8% agarose gel. Neutral Southern capillary transfer was performed overnight, using Amersham Ny+ membrane. OCT4 and SOX2 probes were generated by PCR amplification from OCT4 and SOX2 cDNA plasmids2 using the following primers (OCT4 primer forward: GAGAAGGAGAAGCTGGAGCA, reverse: GTGAAGTGAGGGCTCCCATA, 620 bp product; SOX2, primer forward: AGAACCCCAAGATGCACAAC, reverse: TGGAGTGGGAGGAAGAGGTA, 600 bp product) and Roche PCR DIG Probe Synthesis Kit following manufacturer’s instructions. DNA was bound to membrane by UV, then probe was hybridized overnight (45 °C for OCT4, 55 °C SOX2) using DIG Easy Hyb, followed by immunolabeling with anti-digoxigenin–alkaline phosphatase Fab fragments and detection with CPD-Star chemiluminescent substrate (Roche, following manufacturer’s protocols). After hybridization with OCT4 probe, the blot was stripped in 0.4 M NaOH, 0.1% SDS for 40 min at 65 °C, then washed twice in 2× SSC (Fisher) at 25 °C for 15 min, and reprobed with SOX2 specific probe. Blots were imaged on a KODAK Image Station 4000MM Pro. Flow cytometry for TRA-1-60, SSEA3 and NCAM. Trypsinized suspensions of ~1 M single cells, at day 0 or day 29 of differentiation, were fixed in 4% PFA for 30 min at 4 °C. After washing in PBS, cell suspensions were incubated with the following antibodies obtained from BD Biosciences: SSEA3 PE (1:100, 560237), Tra1-60 AlexaFluor647 (1:100, 560219) or the neural differentiation marker NCAM (CD56 V450 BD biosciences 1:100, 560361) for 30 min protected from light at 4 °C. Stained cells were washed once with PBS and analyzed immediately thereafter on a 5 laser ARIA-IIu ROU Cell Sorter configured with a 100 mm ceramic nozzle and operating at 20 p.s.i. SSEA3+Tra-1-60+ populations were analyzed first by forward and side-scatter properties (FSC, SSC) then analysis gates were set using a combination of fluorescence minus one (FMO) and isotype controls. Cell cycle analysis. Fibroblasts, ESCs and iPSCs were trypsinized to single cells, fixed overnight in cold 70% ethanol, treated with RNaseA (Qiagen) and stained with propidium iodide (PI; 50 μg/ml, Invitrogen) in 0.1% BSA for at least 30 min. Cells were analyzed using the BD Biosystem LSRII FACS analyzer by doublet discrimination, giving rise to a histogram of PI signal with clear 2n and 4n peaks. Spontaneous in vitro three-germ layer differentiation. Whole stem cell colonies were isolated by dispase treatment and plated in suspension in low-cluster 6-well

nature biotechnology

plates (Corning) in hES media without bFGF and plasmanate. Cells aggregated to form embryoid bodies within 24 h. Media was replaced every 48 h, and on day 16 embryoid bodies were trypsin and/or mechanically dissociated and plated on gelatin-coated tissue culture plastic for another 2–7 d of adherent culture before fixation and staining. Teratoma assay. IPSC lines were trypsinized to single cells, washed and resuspended in a minimal volume of CMF-PBS (Cellgro), supplemented with 10% FCS (Invitrogen). At least 1 × 106 cells were injected into the left kidneys of 5- to 6-week-old, severe combined immunodeficient hairless outbred (SHO) mice (3–5 mice/cell line). Xenograft tissue masses formed within 62–131 d, which were extracted, fixed, paraffin-embedded, sectioned and H&E stained. Cells representing all three germ layers were identified after careful examination under the microscope. Further staining images and individual cell line details available upon request. qRT-PCR. Total RNA was isolated using Trizol LS (Invitrogen), 1 μg was treated with DNase (Invitrogen) and was subsequently used to synthesize cDNA with iScript (Bio-Rad). qRT-PCR was then performed using SYBR green (Bio-Rad) and the iCycler system (Bio-Rad). Quantitative levels for all genes were normalized to endogenous GAPDH. For pluripotency genes, levels were expressed relative to the levels in human ES line HuES-3, for motor neuron genes, levels are expressed relative to human ES line HuES-3 hb9-GFP. Standard curves were run to ensure equal efficiency of all primers, and RNA from 293 cells transfected with the plasmids encoding the transgenes was used as a positive control for viral transgene detection. Primer sequences are available upon request. Immunocytochemistry. Pluripotency marker, three-germ layer and OCT3/4– ISL1 stains were applied after fixation overnight in 4% paraformaldehyde at 4 °C, as previously described2. Neuronal cultures were fixed in 4% PFA for 15–30 min at 4 °C, permeabilized and quenched with 0.1–0.2% Triton-X in PBS (wash buffer) and 100 mM glycine (Sigma) for 20 min. Cells were blocked in wash with 10% donkey serum for 30 min and then incubated in primary antibody overnight, secondary antibodies for 1 h. Primary antibodies used in this study are SSEA-3 (1:2, Developmental Studies Hybridoma Bank (DSHB)), SSEA-4 (1:2, DSHB), TRA1-60 (1:500, Chemicon), TRA1-81 (1:500, Chemicon), NANOG (1:500, R&D), OCT3/4 (1:500, Santa Cruz), AFP (1:500, DAKO), α-SMA (1:500, Sigma), ISL (1:200, DSHB, 40.2D6 or 39.4D5, both of which detect Islet1 and Islet2 in the identical pattern in vivo in mouse and chick, Susan Morton, personal communication), HB9 (1:100, DSHB), ChAT (1:100, Chemicon), TUJ1 (1:1,000, Sigma), Ki67 (1:400, Abcam), and Pax6 (1:50, DSHB). Alkaline phosphatase activity was detected in live cultures using the alkaline phosphatase substrate kit (Vector) according to the manufacturer’s instructions. Secondary antibodies used in the Eggan laboratory were AlexaFluor 488, 555, 594 and 647 conjugated (1:300, Invitrogen) and images were acquired on the Opera High-Content Screening System (PerkinElmer) for ISL and HB9 quantifications, and otherwise using an Olympus 1X51 epi-fluorescence microscope, or an LSM 510 META confocal microscope (Zeiss). Secondary antibodies used in the PALS laboratory were DyLight 488, 549, 647 conjugated (1:1,000, Jackson ImmunoResearch) and images (9, 10× fields/sample) were acquired on a fully automated Zeiss Observer Z1 epi-fluorescence microscope. Motor neuron differentiation. Pluripotent stem cell colonies were treated with dispase (1 mg/ml) to separate colonies from feeder cells, then with 10 μM ROCK inhibitor Y-27632 (Sigma) for 1 h in suspension, then followed by trypsinization to single cells, and seeded in low-adherence dishes at 0.2–0.4 million cells/ml in hES medium with 20 ng/ml of bFGF and 10 μM Y-27632 for the first 3 d. At day 4 embryoid bodies were switched to a neural induction medium (DMEM/F12 with l-glutamine, NEAA, penicillin/streptomycin, heparin (2 μg/ml), N2 supplement (Invitrogen) and bFGF (20 ng/ml). At day 10, retinoic acid (RA) (0.1 μM, Sigma), ascorbic acid (0.4 μg/ml, Sigma), db-cAMP (1 μM, Sigma) and 0.1 μM HAg were added. At day 17 the concentration of HAg was increased to 1 μM. At day 25 the base medium was changed to Neurobasal (Invitrogen), with all previous factors and with the addition of 10 ng/ml each of BDNF, GDNF and CNTF (R&D). At day 29 embryoid bodies were dissociated with 0.05% trypsin (Invitrogen), and plated onto poly-d-lysine laminin-coated chamber slides (BD Biosciences) at 0.2– 0.5 million cells/well. Plated neuron cultures were cultured in the same medium

doi:10.1038/nbt.1783

resource with the addition of B27 (Invitrogen), 25 μM β-mercaptoethanol (Millipore) and 25 μM glutamic acid (Sigma), and fixed 3 d later.

© 2011 Nature America, Inc. All rights reserved.

Neuralizing motor neuron differentiation. IPSCs and ESCs were differentiated as described above, but with the following modifications: differentiations were started from dispased colonies triturated to become ~50-cell aggregates of iPSCs, and from days 1–9 were cultured in the presence of SB431542 (10 μM, Sigma-Aldrich) and LDN193189 (0.2 μM, Stemgent) to neuralize the cultures. From day 5 onward, BDNF (10 ng/ml, R&D), ascorbic acid (0.4 μg/ml, Sigma) and RA (Sigma) were added. From day 7 onward, Smoothened agonist 1.3 (SAG) (Calbiochem) was added at 0.5 μM to replace HAg. Aggregates were dissociated, plated and assayed as described above on day 21. Quantitative image analysis. Quantitative image analysis of differentiated neuronal cultures, for DAPI, TUJ1, ISLET and HB9, was done using the multiwavelength cell scoring module in MetaMorph (Molecular Devices) software by the PALS laboratory, or Opera/Acapella software (PerkinElmer) by the Eggan laboratory. In brief, intensity thresholds were set, blinded to sample identity, to selectively identify as positive cells, which displayed unambiguous signal intensity above local background. These parameters were used on all samples, and only minimally adjusted for different staining batches as necessary. Script and Parameter files available upon request. Eggan: a minimum of 20,000–160,000 cells per sample were analyzed from 60–180 20× fields per sample. PALS: a minimum of 4,000 cells per sample were analyzed from nine 10× fields per sample. Total cell number analysis. As different image field sizes were used in different laboratories, total cells/field were normalized as follows. For all cell lines differentiated in parallel in both laboratories, the mean value for each line was averaged with the mean value of the other lines in this set. These values then generated a ratio (mean cells/field in PALS laboratory/mean cells/field in Eggan laboratory), which was then used to normalize the values from the Eggan laboratory to those from the PALS laboratory. Voltage-clamp and current-clamp recordings. Differentiated d44 embryoid bodies were dissociated, plated at 8,000 cells/cm2 on lysine/laminin-coated coverslips, and allowed to mature for 6 d. Whole-cell voltage-clamp or current-clamp recordings were made using a Multiclamp 700B (Molecular Devices) at 21–23 °C. Data were digitized with a Digidata 1440A A/D interface, and recorded and analyzed using pCLAMP 10 software (Molecular Devices). Data were sampled at 20 kHz and low-pass filtered at 2 kHz. Patch pipettes were pulled from borosilicate glass capillaries on a Sutter Instruments P-97 puller and had resistances of 2–4 MΩ. The pipette capacitance was reduced by wrapping the shank with Parafilm

doi:10.1038/nbt.1783

and compensated for using the amplifier circuitry. Series resistance was typically 5–10 MΩ, always <15 MΩ, and compensated by at least 80%. Linear leakage currents were digitally subtracted using a P/4 protocol. Leak currents were typically <100 pA, but occasionally leak currents up to 500 pA were tolerated to accurately document the percentage of cells with voltage-activated sodium currents. Voltages were elicited from a holding potential of –90 mV to test potentials ranging from –90 mV to 20 mV in 10 mV increments. The intracellular solution was a potassium-based solution and contained (mM) KCl, 135; MgCl2, 2; HEPES, 10 (pH 7.4 with KOH). The extracellular was sodium-based and contained NaCl, 135; KCl, 5; CaCl2, 2; MgCl2, 1; glucose, 10; HEPES, 10, pH 7.4 with NaOH). Tetrodotoxin was purchased from Tocris Bioscience. Ca2+ imaging. ES and iPSCs were differentiated under the neuralizing motor neuron differentiation protocol above, dissociated at day 21, cryopreserved, seeded on 15–25 mm diameter coverslips at a density of 125,000–250,000 cells per coverslip in standard media, and matured 12–14 d before Ca2+ imaging. Cells were loaded with 5 μM Fura Red AM and 3 μM Fluo-4 AM (Invitrogen) dissolved in 0.2% dimethyl sulfoxide/0.04% pluronic acid (Sigma-Aldrich) in HEPESbuffered physiological salt solution (PSS) for 1 h at 25 °C. PSS contained (mM): NaCl 145, KCl 5, HEPES 10, CaCl2 2, MgCl2 2 and glucose 5.5, pH 7.4. Cultures were continuously superfused with PSS at a rate of ~0.5 ml/minute. The cultures were imaged on a C-1 inverted confocal microscope (Nikon Instruments). The fluorescent Ca2+ indicators were excited using a 488 nm solid-state laser and emitted light from the Fluo-4 and Fura-Red recorded in separate channels using 500–530 nm band-pass and 650 nm long-pass filters, respectively. We acquired 256 × 256 pixel images using a 20× 0.7 NA air objective (Nikon). For imaging spontaneous Ca2+ transients, single sets of 300 images were acquired at a rate of ~2 Hz from each coverslip. For the kainate and KCl experiments, 36 images were acquired at a rate of 0.033 Hz and the superfusing PSS was replaced with PSS containing kainate (100 μM) or KCl (50 mM). The NaCl concentration of the PSS was reduced to maintain a constant osmolality and Cl– concentration. Image analysis was performed using ImageJ (http://rsb.info.nih.gov/ij/) and custom-written macros. Ca2+ transients were determined from regions of interest encompassing the soma of individual cells, using the ratio of intensities from the Fluo-4 and Fura Red channels. Two cultures obtained from a single differentiation of each cell line were used for the kainate and KCl Ca2+ imaging experiments. Statistical analyses. All quantitative data were analyzed using SigmaPlot. Sample groups were subject to One Way ANOVA, with Holm-Sidak post hoc pairwise comparisons, or, if equal variance tests failed, by Kruskal-Wallis ANOVA on ranks, with Dunn’s post hoc pairwise comparisons. Alpha was set at 0.05 for all ANOVAs, ANOVAs on ranks and post hoc tests.

nature biotechnology

careers and recruitment

A mentoring program for women scientists meets a pressing need Masha Fridkis-Hareli

© 2011 Nature America, Inc. All rights reserved.

An innovative program supports the career development of women scientists in academia and industry.

T

he majority of women scientists who have been successful in their careers share one thing in common: they have all been mentored either formally or informally by a friend, colleague or coach. Mentoring originates from the Greek word meaning ‘enduring’ and is defined as a sustained relationship between a junior and an experienced person. Through continued involvement, the mentor offers support, guidance and assistance as the mentee goes through a difficult period, faces new challenges or works to correct earlier problems. The opportunities to mentor and be mentored have expanded recently with the increase in the number of women’s professional groups who share the common goal of addressing the growing need for career development and equality in the workplace. The Association for Women in Science (AWIS) is a global nonprofit organization dedicated to promoting full participation of women across all scientific fields through mentoring, education and job opportunities. AWIS was established in 1971 by 27 women participating in the annual Federation of American Societies for Experimental Biology meeting who gathered to exchange solutions for overcoming job discrimination, lower pay and professional isolation. Today, after 40 years, with over 50 chapters across the country, AWIS is a premiere national advocacy and leadership organization supporting equity and career advancement for women— from the bench to the board room. Masha Fridkis-Hareli is president of the Massachusetts chapter of the Association for Women in Science (MASS-AWIS) and an associate scientific director at Taligen Therapeutics (acquired by Alexion Pharmaceuticals), Cambridge, Massachusetts, USA. e-mail: [email protected]

The Massachusetts chapter of AWIS (MASS-AWIS) was established in 2004 by Joanne Kamens and Kristin O’Brien and has grown steadily by attracting increasing numbers of local women professionals to its events, which focus on networking and career development opportunities. Membership in the chapter has almost tripled during the past several years, reaching close to 200 members in 2010, almost in direct defiance of the nowinfamous speech at a diversity conference (of all places) by then-president of Harvard University Larry Summers, in which he questioned the innate scientific capabilities of women. MASS-AWIS offers a variety of programs reflecting the increased awareness among women in scientific, technical and medical careers of the need for promotion of equal opportunity in academic and indus-

trial sectors. Some examples include panel discussions with high-profile academic and industry leaders, workshops on career skills, joint events with other local women’s organizations and a mentoring program. Meeting a pressing need The MASS-AWIS Mentoring Circle program, now in its third year, has become a popular venue for junior academic and industry scientists to get advice, support and information from experienced mentors who commit their time to the program. It also provides a safe environment to discuss workplace issues and an opportunity to gain leadership experience. Over 150 participants have been involved in the program since its inception. Kamens, a senior director at RXi Pharmaceuticals and mentoring program cofounder (with Masha

Participants of the 2009 mentoring program listen to Joanne Kamens presenting at the end-of-year event.

nature biotechnology volume 29 number 3 march 2011

287

© 2011 Nature America, Inc. All rights reserved.

careers and recruitment Fridkis-Hareli), recalls: “When I progressed through the industry management ranks, the number of women at the executive level that I met was decreasing exponentially. I felt that there was a pressing need for women scientists to get mentoring and advice starting from the early stages of their career and to the executive level. I wanted to implement what I learned from other women’s organizations like HBA [Healthcare Businesswomen Association, which focuses on more advanced biopharmaceutical industry professionals], which has a mentoring program in place, at AWIS, which is geared toward junior scientists from academia and industry who do not have a mentor or cannot afford a private career coach. Group mentoring seemed to fit best the model of peer support and open communication between the circle members.” The program is free of charge and only requires membership in AWIS at the national and local chapter levels. Each cycle of the program starts in September with an event designed to introduce the concepts of mentoring and encourage goal setting for the year. Planning for the September kickoff event, however, begins as early as spring. Flyers are sent to the MASS-AWIS community and to the wider population of women scientists in the greater Boston area to recruit volunteer mentors and interested mentees. New registrants who have not been enrolled in the AWIS mentoring program in the past are given priority for participating in the current program. Following electronic sign ups, the mentoring committee meets to assign each participant to a circle. The circles, consisting of 3–5 peers and 1–2 mentors, are carefully matched based on mutual interests, career goals and geographical location. The program also has ‘connectors’ in place. The connectors check in several times with program participants during the year to ensure the mentoring circles are meeting, verify that participants are enjoying the experience and provide materials or advice when needed. The program runs from September to May, during which time the groups meet one time each month for several hours to discuss progress toward individual goals and various topics including balance of work and family, getting started in academia, tackling various academic track issues, career change exploration, transition

288

to industry, learning to network, mid-career decisions, challenges faced during work and other career development topics. Karen Page, a senior research scientist at Pfizer and a cochair of the mentoring program, says, “When we advertise the new mentoring program each year, I am unsure of the number of participants that will enroll. We have had good attendance in the past and we hope to expand and grow the program, but you never know. The committee works hard on coordination, organization and planning, and we want our efforts to pay off. We make it clear that attendance of the kickoff event is mandatory. This first face-to-face meeting of the groups is crucial for the introduction and the smooth continuation of the program.” This past September, over 60 women attended the kickoff event. Prior to the

Many past mentoring groups continue to meet after the official completion of their program, indicating that strong and longlasting bonds were created between the circle members. general presentation, a mentor training session took place to help clarify the role of a mentor and to give some tips on facilitating group discussion. This session was followed by an orientation meeting for all program participants, who were also seated in their new circle groups. Every member of a group signed a contract as an official, binding document stating their commitment to maintaining the confidentiality of shared communications, meeting regularly at the times agreed upon and actively participating in circle discussions. Career strategist Sarah Cardozo Duncan then led the groups through a number of exercises aimed at getting the members of each circle to become familiar with one another and to facilitate scheduling their first meeting. The room was filled with lively conversations and laughter, and it was remarkable to watch how a group of strangers became comfortable sharing personal information and transformed into a team of

committed women bound by mutual support, trust and encouragement to succeed. Although it is too early to predict how this year’s mentoring program will influence each participant’s self-growth and development, it is worth mentioning that many past groups continue to meet after the official completion of their program, indicating that strong and long-lasting bonds are created between the circle members. In the words of last year’s participants of a mentoring group, “We gained so much from participating in this mentoring circle, especially because of the diversity of backgrounds, expertise and perspectives of each member. We clicked from the beginning, got along well and everyone was genuinely concerned about each other. Our meetings helped us share ways to network efficiently, communicate effectively and deal with difficult bosses. It was enriching to think and listen about the dynamics of different work environments in our monthly meetings and overall, it also provided a nice change of schedule and an opportunity to unwind among women scientists who could understand where we were coming from.” Inspiring new programs Inspired by her MASS-AWIS mentoring program experience, Page tried to implement ideas of mentoring and adjust them to the specifics of the industry environment at Pfizer. “We started a pilot mentoring program at Pfizer to address employee satisfaction at the junior PhD level. I would have liked to model it after our mentoring circles, but after talking to folks here and to Joanne Kamens, I learned that one-on-one relationships were better for the workplace. Our program ran for six months. We solicited volunteer mentors and mentees and had our HR representative help match them with as much confidentiality as possible. Each pair was expected to meet 1–2 times per month. At the end, the mentoring pairs thought that the program was successful,” she says. MASS-AWIS looks forward to initiating similar mentoring programs in academia and other organizations and is proud that its Mentoring Circle program has helped to grow the community of women scientists who work together to achieve their highest potential. COMPETING FINANCIAL INTERESTS The author declares no competing financial interests.

volume 29 number 3 march 2011 nature biotechnology

© 2011 Nature America, Inc. All rights reserved.

people

AlloCure (Burlington, MA, USA) has announced the appointment of Kevin J. Heyeck (left) as chief business officer. He joins the company with more than 20 years of experience in business development and corporate partnering for emerging pharmaceutical companies, most recently leading business development efforts at Vitae Pharmaceuticals. “Kevin Heyeck has a very impressive track record in business development, and he has been responsible for many major partnerships with large pharmaceutical and biotechnology companies worldwide,” says AlloCure president and CEO Robert M. Brenner. “Kevin’s arrival marks an important milestone in AlloCure’s evolution as we advance into later stage clinical development and chart the course for future commercialization of our novel treatment for acute kidney injury.”

Matthias Baumann has been appointed chief medical officer and a member of the management board of Noxxon Pharma (Berlin). He joins the company from Focus Clinical Drug Development, where he served as CSO and managing director. Alios BioPharma (S. San Francisco, CA, USA) has named Carol L. Brosgart (left) to the position of chief medical officer. Brosgart joins Alios from Children’s Hospital & Research Center (Oakland, CA, USA) where she served as senior vice president and chief medical officer. Previously, she served in several senior management roles at Gilead Sciences including vice president, public health and policy, vice president, clinical research and vice president, medical affairs. Genesis Biopharma (Los Angeles) has named Anthony J. Cataldo president and CEO. Cataldo, who previously served as chairman, CEO and a director of Oxis International, succeeds Robert Brooke, who will serve as a strategic advisor to the company. In addition, Genesis has appointed Michael Handelman CEO, treasurer and secretary, succeeding Richard McKilligan. Affymax (Palo Alto, CA, USA) has announced the promotion of Herb Cross, vice president of finance and chief accounting officer, to CFO. He succeeds Paul Cleveland, who leaves for a 290

new role in a private equity firm. Prior to joining Affymax, Cross was vice president, finance for Facet Biotech. Geron (Menlo Park, CA, USA) has announced its new leadership structure: David Greenwood, CFO, has been named president, interim CEO and a member of the board of directors; Hoyoung Huh, a director of the company, becomes executive chairman; and former chairman Alexander Barkas transitions to the role of lead independent director. Thomas Okarma has stepped down as president, CEO and director of the company and will serve as a consultant. Paul E. Huff has joined Amarin (Dublin and Mystic, CT, USA) as chief commercial officer. He most recently served as vice president of marketing at Reliant Pharmaceuticals. Ken Lisaius has been named to the newly created position of senior advisor and director for public affairs for the Biotechnology Industry Organization (BIO; Washington, DC, USA). He will oversee the development and implementation of BIO’s education and industry branding campaign. Most recently, Lisaius held the position of senior counselor at Brightline Media. He was previously deputy director of the Office of Media Affairs and special assistant to the President of the United States and deputy White House press secretary under President George W. Bush. Scil Technology (Martinsried, Germany) has named Christian Nafe CEO, succeeding Weishui Weiser, who is retiring after 5 years as

managing director. Nafe joined Scil Technology in 2002 as CFO. H. Stewart Parker has been appointed CEO and a member of the board of directors of the Infectious Disease Research Institute (Seattle), a nonprofit organization applying innovative science to prevent, detect and treat infectious diseases of poverty. Parker previously served as president and CEO of Targeted Genetics and vice president, corporate development at Immunex. She is currently a commercialization consultant with the Washington Biotechnology & Biomedical Association and serves on the boards of OncoGenex and C3 Jian. Aeolus Pharmaceuticals (Mission Viejo, CA, USA) has announced the appointment of Russell L. Skibsted as senior vice president and CFO. Skibsted is a seasoned executive with over 25 years of experience. Previously, he was senior vice president and chief business officer of Spectrum Pharmaceuticals. Glenn Tillotson has joined Optimer Pharmaceuticals (San Diego) as senior vice president of medical affairs. He joins the company with more than 25 years of pharma industry experience, most recently heading the medical education programs at ViroPharma. He previously served as executive director of scientific affairs at Replidyne and vice president, scientific and medical relations for Oscient Pharmaceuticals. In addition, Optimer has promoted Sherwood Gorbach to the position of CSO and senior vice president of R&D and Marc Lesnick to the newly created position of vice president of regulatory affairs. Gorbach joined Optimer in 2005 as chief medical officer and Lesnick had served as director of regulatory affairs since 2008. Diane L. Tribble has been named CSO of Aegerion Pharmaceuticals (Cambridge, MA, USA). Most recently, she held the position of vice president of clinical development at Isis Pharmaceuticals, where she led the development of mipomersen, a cholesterol-reducing antisense therapeutic currently in late-stage clinical development. Previously, she was director of clinical research at the Merck Research Laboratories.

volume 29 number 3 march 2011 nature biotechnology

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

No title

Recommend Documents